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> |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 36 | #include <linux/mutex.h> |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 37 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 38 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 39 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 40 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 41 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 42 | #include <linux/eventfd.h> |
| 43 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 44 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 45 | #include <linux/seq_file.h> |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 46 | #include <linux/vmalloc.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 47 | #include <linux/mm_inline.h> |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 48 | #include <linux/page_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 49 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 50 | #include <linux/oom.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 51 | #include "internal.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 52 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 53 | #include <asm/uaccess.h> |
| 54 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 55 | #include <trace/events/vmscan.h> |
| 56 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 57 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 58 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 59 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 60 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 61 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 62 | /* 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] | 63 | int do_swap_account __read_mostly; |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 64 | |
| 65 | /* for remember boot option*/ |
| 66 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED |
| 67 | static int really_do_swap_account __initdata = 1; |
| 68 | #else |
| 69 | static int really_do_swap_account __initdata = 0; |
| 70 | #endif |
| 71 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 72 | #else |
| 73 | #define do_swap_account (0) |
| 74 | #endif |
| 75 | |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 76 | /* |
| 77 | * Per memcg event counter is incremented at every pagein/pageout. This counter |
| 78 | * is used for trigger some periodic events. This is straightforward and better |
| 79 | * than using jiffies etc. to handle periodic memcg event. |
| 80 | * |
| 81 | * These values will be used as !((event) & ((1 <<(thresh)) - 1)) |
| 82 | */ |
| 83 | #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */ |
| 84 | #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */ |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 85 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 86 | /* |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 87 | * Statistics for memory cgroup. |
| 88 | */ |
| 89 | enum mem_cgroup_stat_index { |
| 90 | /* |
| 91 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. |
| 92 | */ |
| 93 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 94 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 95 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 96 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
| 97 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 98 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 99 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
| 100 | /* incremented at every pagein/pageout */ |
| 101 | MEM_CGROUP_EVENTS = MEM_CGROUP_STAT_DATA, |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 102 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 103 | |
| 104 | MEM_CGROUP_STAT_NSTATS, |
| 105 | }; |
| 106 | |
| 107 | struct mem_cgroup_stat_cpu { |
| 108 | s64 count[MEM_CGROUP_STAT_NSTATS]; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 109 | }; |
| 110 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 111 | /* |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 112 | * per-zone information in memory controller. |
| 113 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 114 | struct mem_cgroup_per_zone { |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 115 | /* |
| 116 | * spin_lock to protect the per cgroup LRU |
| 117 | */ |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 118 | struct list_head lists[NR_LRU_LISTS]; |
| 119 | unsigned long count[NR_LRU_LISTS]; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 120 | |
| 121 | struct zone_reclaim_stat reclaim_stat; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 122 | struct rb_node tree_node; /* RB tree node */ |
| 123 | unsigned long long usage_in_excess;/* Set to the value by which */ |
| 124 | /* the soft limit is exceeded*/ |
| 125 | bool on_tree; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 126 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
| 127 | /* use container_of */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 128 | }; |
| 129 | /* Macro for accessing counter */ |
| 130 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) |
| 131 | |
| 132 | struct mem_cgroup_per_node { |
| 133 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 134 | }; |
| 135 | |
| 136 | struct mem_cgroup_lru_info { |
| 137 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; |
| 138 | }; |
| 139 | |
| 140 | /* |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 141 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 142 | * their hierarchy representation |
| 143 | */ |
| 144 | |
| 145 | struct mem_cgroup_tree_per_zone { |
| 146 | struct rb_root rb_root; |
| 147 | spinlock_t lock; |
| 148 | }; |
| 149 | |
| 150 | struct mem_cgroup_tree_per_node { |
| 151 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 152 | }; |
| 153 | |
| 154 | struct mem_cgroup_tree { |
| 155 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 156 | }; |
| 157 | |
| 158 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 159 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 160 | struct mem_cgroup_threshold { |
| 161 | struct eventfd_ctx *eventfd; |
| 162 | u64 threshold; |
| 163 | }; |
| 164 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 165 | /* For threshold */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 166 | struct mem_cgroup_threshold_ary { |
| 167 | /* An array index points to threshold just below usage. */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 168 | int current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 169 | /* Size of entries[] */ |
| 170 | unsigned int size; |
| 171 | /* Array of thresholds */ |
| 172 | struct mem_cgroup_threshold entries[0]; |
| 173 | }; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 174 | |
| 175 | struct mem_cgroup_thresholds { |
| 176 | /* Primary thresholds array */ |
| 177 | struct mem_cgroup_threshold_ary *primary; |
| 178 | /* |
| 179 | * Spare threshold array. |
| 180 | * This is needed to make mem_cgroup_unregister_event() "never fail". |
| 181 | * It must be able to store at least primary->size - 1 entries. |
| 182 | */ |
| 183 | struct mem_cgroup_threshold_ary *spare; |
| 184 | }; |
| 185 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 186 | /* for OOM */ |
| 187 | struct mem_cgroup_eventfd_list { |
| 188 | struct list_head list; |
| 189 | struct eventfd_ctx *eventfd; |
| 190 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 191 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 192 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 193 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 194 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 195 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 196 | * The memory controller data structure. The memory controller controls both |
| 197 | * page cache and RSS per cgroup. We would eventually like to provide |
| 198 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 199 | * to help the administrator determine what knobs to tune. |
| 200 | * |
| 201 | * 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] | 202 | * we hit the water mark. May be even add a low water mark, such that |
| 203 | * no reclaim occurs from a cgroup at it's low water mark, this is |
| 204 | * a feature that will be implemented much later in the future. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 205 | */ |
| 206 | struct mem_cgroup { |
| 207 | struct cgroup_subsys_state css; |
| 208 | /* |
| 209 | * the counter to account for memory usage |
| 210 | */ |
| 211 | struct res_counter res; |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 212 | /* |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 213 | * the counter to account for mem+swap usage. |
| 214 | */ |
| 215 | struct res_counter memsw; |
| 216 | /* |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 217 | * Per cgroup active and inactive list, similar to the |
| 218 | * per zone LRU lists. |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 219 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 220 | struct mem_cgroup_lru_info info; |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 221 | |
KOSAKI Motohiro | 2733c06 | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 222 | /* |
| 223 | protect against reclaim related member. |
| 224 | */ |
| 225 | spinlock_t reclaim_param_lock; |
| 226 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 227 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 228 | * While reclaiming in a hierarchy, we cache the last child we |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 229 | * reclaimed from. |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 230 | */ |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 231 | int last_scanned_child; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 232 | /* |
| 233 | * Should the accounting and control be hierarchical, per subtree? |
| 234 | */ |
| 235 | bool use_hierarchy; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 236 | atomic_t oom_lock; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 237 | atomic_t refcnt; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 238 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 239 | unsigned int swappiness; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 240 | /* OOM-Killer disable */ |
| 241 | int oom_kill_disable; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 242 | |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 243 | /* set when res.limit == memsw.limit */ |
| 244 | bool memsw_is_minimum; |
| 245 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 246 | /* protect arrays of thresholds */ |
| 247 | struct mutex thresholds_lock; |
| 248 | |
| 249 | /* thresholds for memory usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 250 | struct mem_cgroup_thresholds thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 251 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 252 | /* thresholds for mem+swap usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 253 | struct mem_cgroup_thresholds memsw_thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 254 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 255 | /* For oom notifier event fd */ |
| 256 | struct list_head oom_notify; |
| 257 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 258 | /* |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 259 | * Should we move charges of a task when a task is moved into this |
| 260 | * mem_cgroup ? And what type of charges should we move ? |
| 261 | */ |
| 262 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 263 | /* |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 264 | * percpu counter. |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 265 | */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 266 | struct mem_cgroup_stat_cpu *stat; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 267 | /* |
| 268 | * used when a cpu is offlined or other synchronizations |
| 269 | * See mem_cgroup_read_stat(). |
| 270 | */ |
| 271 | struct mem_cgroup_stat_cpu nocpu_base; |
| 272 | spinlock_t pcp_counter_lock; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 273 | }; |
| 274 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 275 | /* Stuffs for move charges at task migration. */ |
| 276 | /* |
| 277 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a |
| 278 | * left-shifted bitmap of these types. |
| 279 | */ |
| 280 | enum move_type { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 281 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 282 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 283 | NR_MOVE_TYPE, |
| 284 | }; |
| 285 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 286 | /* "mc" and its members are protected by cgroup_mutex */ |
| 287 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 288 | spinlock_t lock; /* for from, to */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 289 | struct mem_cgroup *from; |
| 290 | struct mem_cgroup *to; |
| 291 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 292 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 293 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 294 | struct task_struct *moving_task; /* a task moving charges */ |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 295 | struct mm_struct *mm; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 296 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 297 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 298 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 299 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 300 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 301 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 302 | static bool move_anon(void) |
| 303 | { |
| 304 | return test_bit(MOVE_CHARGE_TYPE_ANON, |
| 305 | &mc.to->move_charge_at_immigrate); |
| 306 | } |
| 307 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 308 | static bool move_file(void) |
| 309 | { |
| 310 | return test_bit(MOVE_CHARGE_TYPE_FILE, |
| 311 | &mc.to->move_charge_at_immigrate); |
| 312 | } |
| 313 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 314 | /* |
| 315 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 316 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 317 | */ |
| 318 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) |
| 319 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) |
| 320 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 321 | enum charge_type { |
| 322 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
| 323 | MEM_CGROUP_CHARGE_TYPE_MAPPED, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 324 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 325 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 326 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 327 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 328 | NR_CHARGE_TYPE, |
| 329 | }; |
| 330 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 331 | /* only for here (for easy reading.) */ |
| 332 | #define PCGF_CACHE (1UL << PCG_CACHE) |
| 333 | #define PCGF_USED (1UL << PCG_USED) |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 334 | #define PCGF_LOCK (1UL << PCG_LOCK) |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 335 | /* Not used, but added here for completeness */ |
| 336 | #define PCGF_ACCT (1UL << PCG_ACCT) |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 337 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 338 | /* for encoding cft->private value on file */ |
| 339 | #define _MEM (0) |
| 340 | #define _MEMSWAP (1) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 341 | #define _OOM_TYPE (2) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 342 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
| 343 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) |
| 344 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 345 | /* Used for OOM nofiier */ |
| 346 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 347 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 348 | /* |
| 349 | * Reclaim flags for mem_cgroup_hierarchical_reclaim |
| 350 | */ |
| 351 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 |
| 352 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) |
| 353 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 |
| 354 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 355 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
| 356 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 357 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 358 | static void mem_cgroup_get(struct mem_cgroup *mem); |
| 359 | static void mem_cgroup_put(struct mem_cgroup *mem); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 360 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 361 | static void drain_all_stock_async(void); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 362 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 363 | static struct mem_cgroup_per_zone * |
| 364 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
| 365 | { |
| 366 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
| 367 | } |
| 368 | |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 369 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
| 370 | { |
| 371 | return &mem->css; |
| 372 | } |
| 373 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 374 | static struct mem_cgroup_per_zone * |
| 375 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
| 376 | { |
| 377 | struct mem_cgroup *mem = pc->mem_cgroup; |
| 378 | int nid = page_cgroup_nid(pc); |
| 379 | int zid = page_cgroup_zid(pc); |
| 380 | |
| 381 | if (!mem) |
| 382 | return NULL; |
| 383 | |
| 384 | return mem_cgroup_zoneinfo(mem, nid, zid); |
| 385 | } |
| 386 | |
| 387 | static struct mem_cgroup_tree_per_zone * |
| 388 | soft_limit_tree_node_zone(int nid, int zid) |
| 389 | { |
| 390 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 391 | } |
| 392 | |
| 393 | static struct mem_cgroup_tree_per_zone * |
| 394 | soft_limit_tree_from_page(struct page *page) |
| 395 | { |
| 396 | int nid = page_to_nid(page); |
| 397 | int zid = page_zonenum(page); |
| 398 | |
| 399 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 400 | } |
| 401 | |
| 402 | static void |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 403 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 404 | struct mem_cgroup_per_zone *mz, |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 405 | struct mem_cgroup_tree_per_zone *mctz, |
| 406 | unsigned long long new_usage_in_excess) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 407 | { |
| 408 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 409 | struct rb_node *parent = NULL; |
| 410 | struct mem_cgroup_per_zone *mz_node; |
| 411 | |
| 412 | if (mz->on_tree) |
| 413 | return; |
| 414 | |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 415 | mz->usage_in_excess = new_usage_in_excess; |
| 416 | if (!mz->usage_in_excess) |
| 417 | return; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 418 | while (*p) { |
| 419 | parent = *p; |
| 420 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 421 | tree_node); |
| 422 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 423 | p = &(*p)->rb_left; |
| 424 | /* |
| 425 | * We can't avoid mem cgroups that are over their soft |
| 426 | * limit by the same amount |
| 427 | */ |
| 428 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 429 | p = &(*p)->rb_right; |
| 430 | } |
| 431 | rb_link_node(&mz->tree_node, parent, p); |
| 432 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 433 | mz->on_tree = true; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 434 | } |
| 435 | |
| 436 | static void |
| 437 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
| 438 | struct mem_cgroup_per_zone *mz, |
| 439 | struct mem_cgroup_tree_per_zone *mctz) |
| 440 | { |
| 441 | if (!mz->on_tree) |
| 442 | return; |
| 443 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 444 | mz->on_tree = false; |
| 445 | } |
| 446 | |
| 447 | static void |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 448 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
| 449 | struct mem_cgroup_per_zone *mz, |
| 450 | struct mem_cgroup_tree_per_zone *mctz) |
| 451 | { |
| 452 | spin_lock(&mctz->lock); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 453 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 454 | spin_unlock(&mctz->lock); |
| 455 | } |
| 456 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 457 | |
| 458 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) |
| 459 | { |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 460 | unsigned long long excess; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 461 | struct mem_cgroup_per_zone *mz; |
| 462 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 463 | int nid = page_to_nid(page); |
| 464 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 465 | mctz = soft_limit_tree_from_page(page); |
| 466 | |
| 467 | /* |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 468 | * Necessary to update all ancestors when hierarchy is used. |
| 469 | * because their event counter is not touched. |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 470 | */ |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 471 | for (; mem; mem = parent_mem_cgroup(mem)) { |
| 472 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 473 | excess = res_counter_soft_limit_excess(&mem->res); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 474 | /* |
| 475 | * We have to update the tree if mz is on RB-tree or |
| 476 | * mem is over its softlimit. |
| 477 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 478 | if (excess || mz->on_tree) { |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 479 | spin_lock(&mctz->lock); |
| 480 | /* if on-tree, remove it */ |
| 481 | if (mz->on_tree) |
| 482 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
| 483 | /* |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 484 | * Insert again. mz->usage_in_excess will be updated. |
| 485 | * If excess is 0, no tree ops. |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 486 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 487 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 488 | spin_unlock(&mctz->lock); |
| 489 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 490 | } |
| 491 | } |
| 492 | |
| 493 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) |
| 494 | { |
| 495 | int node, zone; |
| 496 | struct mem_cgroup_per_zone *mz; |
| 497 | struct mem_cgroup_tree_per_zone *mctz; |
| 498 | |
| 499 | for_each_node_state(node, N_POSSIBLE) { |
| 500 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 501 | mz = mem_cgroup_zoneinfo(mem, node, zone); |
| 502 | mctz = soft_limit_tree_node_zone(node, zone); |
| 503 | mem_cgroup_remove_exceeded(mem, mz, mctz); |
| 504 | } |
| 505 | } |
| 506 | } |
| 507 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 508 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
| 509 | { |
| 510 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; |
| 511 | } |
| 512 | |
| 513 | static struct mem_cgroup_per_zone * |
| 514 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 515 | { |
| 516 | struct rb_node *rightmost = NULL; |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 517 | struct mem_cgroup_per_zone *mz; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 518 | |
| 519 | retry: |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 520 | mz = NULL; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 521 | rightmost = rb_last(&mctz->rb_root); |
| 522 | if (!rightmost) |
| 523 | goto done; /* Nothing to reclaim from */ |
| 524 | |
| 525 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 526 | /* |
| 527 | * Remove the node now but someone else can add it back, |
| 528 | * we will to add it back at the end of reclaim to its correct |
| 529 | * position in the tree. |
| 530 | */ |
| 531 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
| 532 | if (!res_counter_soft_limit_excess(&mz->mem->res) || |
| 533 | !css_tryget(&mz->mem->css)) |
| 534 | goto retry; |
| 535 | done: |
| 536 | return mz; |
| 537 | } |
| 538 | |
| 539 | static struct mem_cgroup_per_zone * |
| 540 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 541 | { |
| 542 | struct mem_cgroup_per_zone *mz; |
| 543 | |
| 544 | spin_lock(&mctz->lock); |
| 545 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 546 | spin_unlock(&mctz->lock); |
| 547 | return mz; |
| 548 | } |
| 549 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 550 | /* |
| 551 | * Implementation Note: reading percpu statistics for memcg. |
| 552 | * |
| 553 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 554 | * synchronization to implement "quick" read. There are trade-off between |
| 555 | * reading cost and precision of value. Then, we may have a chance to implement |
| 556 | * a periodic synchronizion of counter in memcg's counter. |
| 557 | * |
| 558 | * But this _read() function is used for user interface now. The user accounts |
| 559 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 560 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 561 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 562 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 563 | * |
| 564 | * If there are kernel internal actions which can make use of some not-exact |
| 565 | * value, and reading all cpu value can be performance bottleneck in some |
| 566 | * common workload, threashold and synchonization as vmstat[] should be |
| 567 | * implemented. |
| 568 | */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 569 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
| 570 | enum mem_cgroup_stat_index idx) |
| 571 | { |
| 572 | int cpu; |
| 573 | s64 val = 0; |
| 574 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 575 | get_online_cpus(); |
| 576 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 577 | val += per_cpu(mem->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 578 | #ifdef CONFIG_HOTPLUG_CPU |
| 579 | spin_lock(&mem->pcp_counter_lock); |
| 580 | val += mem->nocpu_base.count[idx]; |
| 581 | spin_unlock(&mem->pcp_counter_lock); |
| 582 | #endif |
| 583 | put_online_cpus(); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 584 | return val; |
| 585 | } |
| 586 | |
| 587 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) |
| 588 | { |
| 589 | s64 ret; |
| 590 | |
| 591 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
| 592 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
| 593 | return ret; |
| 594 | } |
| 595 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 596 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
| 597 | bool charge) |
| 598 | { |
| 599 | int val = (charge) ? 1 : -1; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 600 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 601 | } |
| 602 | |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 603 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
| 604 | struct page_cgroup *pc, |
| 605 | bool charge) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 606 | { |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 607 | int val = (charge) ? 1 : -1; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 608 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 609 | preempt_disable(); |
| 610 | |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 611 | if (PageCgroupCache(pc)) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 612 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 613 | else |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 614 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 615 | |
| 616 | if (charge) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 617 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 618 | else |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 619 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 620 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 621 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 622 | preempt_enable(); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 623 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 624 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 625 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 626 | enum lru_list idx) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 627 | { |
| 628 | int nid, zid; |
| 629 | struct mem_cgroup_per_zone *mz; |
| 630 | u64 total = 0; |
| 631 | |
| 632 | for_each_online_node(nid) |
| 633 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 634 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
| 635 | total += MEM_CGROUP_ZSTAT(mz, idx); |
| 636 | } |
| 637 | return total; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 638 | } |
| 639 | |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 640 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
| 641 | { |
| 642 | s64 val; |
| 643 | |
| 644 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); |
| 645 | |
| 646 | return !(val & ((1 << event_mask_shift) - 1)); |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * Check events in order. |
| 651 | * |
| 652 | */ |
| 653 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) |
| 654 | { |
| 655 | /* threshold event is triggered in finer grain than soft limit */ |
| 656 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { |
| 657 | mem_cgroup_threshold(mem); |
| 658 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) |
| 659 | mem_cgroup_update_tree(mem, page); |
| 660 | } |
| 661 | } |
| 662 | |
Hugh Dickins | d5b69e3 | 2008-03-04 14:29:10 -0800 | [diff] [blame] | 663 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 664 | { |
| 665 | return container_of(cgroup_subsys_state(cont, |
| 666 | mem_cgroup_subsys_id), struct mem_cgroup, |
| 667 | css); |
| 668 | } |
| 669 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 670 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 671 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 672 | /* |
| 673 | * mm_update_next_owner() may clear mm->owner to NULL |
| 674 | * if it races with swapoff, page migration, etc. |
| 675 | * So this can be called with p == NULL. |
| 676 | */ |
| 677 | if (unlikely(!p)) |
| 678 | return NULL; |
| 679 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 680 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
| 681 | struct mem_cgroup, css); |
| 682 | } |
| 683 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 684 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
| 685 | { |
| 686 | struct mem_cgroup *mem = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 687 | |
| 688 | if (!mm) |
| 689 | return NULL; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 690 | /* |
| 691 | * Because we have no locks, mm->owner's may be being moved to other |
| 692 | * cgroup. We use css_tryget() here even if this looks |
| 693 | * pessimistic (rather than adding locks here). |
| 694 | */ |
| 695 | rcu_read_lock(); |
| 696 | do { |
| 697 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 698 | if (unlikely(!mem)) |
| 699 | break; |
| 700 | } while (!css_tryget(&mem->css)); |
| 701 | rcu_read_unlock(); |
| 702 | return mem; |
| 703 | } |
| 704 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 705 | /* The caller has to guarantee "mem" exists before calling this */ |
| 706 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 707 | { |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 708 | struct cgroup_subsys_state *css; |
| 709 | int found; |
| 710 | |
| 711 | if (!mem) /* ROOT cgroup has the smallest ID */ |
| 712 | return root_mem_cgroup; /*css_put/get against root is ignored*/ |
| 713 | if (!mem->use_hierarchy) { |
| 714 | if (css_tryget(&mem->css)) |
| 715 | return mem; |
| 716 | return NULL; |
| 717 | } |
| 718 | rcu_read_lock(); |
| 719 | /* |
| 720 | * searching a memory cgroup which has the smallest ID under given |
| 721 | * ROOT cgroup. (ID >= 1) |
| 722 | */ |
| 723 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); |
| 724 | if (css && css_tryget(css)) |
| 725 | mem = container_of(css, struct mem_cgroup, css); |
| 726 | else |
| 727 | mem = NULL; |
| 728 | rcu_read_unlock(); |
| 729 | return mem; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 730 | } |
| 731 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 732 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, |
| 733 | struct mem_cgroup *root, |
| 734 | bool cond) |
| 735 | { |
| 736 | int nextid = css_id(&iter->css) + 1; |
| 737 | int found; |
| 738 | int hierarchy_used; |
| 739 | struct cgroup_subsys_state *css; |
| 740 | |
| 741 | hierarchy_used = iter->use_hierarchy; |
| 742 | |
| 743 | css_put(&iter->css); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 744 | /* If no ROOT, walk all, ignore hierarchy */ |
| 745 | if (!cond || (root && !hierarchy_used)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 746 | return NULL; |
| 747 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 748 | if (!root) |
| 749 | root = root_mem_cgroup; |
| 750 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 751 | do { |
| 752 | iter = NULL; |
| 753 | rcu_read_lock(); |
| 754 | |
| 755 | css = css_get_next(&mem_cgroup_subsys, nextid, |
| 756 | &root->css, &found); |
| 757 | if (css && css_tryget(css)) |
| 758 | iter = container_of(css, struct mem_cgroup, css); |
| 759 | rcu_read_unlock(); |
| 760 | /* If css is NULL, no more cgroups will be found */ |
| 761 | nextid = found + 1; |
| 762 | } while (css && !iter); |
| 763 | |
| 764 | return iter; |
| 765 | } |
| 766 | /* |
| 767 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please |
| 768 | * be careful that "break" loop is not allowed. We have reference count. |
| 769 | * Instead of that modify "cond" to be false and "continue" to exit the loop. |
| 770 | */ |
| 771 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ |
| 772 | for (iter = mem_cgroup_start_loop(root);\ |
| 773 | iter != NULL;\ |
| 774 | iter = mem_cgroup_get_next(iter, root, cond)) |
| 775 | |
| 776 | #define for_each_mem_cgroup_tree(iter, root) \ |
| 777 | for_each_mem_cgroup_tree_cond(iter, root, true) |
| 778 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 779 | #define for_each_mem_cgroup_all(iter) \ |
| 780 | for_each_mem_cgroup_tree_cond(iter, NULL, true) |
| 781 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 782 | |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 783 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
| 784 | { |
| 785 | return (mem == root_mem_cgroup); |
| 786 | } |
| 787 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 788 | /* |
| 789 | * Following LRU functions are allowed to be used without PCG_LOCK. |
| 790 | * Operations are called by routine of global LRU independently from memcg. |
| 791 | * What we have to take care of here is validness of pc->mem_cgroup. |
| 792 | * |
| 793 | * Changes to pc->mem_cgroup happens when |
| 794 | * 1. charge |
| 795 | * 2. moving account |
| 796 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. |
| 797 | * It is added to LRU before charge. |
| 798 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. |
| 799 | * When moving account, the page is not on LRU. It's isolated. |
| 800 | */ |
| 801 | |
| 802 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 803 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 804 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 805 | struct mem_cgroup_per_zone *mz; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 806 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 807 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 808 | return; |
| 809 | pc = lookup_page_cgroup(page); |
| 810 | /* can happen while we handle swapcache. */ |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 811 | if (!TestClearPageCgroupAcctLRU(pc)) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 812 | return; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 813 | VM_BUG_ON(!pc->mem_cgroup); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 814 | /* |
| 815 | * We don't check PCG_USED bit. It's cleared when the "page" is finally |
| 816 | * removed from global LRU. |
| 817 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 818 | mz = page_cgroup_zoneinfo(pc); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 819 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 820 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
| 821 | return; |
| 822 | VM_BUG_ON(list_empty(&pc->lru)); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 823 | list_del_init(&pc->lru); |
| 824 | return; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 825 | } |
| 826 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 827 | void mem_cgroup_del_lru(struct page *page) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 828 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 829 | mem_cgroup_del_lru_list(page, page_lru(page)); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 830 | } |
| 831 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 832 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 833 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 834 | struct mem_cgroup_per_zone *mz; |
| 835 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 836 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 837 | if (mem_cgroup_disabled()) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 838 | return; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 839 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 840 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | bd112db | 2009-01-15 13:51:11 -0800 | [diff] [blame] | 841 | /* |
| 842 | * Used bit is set without atomic ops but after smp_wmb(). |
| 843 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
| 844 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 845 | smp_rmb(); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 846 | /* unused or root page is not rotated. */ |
| 847 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 848 | return; |
| 849 | mz = page_cgroup_zoneinfo(pc); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 850 | list_move(&pc->lru, &mz->lists[lru]); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 851 | } |
| 852 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 853 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
| 854 | { |
| 855 | struct page_cgroup *pc; |
| 856 | struct mem_cgroup_per_zone *mz; |
| 857 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 858 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 859 | return; |
| 860 | pc = lookup_page_cgroup(page); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 861 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
Daisuke Nishimura | bd112db | 2009-01-15 13:51:11 -0800 | [diff] [blame] | 862 | /* |
| 863 | * Used bit is set without atomic ops but after smp_wmb(). |
| 864 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
| 865 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 866 | smp_rmb(); |
| 867 | if (!PageCgroupUsed(pc)) |
| 868 | return; |
| 869 | |
| 870 | mz = page_cgroup_zoneinfo(pc); |
| 871 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 872 | SetPageCgroupAcctLRU(pc); |
| 873 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
| 874 | return; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 875 | list_add(&pc->lru, &mz->lists[lru]); |
| 876 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 877 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 878 | /* |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 879 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
| 880 | * lru because the page may.be reused after it's fully uncharged (because of |
| 881 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge |
| 882 | * it again. This function is only used to charge SwapCache. It's done under |
| 883 | * lock_page and expected that zone->lru_lock is never held. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 884 | */ |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 885 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 886 | { |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 887 | unsigned long flags; |
| 888 | struct zone *zone = page_zone(page); |
| 889 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 890 | |
| 891 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 892 | /* |
| 893 | * Forget old LRU when this page_cgroup is *not* used. This Used bit |
| 894 | * is guarded by lock_page() because the page is SwapCache. |
| 895 | */ |
| 896 | if (!PageCgroupUsed(pc)) |
| 897 | mem_cgroup_del_lru_list(page, page_lru(page)); |
| 898 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 899 | } |
| 900 | |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 901 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
| 902 | { |
| 903 | unsigned long flags; |
| 904 | struct zone *zone = page_zone(page); |
| 905 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 906 | |
| 907 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 908 | /* link when the page is linked to LRU but page_cgroup isn't */ |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 909 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 910 | mem_cgroup_add_lru_list(page, page_lru(page)); |
| 911 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 912 | } |
| 913 | |
| 914 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 915 | void mem_cgroup_move_lists(struct page *page, |
| 916 | enum lru_list from, enum lru_list to) |
| 917 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 918 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 919 | return; |
| 920 | mem_cgroup_del_lru_list(page, from); |
| 921 | mem_cgroup_add_lru_list(page, to); |
| 922 | } |
| 923 | |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 924 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
| 925 | { |
| 926 | int ret; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 927 | struct mem_cgroup *curr = NULL; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 928 | struct task_struct *p; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 929 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 930 | p = find_lock_task_mm(task); |
| 931 | if (!p) |
| 932 | return 0; |
| 933 | curr = try_get_mem_cgroup_from_mm(p->mm); |
| 934 | task_unlock(p); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 935 | if (!curr) |
| 936 | return 0; |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 937 | /* |
| 938 | * We should check use_hierarchy of "mem" not "curr". Because checking |
| 939 | * use_hierarchy of "curr" here make this function true if hierarchy is |
| 940 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* |
| 941 | * hierarchy(even if use_hierarchy is disabled in "mem"). |
| 942 | */ |
| 943 | if (mem->use_hierarchy) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 944 | ret = css_is_ancestor(&curr->css, &mem->css); |
| 945 | else |
| 946 | ret = (curr == mem); |
| 947 | css_put(&curr->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 948 | return ret; |
| 949 | } |
| 950 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 951 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 952 | { |
| 953 | unsigned long active; |
| 954 | unsigned long inactive; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 955 | unsigned long gb; |
| 956 | unsigned long inactive_ratio; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 957 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 958 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
| 959 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 960 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 961 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 962 | if (gb) |
| 963 | inactive_ratio = int_sqrt(10 * gb); |
| 964 | else |
| 965 | inactive_ratio = 1; |
| 966 | |
| 967 | if (present_pages) { |
| 968 | present_pages[0] = inactive; |
| 969 | present_pages[1] = active; |
| 970 | } |
| 971 | |
| 972 | return inactive_ratio; |
| 973 | } |
| 974 | |
| 975 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) |
| 976 | { |
| 977 | unsigned long active; |
| 978 | unsigned long inactive; |
| 979 | unsigned long present_pages[2]; |
| 980 | unsigned long inactive_ratio; |
| 981 | |
| 982 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); |
| 983 | |
| 984 | inactive = present_pages[0]; |
| 985 | active = present_pages[1]; |
| 986 | |
| 987 | if (inactive * inactive_ratio < active) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 988 | return 1; |
| 989 | |
| 990 | return 0; |
| 991 | } |
| 992 | |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 993 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
| 994 | { |
| 995 | unsigned long active; |
| 996 | unsigned long inactive; |
| 997 | |
| 998 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); |
| 999 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); |
| 1000 | |
| 1001 | return (active > inactive); |
| 1002 | } |
| 1003 | |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 1004 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
| 1005 | struct zone *zone, |
| 1006 | enum lru_list lru) |
| 1007 | { |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1008 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 1009 | int zid = zone_idx(zone); |
| 1010 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 1011 | |
| 1012 | return MEM_CGROUP_ZSTAT(mz, lru); |
| 1013 | } |
| 1014 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1015 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
| 1016 | struct zone *zone) |
| 1017 | { |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1018 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1019 | int zid = zone_idx(zone); |
| 1020 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 1021 | |
| 1022 | return &mz->reclaim_stat; |
| 1023 | } |
| 1024 | |
| 1025 | struct zone_reclaim_stat * |
| 1026 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) |
| 1027 | { |
| 1028 | struct page_cgroup *pc; |
| 1029 | struct mem_cgroup_per_zone *mz; |
| 1030 | |
| 1031 | if (mem_cgroup_disabled()) |
| 1032 | return NULL; |
| 1033 | |
| 1034 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | bd112db | 2009-01-15 13:51:11 -0800 | [diff] [blame] | 1035 | /* |
| 1036 | * Used bit is set without atomic ops but after smp_wmb(). |
| 1037 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
| 1038 | */ |
| 1039 | smp_rmb(); |
| 1040 | if (!PageCgroupUsed(pc)) |
| 1041 | return NULL; |
| 1042 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1043 | mz = page_cgroup_zoneinfo(pc); |
| 1044 | if (!mz) |
| 1045 | return NULL; |
| 1046 | |
| 1047 | return &mz->reclaim_stat; |
| 1048 | } |
| 1049 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1050 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
| 1051 | struct list_head *dst, |
| 1052 | unsigned long *scanned, int order, |
| 1053 | int mode, struct zone *z, |
| 1054 | struct mem_cgroup *mem_cont, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1055 | int active, int file) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1056 | { |
| 1057 | unsigned long nr_taken = 0; |
| 1058 | struct page *page; |
| 1059 | unsigned long scan; |
| 1060 | LIST_HEAD(pc_list); |
| 1061 | struct list_head *src; |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1062 | struct page_cgroup *pc, *tmp; |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1063 | int nid = zone_to_nid(z); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 1064 | int zid = zone_idx(z); |
| 1065 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1066 | int lru = LRU_FILE * file + active; |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1067 | int ret; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1068 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 1069 | BUG_ON(!mem_cont); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 1070 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1071 | src = &mz->lists[lru]; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1072 | |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1073 | scan = 0; |
| 1074 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1075 | if (scan >= nr_to_scan) |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1076 | break; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1077 | |
| 1078 | page = pc->page; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 1079 | if (unlikely(!PageCgroupUsed(pc))) |
| 1080 | continue; |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1081 | if (unlikely(!PageLRU(page))) |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1082 | continue; |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1083 | |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1084 | scan++; |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1085 | ret = __isolate_lru_page(page, mode, file); |
| 1086 | switch (ret) { |
| 1087 | case 0: |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1088 | list_move(&page->lru, dst); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1089 | mem_cgroup_del_lru(page); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1090 | nr_taken += hpage_nr_pages(page); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1091 | break; |
| 1092 | case -EBUSY: |
| 1093 | /* we don't affect global LRU but rotate in our LRU */ |
| 1094 | mem_cgroup_rotate_lru_list(page, page_lru(page)); |
| 1095 | break; |
| 1096 | default: |
| 1097 | break; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1098 | } |
| 1099 | } |
| 1100 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1101 | *scanned = scan; |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 1102 | |
| 1103 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, |
| 1104 | 0, 0, 0, mode); |
| 1105 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1106 | return nr_taken; |
| 1107 | } |
| 1108 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1109 | #define mem_cgroup_from_res_counter(counter, member) \ |
| 1110 | container_of(counter, struct mem_cgroup, member) |
| 1111 | |
Daisuke Nishimura | b85a96c | 2009-01-07 18:08:12 -0800 | [diff] [blame] | 1112 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
| 1113 | { |
| 1114 | if (do_swap_account) { |
| 1115 | if (res_counter_check_under_limit(&mem->res) && |
| 1116 | res_counter_check_under_limit(&mem->memsw)) |
| 1117 | return true; |
| 1118 | } else |
| 1119 | if (res_counter_check_under_limit(&mem->res)) |
| 1120 | return true; |
| 1121 | return false; |
| 1122 | } |
| 1123 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1124 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
| 1125 | { |
| 1126 | struct cgroup *cgrp = memcg->css.cgroup; |
| 1127 | unsigned int swappiness; |
| 1128 | |
| 1129 | /* root ? */ |
| 1130 | if (cgrp->parent == NULL) |
| 1131 | return vm_swappiness; |
| 1132 | |
| 1133 | spin_lock(&memcg->reclaim_param_lock); |
| 1134 | swappiness = memcg->swappiness; |
| 1135 | spin_unlock(&memcg->reclaim_param_lock); |
| 1136 | |
| 1137 | return swappiness; |
| 1138 | } |
| 1139 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1140 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
| 1141 | { |
| 1142 | int cpu; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1143 | |
| 1144 | get_online_cpus(); |
| 1145 | spin_lock(&mem->pcp_counter_lock); |
| 1146 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1147 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1148 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
| 1149 | spin_unlock(&mem->pcp_counter_lock); |
| 1150 | put_online_cpus(); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1151 | |
| 1152 | synchronize_rcu(); |
| 1153 | } |
| 1154 | |
| 1155 | static void mem_cgroup_end_move(struct mem_cgroup *mem) |
| 1156 | { |
| 1157 | int cpu; |
| 1158 | |
| 1159 | if (!mem) |
| 1160 | return; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1161 | get_online_cpus(); |
| 1162 | spin_lock(&mem->pcp_counter_lock); |
| 1163 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1164 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1165 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
| 1166 | spin_unlock(&mem->pcp_counter_lock); |
| 1167 | put_online_cpus(); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1168 | } |
| 1169 | /* |
| 1170 | * 2 routines for checking "mem" is under move_account() or not. |
| 1171 | * |
| 1172 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used |
| 1173 | * for avoiding race in accounting. If true, |
| 1174 | * pc->mem_cgroup may be overwritten. |
| 1175 | * |
| 1176 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or |
| 1177 | * under hierarchy of moving cgroups. This is for |
| 1178 | * waiting at hith-memory prressure caused by "move". |
| 1179 | */ |
| 1180 | |
| 1181 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) |
| 1182 | { |
| 1183 | VM_BUG_ON(!rcu_read_lock_held()); |
| 1184 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
| 1185 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1186 | |
| 1187 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) |
| 1188 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1189 | struct mem_cgroup *from; |
| 1190 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1191 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1192 | /* |
| 1193 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1194 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1195 | */ |
| 1196 | spin_lock(&mc.lock); |
| 1197 | from = mc.from; |
| 1198 | to = mc.to; |
| 1199 | if (!from) |
| 1200 | goto unlock; |
| 1201 | if (from == mem || to == mem |
| 1202 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) |
| 1203 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) |
| 1204 | ret = true; |
| 1205 | unlock: |
| 1206 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1207 | return ret; |
| 1208 | } |
| 1209 | |
| 1210 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) |
| 1211 | { |
| 1212 | if (mc.moving_task && current != mc.moving_task) { |
| 1213 | if (mem_cgroup_under_move(mem)) { |
| 1214 | DEFINE_WAIT(wait); |
| 1215 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1216 | /* moving charge context might have finished. */ |
| 1217 | if (mc.moving_task) |
| 1218 | schedule(); |
| 1219 | finish_wait(&mc.waitq, &wait); |
| 1220 | return true; |
| 1221 | } |
| 1222 | } |
| 1223 | return false; |
| 1224 | } |
| 1225 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1226 | /** |
Kirill A. Shutemov | 6a6135b | 2010-03-10 15:22:25 -0800 | [diff] [blame] | 1227 | * 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] | 1228 | * @memcg: The memory cgroup that went over limit |
| 1229 | * @p: Task that is going to be killed |
| 1230 | * |
| 1231 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1232 | * enabled |
| 1233 | */ |
| 1234 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1235 | { |
| 1236 | struct cgroup *task_cgrp; |
| 1237 | struct cgroup *mem_cgrp; |
| 1238 | /* |
| 1239 | * Need a buffer in BSS, can't rely on allocations. The code relies |
| 1240 | * on the assumption that OOM is serialized for memory controller. |
| 1241 | * If this assumption is broken, revisit this code. |
| 1242 | */ |
| 1243 | static char memcg_name[PATH_MAX]; |
| 1244 | int ret; |
| 1245 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1246 | if (!memcg || !p) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1247 | return; |
| 1248 | |
| 1249 | |
| 1250 | rcu_read_lock(); |
| 1251 | |
| 1252 | mem_cgrp = memcg->css.cgroup; |
| 1253 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); |
| 1254 | |
| 1255 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); |
| 1256 | if (ret < 0) { |
| 1257 | /* |
| 1258 | * Unfortunately, we are unable to convert to a useful name |
| 1259 | * But we'll still print out the usage information |
| 1260 | */ |
| 1261 | rcu_read_unlock(); |
| 1262 | goto done; |
| 1263 | } |
| 1264 | rcu_read_unlock(); |
| 1265 | |
| 1266 | printk(KERN_INFO "Task in %s killed", memcg_name); |
| 1267 | |
| 1268 | rcu_read_lock(); |
| 1269 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); |
| 1270 | if (ret < 0) { |
| 1271 | rcu_read_unlock(); |
| 1272 | goto done; |
| 1273 | } |
| 1274 | rcu_read_unlock(); |
| 1275 | |
| 1276 | /* |
| 1277 | * Continues from above, so we don't need an KERN_ level |
| 1278 | */ |
| 1279 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); |
| 1280 | done: |
| 1281 | |
| 1282 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", |
| 1283 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
| 1284 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, |
| 1285 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); |
| 1286 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " |
| 1287 | "failcnt %llu\n", |
| 1288 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
| 1289 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, |
| 1290 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); |
| 1291 | } |
| 1292 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1293 | /* |
| 1294 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1295 | * 1(self count) if no children. |
| 1296 | */ |
| 1297 | static int mem_cgroup_count_children(struct mem_cgroup *mem) |
| 1298 | { |
| 1299 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1300 | struct mem_cgroup *iter; |
| 1301 | |
| 1302 | for_each_mem_cgroup_tree(iter, mem) |
| 1303 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1304 | return num; |
| 1305 | } |
| 1306 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1307 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1308 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1309 | */ |
| 1310 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
| 1311 | { |
| 1312 | u64 limit; |
| 1313 | u64 memsw; |
| 1314 | |
Johannes Weiner | f3e8eb7 | 2011-01-13 15:47:39 -0800 | [diff] [blame^] | 1315 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 1316 | limit += total_swap_pages << PAGE_SHIFT; |
| 1317 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1318 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 1319 | /* |
| 1320 | * If memsw is finite and limits the amount of swap space available |
| 1321 | * to this memcg, return that limit. |
| 1322 | */ |
| 1323 | return min(limit, memsw); |
| 1324 | } |
| 1325 | |
| 1326 | /* |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1327 | * Visit the first child (need not be the first child as per the ordering |
| 1328 | * of the cgroup list, since we track last_scanned_child) of @mem and use |
| 1329 | * that to reclaim free pages from. |
| 1330 | */ |
| 1331 | static struct mem_cgroup * |
| 1332 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) |
| 1333 | { |
| 1334 | struct mem_cgroup *ret = NULL; |
| 1335 | struct cgroup_subsys_state *css; |
| 1336 | int nextid, found; |
| 1337 | |
| 1338 | if (!root_mem->use_hierarchy) { |
| 1339 | css_get(&root_mem->css); |
| 1340 | ret = root_mem; |
| 1341 | } |
| 1342 | |
| 1343 | while (!ret) { |
| 1344 | rcu_read_lock(); |
| 1345 | nextid = root_mem->last_scanned_child + 1; |
| 1346 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, |
| 1347 | &found); |
| 1348 | if (css && css_tryget(css)) |
| 1349 | ret = container_of(css, struct mem_cgroup, css); |
| 1350 | |
| 1351 | rcu_read_unlock(); |
| 1352 | /* Updates scanning parameter */ |
| 1353 | spin_lock(&root_mem->reclaim_param_lock); |
| 1354 | if (!css) { |
| 1355 | /* this means start scan from ID:1 */ |
| 1356 | root_mem->last_scanned_child = 0; |
| 1357 | } else |
| 1358 | root_mem->last_scanned_child = found; |
| 1359 | spin_unlock(&root_mem->reclaim_param_lock); |
| 1360 | } |
| 1361 | |
| 1362 | return ret; |
| 1363 | } |
| 1364 | |
| 1365 | /* |
| 1366 | * Scan the hierarchy if needed to reclaim memory. We remember the last child |
| 1367 | * we reclaimed from, so that we don't end up penalizing one child extensively |
| 1368 | * based on its position in the children list. |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1369 | * |
| 1370 | * root_mem is the original ancestor that we've been reclaim from. |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1371 | * |
| 1372 | * We give up and return to the caller when we visit root_mem twice. |
| 1373 | * (other groups can be removed while we're walking....) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1374 | * |
| 1375 | * If shrink==true, for avoiding to free too much, this returns immedieately. |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1376 | */ |
| 1377 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1378 | struct zone *zone, |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 1379 | gfp_t gfp_mask, |
| 1380 | unsigned long reclaim_options) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1381 | { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1382 | struct mem_cgroup *victim; |
| 1383 | int ret, total = 0; |
| 1384 | int loop = 0; |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 1385 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
| 1386 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1387 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
| 1388 | unsigned long excess = mem_cgroup_get_excess(root_mem); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1389 | |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 1390 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
| 1391 | if (root_mem->memsw_is_minimum) |
| 1392 | noswap = true; |
| 1393 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1394 | while (1) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1395 | victim = mem_cgroup_select_victim(root_mem); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1396 | if (victim == root_mem) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1397 | loop++; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1398 | if (loop >= 1) |
| 1399 | drain_all_stock_async(); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1400 | if (loop >= 2) { |
| 1401 | /* |
| 1402 | * If we have not been able to reclaim |
| 1403 | * anything, it might because there are |
| 1404 | * no reclaimable pages under this hierarchy |
| 1405 | */ |
| 1406 | if (!check_soft || !total) { |
| 1407 | css_put(&victim->css); |
| 1408 | break; |
| 1409 | } |
| 1410 | /* |
| 1411 | * We want to do more targetted reclaim. |
| 1412 | * excess >> 2 is not to excessive so as to |
| 1413 | * reclaim too much, nor too less that we keep |
| 1414 | * coming back to reclaim from this cgroup |
| 1415 | */ |
| 1416 | if (total >= (excess >> 2) || |
| 1417 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { |
| 1418 | css_put(&victim->css); |
| 1419 | break; |
| 1420 | } |
| 1421 | } |
| 1422 | } |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 1423 | if (!mem_cgroup_local_usage(victim)) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1424 | /* this cgroup's local usage == 0 */ |
| 1425 | css_put(&victim->css); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1426 | continue; |
| 1427 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1428 | /* we use swappiness of local cgroup */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1429 | if (check_soft) |
| 1430 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
KOSAKI Motohiro | 14fec79 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 1431 | noswap, get_swappiness(victim), zone); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1432 | else |
| 1433 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
| 1434 | noswap, get_swappiness(victim)); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1435 | css_put(&victim->css); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1436 | /* |
| 1437 | * At shrinking usage, we can't check we should stop here or |
| 1438 | * reclaim more. It's depends on callers. last_scanned_child |
| 1439 | * will work enough for keeping fairness under tree. |
| 1440 | */ |
| 1441 | if (shrink) |
| 1442 | return ret; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1443 | total += ret; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1444 | if (check_soft) { |
| 1445 | if (res_counter_check_under_soft_limit(&root_mem->res)) |
| 1446 | return total; |
| 1447 | } else if (mem_cgroup_check_under_limit(root_mem)) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1448 | return 1 + total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1449 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1450 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1451 | } |
| 1452 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1453 | /* |
| 1454 | * Check OOM-Killer is already running under our hierarchy. |
| 1455 | * If someone is running, return false. |
| 1456 | */ |
| 1457 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) |
| 1458 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1459 | int x, lock_count = 0; |
| 1460 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1461 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1462 | for_each_mem_cgroup_tree(iter, mem) { |
| 1463 | x = atomic_inc_return(&iter->oom_lock); |
| 1464 | lock_count = max(x, lock_count); |
| 1465 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1466 | |
| 1467 | if (lock_count == 1) |
| 1468 | return true; |
| 1469 | return false; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1470 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1471 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1472 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1473 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1474 | struct mem_cgroup *iter; |
| 1475 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1476 | /* |
| 1477 | * When a new child is created while the hierarchy is under oom, |
| 1478 | * mem_cgroup_oom_lock() may not be called. We have to use |
| 1479 | * atomic_add_unless() here. |
| 1480 | */ |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1481 | for_each_mem_cgroup_tree(iter, mem) |
| 1482 | atomic_add_unless(&iter->oom_lock, -1, 0); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1483 | return 0; |
| 1484 | } |
| 1485 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1486 | |
| 1487 | static DEFINE_MUTEX(memcg_oom_mutex); |
| 1488 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1489 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1490 | struct oom_wait_info { |
| 1491 | struct mem_cgroup *mem; |
| 1492 | wait_queue_t wait; |
| 1493 | }; |
| 1494 | |
| 1495 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1496 | unsigned mode, int sync, void *arg) |
| 1497 | { |
| 1498 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; |
| 1499 | struct oom_wait_info *oom_wait_info; |
| 1500 | |
| 1501 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
| 1502 | |
| 1503 | if (oom_wait_info->mem == wake_mem) |
| 1504 | goto wakeup; |
| 1505 | /* if no hierarchy, no match */ |
| 1506 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) |
| 1507 | return 0; |
| 1508 | /* |
| 1509 | * Both of oom_wait_info->mem and wake_mem are stable under us. |
| 1510 | * Then we can use css_is_ancestor without taking care of RCU. |
| 1511 | */ |
| 1512 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && |
| 1513 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) |
| 1514 | return 0; |
| 1515 | |
| 1516 | wakeup: |
| 1517 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1518 | } |
| 1519 | |
| 1520 | static void memcg_wakeup_oom(struct mem_cgroup *mem) |
| 1521 | { |
| 1522 | /* for filtering, pass "mem" as argument. */ |
| 1523 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); |
| 1524 | } |
| 1525 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1526 | static void memcg_oom_recover(struct mem_cgroup *mem) |
| 1527 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1528 | if (mem && atomic_read(&mem->oom_lock)) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1529 | memcg_wakeup_oom(mem); |
| 1530 | } |
| 1531 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1532 | /* |
| 1533 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. |
| 1534 | */ |
| 1535 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) |
| 1536 | { |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1537 | struct oom_wait_info owait; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1538 | bool locked, need_to_kill; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1539 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1540 | owait.mem = mem; |
| 1541 | owait.wait.flags = 0; |
| 1542 | owait.wait.func = memcg_oom_wake_function; |
| 1543 | owait.wait.private = current; |
| 1544 | INIT_LIST_HEAD(&owait.wait.task_list); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1545 | need_to_kill = true; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1546 | /* At first, try to OOM lock hierarchy under mem.*/ |
| 1547 | mutex_lock(&memcg_oom_mutex); |
| 1548 | locked = mem_cgroup_oom_lock(mem); |
| 1549 | /* |
| 1550 | * Even if signal_pending(), we can't quit charge() loop without |
| 1551 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL |
| 1552 | * under OOM is always welcomed, use TASK_KILLABLE here. |
| 1553 | */ |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1554 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
| 1555 | if (!locked || mem->oom_kill_disable) |
| 1556 | need_to_kill = false; |
| 1557 | if (locked) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1558 | mem_cgroup_oom_notify(mem); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1559 | mutex_unlock(&memcg_oom_mutex); |
| 1560 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1561 | if (need_to_kill) { |
| 1562 | finish_wait(&memcg_oom_waitq, &owait.wait); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1563 | mem_cgroup_out_of_memory(mem, mask); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1564 | } else { |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1565 | schedule(); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1566 | finish_wait(&memcg_oom_waitq, &owait.wait); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1567 | } |
| 1568 | mutex_lock(&memcg_oom_mutex); |
| 1569 | mem_cgroup_oom_unlock(mem); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1570 | memcg_wakeup_oom(mem); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1571 | mutex_unlock(&memcg_oom_mutex); |
| 1572 | |
| 1573 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
| 1574 | return false; |
| 1575 | /* Give chance to dying process */ |
| 1576 | schedule_timeout(1); |
| 1577 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1578 | } |
| 1579 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1580 | /* |
| 1581 | * Currently used to update mapped file statistics, but the routine can be |
| 1582 | * generalized to update other statistics as well. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1583 | * |
| 1584 | * Notes: Race condition |
| 1585 | * |
| 1586 | * We usually use page_cgroup_lock() for accessing page_cgroup member but |
| 1587 | * it tends to be costly. But considering some conditions, we doesn't need |
| 1588 | * to do so _always_. |
| 1589 | * |
| 1590 | * Considering "charge", lock_page_cgroup() is not required because all |
| 1591 | * file-stat operations happen after a page is attached to radix-tree. There |
| 1592 | * are no race with "charge". |
| 1593 | * |
| 1594 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup |
| 1595 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even |
| 1596 | * if there are race with "uncharge". Statistics itself is properly handled |
| 1597 | * by flags. |
| 1598 | * |
| 1599 | * Considering "move", this is an only case we see a race. To make the race |
| 1600 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are |
| 1601 | * possibility of race condition. If there is, we take a lock. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1602 | */ |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1603 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1604 | void mem_cgroup_update_page_stat(struct page *page, |
| 1605 | enum mem_cgroup_page_stat_item idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1606 | { |
| 1607 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1608 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 1609 | bool need_unlock = false; |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1610 | unsigned long uninitialized_var(flags); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1611 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1612 | if (unlikely(!pc)) |
| 1613 | return; |
| 1614 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1615 | rcu_read_lock(); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1616 | mem = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1617 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
| 1618 | goto out; |
| 1619 | /* pc->mem_cgroup is unstable ? */ |
| 1620 | if (unlikely(mem_cgroup_stealed(mem))) { |
| 1621 | /* take a lock against to access pc->mem_cgroup */ |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1622 | move_lock_page_cgroup(pc, &flags); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1623 | need_unlock = true; |
| 1624 | mem = pc->mem_cgroup; |
| 1625 | if (!mem || !PageCgroupUsed(pc)) |
| 1626 | goto out; |
| 1627 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1628 | |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1629 | switch (idx) { |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1630 | case MEMCG_NR_FILE_MAPPED: |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1631 | if (val > 0) |
| 1632 | SetPageCgroupFileMapped(pc); |
| 1633 | else if (!page_mapped(page)) |
KAMEZAWA Hiroyuki | 0c270f8 | 2010-10-27 15:33:39 -0700 | [diff] [blame] | 1634 | ClearPageCgroupFileMapped(pc); |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1635 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1636 | break; |
| 1637 | default: |
| 1638 | BUG(); |
KAMEZAWA Hiroyuki | 8725d54 | 2010-04-06 14:35:05 -0700 | [diff] [blame] | 1639 | } |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1640 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1641 | this_cpu_add(mem->stat->count[idx], val); |
| 1642 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1643 | out: |
| 1644 | if (unlikely(need_unlock)) |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1645 | move_unlock_page_cgroup(pc, &flags); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1646 | rcu_read_unlock(); |
| 1647 | return; |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1648 | } |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1649 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1650 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1651 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1652 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1653 | * TODO: maybe necessary to use big numbers in big irons. |
| 1654 | */ |
| 1655 | #define CHARGE_SIZE (32 * PAGE_SIZE) |
| 1656 | struct memcg_stock_pcp { |
| 1657 | struct mem_cgroup *cached; /* this never be root cgroup */ |
| 1658 | int charge; |
| 1659 | struct work_struct work; |
| 1660 | }; |
| 1661 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
| 1662 | static atomic_t memcg_drain_count; |
| 1663 | |
| 1664 | /* |
| 1665 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed |
| 1666 | * from local stock and true is returned. If the stock is 0 or charges from a |
| 1667 | * cgroup which is not current target, returns false. This stock will be |
| 1668 | * refilled. |
| 1669 | */ |
| 1670 | static bool consume_stock(struct mem_cgroup *mem) |
| 1671 | { |
| 1672 | struct memcg_stock_pcp *stock; |
| 1673 | bool ret = true; |
| 1674 | |
| 1675 | stock = &get_cpu_var(memcg_stock); |
| 1676 | if (mem == stock->cached && stock->charge) |
| 1677 | stock->charge -= PAGE_SIZE; |
| 1678 | else /* need to call res_counter_charge */ |
| 1679 | ret = false; |
| 1680 | put_cpu_var(memcg_stock); |
| 1681 | return ret; |
| 1682 | } |
| 1683 | |
| 1684 | /* |
| 1685 | * Returns stocks cached in percpu to res_counter and reset cached information. |
| 1686 | */ |
| 1687 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 1688 | { |
| 1689 | struct mem_cgroup *old = stock->cached; |
| 1690 | |
| 1691 | if (stock->charge) { |
| 1692 | res_counter_uncharge(&old->res, stock->charge); |
| 1693 | if (do_swap_account) |
| 1694 | res_counter_uncharge(&old->memsw, stock->charge); |
| 1695 | } |
| 1696 | stock->cached = NULL; |
| 1697 | stock->charge = 0; |
| 1698 | } |
| 1699 | |
| 1700 | /* |
| 1701 | * This must be called under preempt disabled or must be called by |
| 1702 | * a thread which is pinned to local cpu. |
| 1703 | */ |
| 1704 | static void drain_local_stock(struct work_struct *dummy) |
| 1705 | { |
| 1706 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); |
| 1707 | drain_stock(stock); |
| 1708 | } |
| 1709 | |
| 1710 | /* |
| 1711 | * 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] | 1712 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1713 | */ |
| 1714 | static void refill_stock(struct mem_cgroup *mem, int val) |
| 1715 | { |
| 1716 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 1717 | |
| 1718 | if (stock->cached != mem) { /* reset if necessary */ |
| 1719 | drain_stock(stock); |
| 1720 | stock->cached = mem; |
| 1721 | } |
| 1722 | stock->charge += val; |
| 1723 | put_cpu_var(memcg_stock); |
| 1724 | } |
| 1725 | |
| 1726 | /* |
| 1727 | * Tries to drain stocked charges in other cpus. This function is asynchronous |
| 1728 | * and just put a work per cpu for draining localy on each cpu. Caller can |
| 1729 | * expects some charges will be back to res_counter later but cannot wait for |
| 1730 | * it. |
| 1731 | */ |
| 1732 | static void drain_all_stock_async(void) |
| 1733 | { |
| 1734 | int cpu; |
| 1735 | /* This function is for scheduling "drain" in asynchronous way. |
| 1736 | * The result of "drain" is not directly handled by callers. Then, |
| 1737 | * if someone is calling drain, we don't have to call drain more. |
| 1738 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if |
| 1739 | * there is a race. We just do loose check here. |
| 1740 | */ |
| 1741 | if (atomic_read(&memcg_drain_count)) |
| 1742 | return; |
| 1743 | /* Notify other cpus that system-wide "drain" is running */ |
| 1744 | atomic_inc(&memcg_drain_count); |
| 1745 | get_online_cpus(); |
| 1746 | for_each_online_cpu(cpu) { |
| 1747 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
| 1748 | schedule_work_on(cpu, &stock->work); |
| 1749 | } |
| 1750 | put_online_cpus(); |
| 1751 | atomic_dec(&memcg_drain_count); |
| 1752 | /* We don't wait for flush_work */ |
| 1753 | } |
| 1754 | |
| 1755 | /* This is a synchronous drain interface. */ |
| 1756 | static void drain_all_stock_sync(void) |
| 1757 | { |
| 1758 | /* called when force_empty is called */ |
| 1759 | atomic_inc(&memcg_drain_count); |
| 1760 | schedule_on_each_cpu(drain_local_stock); |
| 1761 | atomic_dec(&memcg_drain_count); |
| 1762 | } |
| 1763 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1764 | /* |
| 1765 | * This function drains percpu counter value from DEAD cpu and |
| 1766 | * move it to local cpu. Note that this function can be preempted. |
| 1767 | */ |
| 1768 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) |
| 1769 | { |
| 1770 | int i; |
| 1771 | |
| 1772 | spin_lock(&mem->pcp_counter_lock); |
| 1773 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
| 1774 | s64 x = per_cpu(mem->stat->count[i], cpu); |
| 1775 | |
| 1776 | per_cpu(mem->stat->count[i], cpu) = 0; |
| 1777 | mem->nocpu_base.count[i] += x; |
| 1778 | } |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1779 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
| 1780 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
| 1781 | spin_unlock(&mem->pcp_counter_lock); |
| 1782 | } |
| 1783 | |
| 1784 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) |
| 1785 | { |
| 1786 | int idx = MEM_CGROUP_ON_MOVE; |
| 1787 | |
| 1788 | spin_lock(&mem->pcp_counter_lock); |
| 1789 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1790 | spin_unlock(&mem->pcp_counter_lock); |
| 1791 | } |
| 1792 | |
| 1793 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1794 | unsigned long action, |
| 1795 | void *hcpu) |
| 1796 | { |
| 1797 | int cpu = (unsigned long)hcpu; |
| 1798 | struct memcg_stock_pcp *stock; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1799 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1800 | |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1801 | if ((action == CPU_ONLINE)) { |
| 1802 | for_each_mem_cgroup_all(iter) |
| 1803 | synchronize_mem_cgroup_on_move(iter, cpu); |
| 1804 | return NOTIFY_OK; |
| 1805 | } |
| 1806 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1807 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1808 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1809 | |
| 1810 | for_each_mem_cgroup_all(iter) |
| 1811 | mem_cgroup_drain_pcp_counter(iter, cpu); |
| 1812 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1813 | stock = &per_cpu(memcg_stock, cpu); |
| 1814 | drain_stock(stock); |
| 1815 | return NOTIFY_OK; |
| 1816 | } |
| 1817 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1818 | |
| 1819 | /* See __mem_cgroup_try_charge() for details */ |
| 1820 | enum { |
| 1821 | CHARGE_OK, /* success */ |
| 1822 | CHARGE_RETRY, /* need to retry but retry is not bad */ |
| 1823 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ |
| 1824 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ |
| 1825 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ |
| 1826 | }; |
| 1827 | |
| 1828 | static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, |
| 1829 | int csize, bool oom_check) |
| 1830 | { |
| 1831 | struct mem_cgroup *mem_over_limit; |
| 1832 | struct res_counter *fail_res; |
| 1833 | unsigned long flags = 0; |
| 1834 | int ret; |
| 1835 | |
| 1836 | ret = res_counter_charge(&mem->res, csize, &fail_res); |
| 1837 | |
| 1838 | if (likely(!ret)) { |
| 1839 | if (!do_swap_account) |
| 1840 | return CHARGE_OK; |
| 1841 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
| 1842 | if (likely(!ret)) |
| 1843 | return CHARGE_OK; |
| 1844 | |
| 1845 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
| 1846 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
| 1847 | } else |
| 1848 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); |
| 1849 | |
| 1850 | if (csize > PAGE_SIZE) /* change csize and retry */ |
| 1851 | return CHARGE_RETRY; |
| 1852 | |
| 1853 | if (!(gfp_mask & __GFP_WAIT)) |
| 1854 | return CHARGE_WOULDBLOCK; |
| 1855 | |
| 1856 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
| 1857 | gfp_mask, flags); |
| 1858 | /* |
| 1859 | * try_to_free_mem_cgroup_pages() might not give us a full |
| 1860 | * picture of reclaim. Some pages are reclaimed and might be |
| 1861 | * moved to swap cache or just unmapped from the cgroup. |
| 1862 | * Check the limit again to see if the reclaim reduced the |
| 1863 | * current usage of the cgroup before giving up |
| 1864 | */ |
| 1865 | if (ret || mem_cgroup_check_under_limit(mem_over_limit)) |
| 1866 | return CHARGE_RETRY; |
| 1867 | |
| 1868 | /* |
| 1869 | * At task move, charge accounts can be doubly counted. So, it's |
| 1870 | * better to wait until the end of task_move if something is going on. |
| 1871 | */ |
| 1872 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 1873 | return CHARGE_RETRY; |
| 1874 | |
| 1875 | /* If we don't need to call oom-killer at el, return immediately */ |
| 1876 | if (!oom_check) |
| 1877 | return CHARGE_NOMEM; |
| 1878 | /* check OOM */ |
| 1879 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) |
| 1880 | return CHARGE_OOM_DIE; |
| 1881 | |
| 1882 | return CHARGE_RETRY; |
| 1883 | } |
| 1884 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1885 | /* |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1886 | * Unlike exported interface, "oom" parameter is added. if oom==true, |
| 1887 | * oom-killer can be invoked. |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1888 | */ |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1889 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1890 | gfp_t gfp_mask, |
| 1891 | struct mem_cgroup **memcg, bool oom, |
| 1892 | int page_size) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1893 | { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1894 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 1895 | struct mem_cgroup *mem = NULL; |
| 1896 | int ret; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1897 | int csize = max(CHARGE_SIZE, (unsigned long) page_size); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1898 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1899 | /* |
| 1900 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage |
| 1901 | * in system level. So, allow to go ahead dying process in addition to |
| 1902 | * MEMDIE process. |
| 1903 | */ |
| 1904 | if (unlikely(test_thread_flag(TIF_MEMDIE) |
| 1905 | || fatal_signal_pending(current))) |
| 1906 | goto bypass; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1907 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1908 | /* |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 1909 | * We always charge the cgroup the mm_struct belongs to. |
| 1910 | * The mm_struct's mem_cgroup changes on task migration if the |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1911 | * thread group leader migrates. It's possible that mm is not |
| 1912 | * set, if so charge the init_mm (happens for pagecache usage). |
| 1913 | */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1914 | if (!*memcg && !mm) |
| 1915 | goto bypass; |
| 1916 | again: |
| 1917 | if (*memcg) { /* css should be a valid one */ |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1918 | mem = *memcg; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1919 | VM_BUG_ON(css_is_removed(&mem->css)); |
| 1920 | if (mem_cgroup_is_root(mem)) |
| 1921 | goto done; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1922 | if (page_size == PAGE_SIZE && consume_stock(mem)) |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1923 | goto done; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1924 | css_get(&mem->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1925 | } else { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1926 | struct task_struct *p; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1927 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1928 | rcu_read_lock(); |
| 1929 | p = rcu_dereference(mm->owner); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1930 | /* |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 1931 | * Because we don't have task_lock(), "p" can exit. |
| 1932 | * In that case, "mem" can point to root or p can be NULL with |
| 1933 | * race with swapoff. Then, we have small risk of mis-accouning. |
| 1934 | * But such kind of mis-account by race always happens because |
| 1935 | * we don't have cgroup_mutex(). It's overkill and we allo that |
| 1936 | * small race, here. |
| 1937 | * (*) swapoff at el will charge against mm-struct not against |
| 1938 | * task-struct. So, mm->owner can be NULL. |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1939 | */ |
| 1940 | mem = mem_cgroup_from_task(p); |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 1941 | if (!mem || mem_cgroup_is_root(mem)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1942 | rcu_read_unlock(); |
| 1943 | goto done; |
| 1944 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1945 | if (page_size == PAGE_SIZE && consume_stock(mem)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1946 | /* |
| 1947 | * It seems dagerous to access memcg without css_get(). |
| 1948 | * But considering how consume_stok works, it's not |
| 1949 | * necessary. If consume_stock success, some charges |
| 1950 | * from this memcg are cached on this cpu. So, we |
| 1951 | * don't need to call css_get()/css_tryget() before |
| 1952 | * calling consume_stock(). |
| 1953 | */ |
| 1954 | rcu_read_unlock(); |
| 1955 | goto done; |
| 1956 | } |
| 1957 | /* after here, we may be blocked. we need to get refcnt */ |
| 1958 | if (!css_tryget(&mem->css)) { |
| 1959 | rcu_read_unlock(); |
| 1960 | goto again; |
| 1961 | } |
| 1962 | rcu_read_unlock(); |
| 1963 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1964 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1965 | do { |
| 1966 | bool oom_check; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1967 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1968 | /* If killed, bypass charge */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1969 | if (fatal_signal_pending(current)) { |
| 1970 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1971 | goto bypass; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1972 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1973 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1974 | oom_check = false; |
| 1975 | if (oom && !nr_oom_retries) { |
| 1976 | oom_check = true; |
| 1977 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 1978 | } |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1979 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1980 | ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check); |
| 1981 | |
| 1982 | switch (ret) { |
| 1983 | case CHARGE_OK: |
| 1984 | break; |
| 1985 | case CHARGE_RETRY: /* not in OOM situation but retry */ |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1986 | csize = page_size; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1987 | css_put(&mem->css); |
| 1988 | mem = NULL; |
| 1989 | goto again; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1990 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1991 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1992 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1993 | case CHARGE_NOMEM: /* OOM routine works */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1994 | if (!oom) { |
| 1995 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1996 | goto nomem; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1997 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1998 | /* If oom, we never return -ENOMEM */ |
| 1999 | nr_oom_retries--; |
| 2000 | break; |
| 2001 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2002 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2003 | goto bypass; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2004 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2005 | } while (ret != CHARGE_OK); |
| 2006 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2007 | if (csize > page_size) |
| 2008 | refill_stock(mem, csize - page_size); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2009 | css_put(&mem->css); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2010 | done: |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2011 | *memcg = mem; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2012 | return 0; |
| 2013 | nomem: |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2014 | *memcg = NULL; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2015 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2016 | bypass: |
| 2017 | *memcg = NULL; |
| 2018 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2019 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2020 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2021 | /* |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2022 | * Somemtimes we have to undo a charge we got by try_charge(). |
| 2023 | * This function is for that and do uncharge, put css's refcnt. |
| 2024 | * gotten by try_charge(). |
| 2025 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2026 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
| 2027 | unsigned long count) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2028 | { |
| 2029 | if (!mem_cgroup_is_root(mem)) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2030 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2031 | if (do_swap_account) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2032 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2033 | } |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2034 | } |
| 2035 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2036 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
| 2037 | int page_size) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2038 | { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2039 | __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2040 | } |
| 2041 | |
| 2042 | /* |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2043 | * A helper function to get mem_cgroup from ID. must be called under |
| 2044 | * rcu_read_lock(). The caller must check css_is_removed() or some if |
| 2045 | * it's concern. (dropping refcnt from swap can be called against removed |
| 2046 | * memcg.) |
| 2047 | */ |
| 2048 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
| 2049 | { |
| 2050 | struct cgroup_subsys_state *css; |
| 2051 | |
| 2052 | /* ID 0 is unused ID */ |
| 2053 | if (!id) |
| 2054 | return NULL; |
| 2055 | css = css_lookup(&mem_cgroup_subsys, id); |
| 2056 | if (!css) |
| 2057 | return NULL; |
| 2058 | return container_of(css, struct mem_cgroup, css); |
| 2059 | } |
| 2060 | |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2061 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2062 | { |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2063 | struct mem_cgroup *mem = NULL; |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2064 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2065 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2066 | swp_entry_t ent; |
| 2067 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2068 | VM_BUG_ON(!PageLocked(page)); |
| 2069 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2070 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2071 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2072 | if (PageCgroupUsed(pc)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2073 | mem = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2074 | if (mem && !css_tryget(&mem->css)) |
| 2075 | mem = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2076 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2077 | ent.val = page_private(page); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2078 | id = lookup_swap_cgroup(ent); |
| 2079 | rcu_read_lock(); |
| 2080 | mem = mem_cgroup_lookup(id); |
| 2081 | if (mem && !css_tryget(&mem->css)) |
| 2082 | mem = NULL; |
| 2083 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2084 | } |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2085 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2086 | return mem; |
| 2087 | } |
| 2088 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2089 | /* |
Daisuke Nishimura | a5e924f | 2009-01-07 18:08:28 -0800 | [diff] [blame] | 2090 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2091 | * USED state. If already USED, uncharge and return. |
| 2092 | */ |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2093 | static void ____mem_cgroup_commit_charge(struct mem_cgroup *mem, |
| 2094 | struct page_cgroup *pc, |
| 2095 | enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2096 | { |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2097 | pc->mem_cgroup = mem; |
KAMEZAWA Hiroyuki | 261fb61 | 2009-09-23 15:56:33 -0700 | [diff] [blame] | 2098 | /* |
| 2099 | * We access a page_cgroup asynchronously without lock_page_cgroup(). |
| 2100 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup |
| 2101 | * is accessed after testing USED bit. To make pc->mem_cgroup visible |
| 2102 | * before USED bit, we need memory barrier here. |
| 2103 | * See mem_cgroup_add_lru_list(), etc. |
| 2104 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2105 | smp_wmb(); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 2106 | switch (ctype) { |
| 2107 | case MEM_CGROUP_CHARGE_TYPE_CACHE: |
| 2108 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: |
| 2109 | SetPageCgroupCache(pc); |
| 2110 | SetPageCgroupUsed(pc); |
| 2111 | break; |
| 2112 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
| 2113 | ClearPageCgroupCache(pc); |
| 2114 | SetPageCgroupUsed(pc); |
| 2115 | break; |
| 2116 | default: |
| 2117 | break; |
| 2118 | } |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2119 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2120 | mem_cgroup_charge_statistics(mem, pc, true); |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2121 | } |
| 2122 | |
| 2123 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
| 2124 | struct page_cgroup *pc, |
| 2125 | enum charge_type ctype, |
| 2126 | int page_size) |
| 2127 | { |
| 2128 | int i; |
| 2129 | int count = page_size >> PAGE_SHIFT; |
| 2130 | |
| 2131 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
| 2132 | if (!mem) |
| 2133 | return; |
| 2134 | |
| 2135 | lock_page_cgroup(pc); |
| 2136 | if (unlikely(PageCgroupUsed(pc))) { |
| 2137 | unlock_page_cgroup(pc); |
| 2138 | mem_cgroup_cancel_charge(mem, page_size); |
| 2139 | return; |
| 2140 | } |
| 2141 | |
| 2142 | /* |
| 2143 | * we don't need page_cgroup_lock about tail pages, becase they are not |
| 2144 | * accessed by any other context at this point. |
| 2145 | */ |
| 2146 | for (i = 0; i < count; i++) |
| 2147 | ____mem_cgroup_commit_charge(mem, pc + i, ctype); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2148 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2149 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 430e4863 | 2010-03-10 15:22:30 -0800 | [diff] [blame] | 2150 | /* |
| 2151 | * "charge_statistics" updated event counter. Then, check it. |
| 2152 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
| 2153 | * if they exceeds softlimit. |
| 2154 | */ |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2155 | memcg_check_events(mem, pc->page); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2156 | } |
| 2157 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2158 | /** |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2159 | * __mem_cgroup_move_account - move account of the page |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2160 | * @pc: page_cgroup of the page. |
| 2161 | * @from: mem_cgroup which the page is moved from. |
| 2162 | * @to: mem_cgroup which the page is moved to. @from != @to. |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2163 | * @uncharge: whether we should call uncharge and css_put against @from. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2164 | * |
| 2165 | * The caller must confirm following. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2166 | * - page is not on LRU (isolate_page() is useful.) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2167 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2168 | * |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2169 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
| 2170 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is |
| 2171 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
| 2172 | * @uncharge is false, so a caller should do "uncharge". |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2173 | */ |
| 2174 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2175 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2176 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2177 | { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2178 | VM_BUG_ON(from == to); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2179 | VM_BUG_ON(PageLRU(pc->page)); |
Kirill A. Shutemov | 112bc2e | 2010-11-24 12:56:58 -0800 | [diff] [blame] | 2180 | VM_BUG_ON(!page_is_cgroup_locked(pc)); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2181 | VM_BUG_ON(!PageCgroupUsed(pc)); |
| 2182 | VM_BUG_ON(pc->mem_cgroup != from); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2183 | |
KAMEZAWA Hiroyuki | 8725d54 | 2010-04-06 14:35:05 -0700 | [diff] [blame] | 2184 | if (PageCgroupFileMapped(pc)) { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 2185 | /* Update mapped_file data for mem_cgroup */ |
| 2186 | preempt_disable(); |
| 2187 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2188 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2189 | preempt_enable(); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2190 | } |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2191 | mem_cgroup_charge_statistics(from, pc, false); |
| 2192 | if (uncharge) |
| 2193 | /* This is not "cancel", but cancel_charge does all we need. */ |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2194 | mem_cgroup_cancel_charge(from, PAGE_SIZE); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2195 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2196 | /* caller should have done css_get */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2197 | pc->mem_cgroup = to; |
| 2198 | mem_cgroup_charge_statistics(to, pc, true); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2199 | /* |
| 2200 | * We charges against "to" which may not have any tasks. Then, "to" |
| 2201 | * can be under rmdir(). But in current implementation, caller of |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 2202 | * this function is just force_empty() and move charge, so it's |
| 2203 | * garanteed that "to" is never removed. So, we don't check rmdir |
| 2204 | * status here. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2205 | */ |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2206 | } |
| 2207 | |
| 2208 | /* |
| 2209 | * check whether the @pc is valid for moving account and call |
| 2210 | * __mem_cgroup_move_account() |
| 2211 | */ |
| 2212 | static int mem_cgroup_move_account(struct page_cgroup *pc, |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2213 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2214 | { |
| 2215 | int ret = -EINVAL; |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2216 | unsigned long flags; |
| 2217 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2218 | lock_page_cgroup(pc); |
| 2219 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2220 | move_lock_page_cgroup(pc, &flags); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2221 | __mem_cgroup_move_account(pc, from, to, uncharge); |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2222 | move_unlock_page_cgroup(pc, &flags); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2223 | ret = 0; |
| 2224 | } |
| 2225 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2226 | /* |
| 2227 | * check events |
| 2228 | */ |
| 2229 | memcg_check_events(to, pc->page); |
| 2230 | memcg_check_events(from, pc->page); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2231 | return ret; |
| 2232 | } |
| 2233 | |
| 2234 | /* |
| 2235 | * move charges to its parent. |
| 2236 | */ |
| 2237 | |
| 2238 | static int mem_cgroup_move_parent(struct page_cgroup *pc, |
| 2239 | struct mem_cgroup *child, |
| 2240 | gfp_t gfp_mask) |
| 2241 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2242 | struct page *page = pc->page; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2243 | struct cgroup *cg = child->css.cgroup; |
| 2244 | struct cgroup *pcg = cg->parent; |
| 2245 | struct mem_cgroup *parent; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2246 | int ret; |
| 2247 | |
| 2248 | /* Is ROOT ? */ |
| 2249 | if (!pcg) |
| 2250 | return -EINVAL; |
| 2251 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2252 | ret = -EBUSY; |
| 2253 | if (!get_page_unless_zero(page)) |
| 2254 | goto out; |
| 2255 | if (isolate_lru_page(page)) |
| 2256 | goto put; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2257 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2258 | parent = mem_cgroup_from_cont(pcg); |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2259 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, |
| 2260 | PAGE_SIZE); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2261 | if (ret || !parent) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2262 | goto put_back; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2263 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2264 | ret = mem_cgroup_move_account(pc, child, parent, true); |
| 2265 | if (ret) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2266 | mem_cgroup_cancel_charge(parent, PAGE_SIZE); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2267 | put_back: |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2268 | putback_lru_page(page); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2269 | put: |
Daisuke Nishimura | 40d5813 | 2009-01-15 13:51:12 -0800 | [diff] [blame] | 2270 | put_page(page); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2271 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2272 | return ret; |
| 2273 | } |
| 2274 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2275 | /* |
| 2276 | * Charge the memory controller for page usage. |
| 2277 | * Return |
| 2278 | * 0 if the charge was successful |
| 2279 | * < 0 if the cgroup is over its limit |
| 2280 | */ |
| 2281 | 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] | 2282 | gfp_t gfp_mask, enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2283 | { |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2284 | struct mem_cgroup *mem = NULL; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2285 | struct page_cgroup *pc; |
| 2286 | int ret; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2287 | int page_size = PAGE_SIZE; |
| 2288 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2289 | if (PageTransHuge(page)) { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2290 | page_size <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2291 | VM_BUG_ON(!PageTransHuge(page)); |
| 2292 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2293 | |
| 2294 | pc = lookup_page_cgroup(page); |
| 2295 | /* can happen at boot */ |
| 2296 | if (unlikely(!pc)) |
| 2297 | return 0; |
| 2298 | prefetchw(pc); |
| 2299 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2300 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page_size); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2301 | if (ret || !mem) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2302 | return ret; |
| 2303 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2304 | __mem_cgroup_commit_charge(mem, pc, ctype, page_size); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2305 | return 0; |
| 2306 | } |
| 2307 | |
| 2308 | int mem_cgroup_newpage_charge(struct page *page, |
| 2309 | struct mm_struct *mm, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2310 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2311 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2312 | return 0; |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2313 | /* |
| 2314 | * If already mapped, we don't have to account. |
| 2315 | * If page cache, page->mapping has address_space. |
| 2316 | * But page->mapping may have out-of-use anon_vma pointer, |
| 2317 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping |
| 2318 | * is NULL. |
| 2319 | */ |
| 2320 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) |
| 2321 | return 0; |
| 2322 | if (unlikely(!mm)) |
| 2323 | mm = &init_mm; |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2324 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2325 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2326 | } |
| 2327 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2328 | static void |
| 2329 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
| 2330 | enum charge_type ctype); |
| 2331 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2332 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
| 2333 | gfp_t gfp_mask) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2334 | { |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2335 | int ret; |
| 2336 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2337 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2338 | return 0; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2339 | if (PageCompound(page)) |
| 2340 | return 0; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2341 | /* |
| 2342 | * Corner case handling. This is called from add_to_page_cache() |
| 2343 | * in usual. But some FS (shmem) precharges this page before calling it |
| 2344 | * and call add_to_page_cache() with GFP_NOWAIT. |
| 2345 | * |
| 2346 | * For GFP_NOWAIT case, the page may be pre-charged before calling |
| 2347 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call |
| 2348 | * charge twice. (It works but has to pay a bit larger cost.) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2349 | * And when the page is SwapCache, it should take swap information |
| 2350 | * into account. This is under lock_page() now. |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2351 | */ |
| 2352 | if (!(gfp_mask & __GFP_WAIT)) { |
| 2353 | struct page_cgroup *pc; |
| 2354 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2355 | pc = lookup_page_cgroup(page); |
| 2356 | if (!pc) |
| 2357 | return 0; |
| 2358 | lock_page_cgroup(pc); |
| 2359 | if (PageCgroupUsed(pc)) { |
| 2360 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2361 | return 0; |
| 2362 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2363 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2364 | } |
| 2365 | |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2366 | if (unlikely(!mm)) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2367 | mm = &init_mm; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2368 | |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 2369 | if (page_is_file_cache(page)) |
| 2370 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2371 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2372 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2373 | /* shmem */ |
| 2374 | if (PageSwapCache(page)) { |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2375 | struct mem_cgroup *mem = NULL; |
| 2376 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2377 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
| 2378 | if (!ret) |
| 2379 | __mem_cgroup_commit_charge_swapin(page, mem, |
| 2380 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
| 2381 | } else |
| 2382 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2383 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2384 | |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2385 | return ret; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2386 | } |
| 2387 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2388 | /* |
| 2389 | * While swap-in, try_charge -> commit or cancel, the page is locked. |
| 2390 | * And when try_charge() successfully returns, one refcnt to memcg without |
Uwe Kleine-König | 21ae295 | 2009-10-07 15:21:09 +0200 | [diff] [blame] | 2391 | * struct page_cgroup is acquired. This refcnt will be consumed by |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2392 | * "commit()" or removed by "cancel()" |
| 2393 | */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2394 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
| 2395 | struct page *page, |
| 2396 | gfp_t mask, struct mem_cgroup **ptr) |
| 2397 | { |
| 2398 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2399 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2400 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2401 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2402 | return 0; |
| 2403 | |
| 2404 | if (!do_swap_account) |
| 2405 | goto charge_cur_mm; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2406 | /* |
| 2407 | * A racing thread's fault, or swapoff, may have already updated |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2408 | * the pte, and even removed page from swap cache: in those cases |
| 2409 | * do_swap_page()'s pte_same() test will fail; but there's also a |
| 2410 | * KSM case which does need to charge the page. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2411 | */ |
| 2412 | if (!PageSwapCache(page)) |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2413 | goto charge_cur_mm; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2414 | mem = try_get_mem_cgroup_from_page(page); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2415 | if (!mem) |
| 2416 | goto charge_cur_mm; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2417 | *ptr = mem; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2418 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2419 | css_put(&mem->css); |
| 2420 | return ret; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2421 | charge_cur_mm: |
| 2422 | if (unlikely(!mm)) |
| 2423 | mm = &init_mm; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2424 | return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2425 | } |
| 2426 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2427 | static void |
| 2428 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
| 2429 | enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2430 | { |
| 2431 | struct page_cgroup *pc; |
| 2432 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2433 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2434 | return; |
| 2435 | if (!ptr) |
| 2436 | return; |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2437 | cgroup_exclude_rmdir(&ptr->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2438 | pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2439 | mem_cgroup_lru_del_before_commit_swapcache(page); |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2440 | __mem_cgroup_commit_charge(ptr, pc, ctype, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2441 | mem_cgroup_lru_add_after_commit_swapcache(page); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2442 | /* |
| 2443 | * Now swap is on-memory. This means this page may be |
| 2444 | * counted both as mem and swap....double count. |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2445 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
| 2446 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() |
| 2447 | * may call delete_from_swap_cache() before reach here. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2448 | */ |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2449 | if (do_swap_account && PageSwapCache(page)) { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2450 | swp_entry_t ent = {.val = page_private(page)}; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2451 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2452 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2453 | |
| 2454 | id = swap_cgroup_record(ent, 0); |
| 2455 | rcu_read_lock(); |
| 2456 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2457 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2458 | /* |
| 2459 | * This recorded memcg can be obsolete one. So, avoid |
| 2460 | * calling css_tryget |
| 2461 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2462 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 2463 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2464 | mem_cgroup_swap_statistics(memcg, false); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2465 | mem_cgroup_put(memcg); |
| 2466 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2467 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2468 | } |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2469 | /* |
| 2470 | * At swapin, we may charge account against cgroup which has no tasks. |
| 2471 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 2472 | * In that case, we need to call pre_destroy() again. check it here. |
| 2473 | */ |
| 2474 | cgroup_release_and_wakeup_rmdir(&ptr->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2475 | } |
| 2476 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2477 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
| 2478 | { |
| 2479 | __mem_cgroup_commit_charge_swapin(page, ptr, |
| 2480 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 2481 | } |
| 2482 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2483 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
| 2484 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2485 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2486 | return; |
| 2487 | if (!mem) |
| 2488 | return; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2489 | mem_cgroup_cancel_charge(mem, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2490 | } |
| 2491 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2492 | static void |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2493 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype, |
| 2494 | int page_size) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2495 | { |
| 2496 | struct memcg_batch_info *batch = NULL; |
| 2497 | bool uncharge_memsw = true; |
| 2498 | /* If swapout, usage of swap doesn't decrease */ |
| 2499 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
| 2500 | uncharge_memsw = false; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2501 | |
| 2502 | batch = ¤t->memcg_batch; |
| 2503 | /* |
| 2504 | * In usual, we do css_get() when we remember memcg pointer. |
| 2505 | * But in this case, we keep res->usage until end of a series of |
| 2506 | * uncharges. Then, it's ok to ignore memcg's refcnt. |
| 2507 | */ |
| 2508 | if (!batch->memcg) |
| 2509 | batch->memcg = mem; |
| 2510 | /* |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2511 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. |
| 2512 | * In those cases, all pages freed continously can be expected to be in |
| 2513 | * the same cgroup and we have chance to coalesce uncharges. |
| 2514 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) |
| 2515 | * because we want to do uncharge as soon as possible. |
| 2516 | */ |
| 2517 | |
| 2518 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) |
| 2519 | goto direct_uncharge; |
| 2520 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2521 | if (page_size != PAGE_SIZE) |
| 2522 | goto direct_uncharge; |
| 2523 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2524 | /* |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2525 | * In typical case, batch->memcg == mem. This means we can |
| 2526 | * merge a series of uncharges to an uncharge of res_counter. |
| 2527 | * If not, we uncharge res_counter ony by one. |
| 2528 | */ |
| 2529 | if (batch->memcg != mem) |
| 2530 | goto direct_uncharge; |
| 2531 | /* remember freed charge and uncharge it later */ |
| 2532 | batch->bytes += PAGE_SIZE; |
| 2533 | if (uncharge_memsw) |
| 2534 | batch->memsw_bytes += PAGE_SIZE; |
| 2535 | return; |
| 2536 | direct_uncharge: |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2537 | res_counter_uncharge(&mem->res, page_size); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2538 | if (uncharge_memsw) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2539 | res_counter_uncharge(&mem->memsw, page_size); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2540 | if (unlikely(batch->memcg != mem)) |
| 2541 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2542 | return; |
| 2543 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2544 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2545 | /* |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2546 | * uncharge if !page_mapped(page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2547 | */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2548 | static struct mem_cgroup * |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2549 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2550 | { |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2551 | int i; |
| 2552 | int count; |
Hugh Dickins | 8289546 | 2008-03-04 14:29:08 -0800 | [diff] [blame] | 2553 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2554 | struct mem_cgroup *mem = NULL; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2555 | int page_size = PAGE_SIZE; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2556 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2557 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2558 | return NULL; |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 2559 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2560 | if (PageSwapCache(page)) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2561 | return NULL; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2562 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2563 | if (PageTransHuge(page)) { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2564 | page_size <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2565 | VM_BUG_ON(!PageTransHuge(page)); |
| 2566 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2567 | |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2568 | count = page_size >> PAGE_SHIFT; |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2569 | /* |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 2570 | * Check if our page_cgroup is valid |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2571 | */ |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2572 | pc = lookup_page_cgroup(page); |
| 2573 | if (unlikely(!pc || !PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2574 | return NULL; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2575 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2576 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2577 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2578 | mem = pc->mem_cgroup; |
| 2579 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2580 | if (!PageCgroupUsed(pc)) |
| 2581 | goto unlock_out; |
| 2582 | |
| 2583 | switch (ctype) { |
| 2584 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2585 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2586 | /* See mem_cgroup_prepare_migration() */ |
| 2587 | if (page_mapped(page) || PageCgroupMigration(pc)) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2588 | goto unlock_out; |
| 2589 | break; |
| 2590 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: |
| 2591 | if (!PageAnon(page)) { /* Shared memory */ |
| 2592 | if (page->mapping && !page_is_file_cache(page)) |
| 2593 | goto unlock_out; |
| 2594 | } else if (page_mapped(page)) /* Anon */ |
| 2595 | goto unlock_out; |
| 2596 | break; |
| 2597 | default: |
| 2598 | break; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2599 | } |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2600 | |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2601 | for (i = 0; i < count; i++) |
| 2602 | mem_cgroup_charge_statistics(mem, pc + i, false); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 2603 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2604 | ClearPageCgroupUsed(pc); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2605 | /* |
| 2606 | * pc->mem_cgroup is not cleared here. It will be accessed when it's |
| 2607 | * freed from LRU. This is safe because uncharged page is expected not |
| 2608 | * to be reused (freed soon). Exception is SwapCache, it's handled by |
| 2609 | * special functions. |
| 2610 | */ |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 2611 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2612 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2613 | /* |
| 2614 | * even after unlock, we have mem->res.usage here and this memcg |
| 2615 | * will never be freed. |
| 2616 | */ |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2617 | memcg_check_events(mem, page); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2618 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
| 2619 | mem_cgroup_swap_statistics(mem, true); |
| 2620 | mem_cgroup_get(mem); |
| 2621 | } |
| 2622 | if (!mem_cgroup_is_root(mem)) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2623 | __do_uncharge(mem, ctype, page_size); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 2624 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2625 | return mem; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2626 | |
| 2627 | unlock_out: |
| 2628 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2629 | return NULL; |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 2630 | } |
| 2631 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2632 | void mem_cgroup_uncharge_page(struct page *page) |
| 2633 | { |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2634 | /* early check. */ |
| 2635 | if (page_mapped(page)) |
| 2636 | return; |
| 2637 | if (page->mapping && !PageAnon(page)) |
| 2638 | return; |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2639 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 2640 | } |
| 2641 | |
| 2642 | void mem_cgroup_uncharge_cache_page(struct page *page) |
| 2643 | { |
| 2644 | VM_BUG_ON(page_mapped(page)); |
KAMEZAWA Hiroyuki | b7abea9 | 2008-10-18 20:28:09 -0700 | [diff] [blame] | 2645 | VM_BUG_ON(page->mapping); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2646 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
| 2647 | } |
| 2648 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2649 | /* |
| 2650 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. |
| 2651 | * In that cases, pages are freed continuously and we can expect pages |
| 2652 | * are in the same memcg. All these calls itself limits the number of |
| 2653 | * pages freed at once, then uncharge_start/end() is called properly. |
| 2654 | * This may be called prural(2) times in a context, |
| 2655 | */ |
| 2656 | |
| 2657 | void mem_cgroup_uncharge_start(void) |
| 2658 | { |
| 2659 | current->memcg_batch.do_batch++; |
| 2660 | /* We can do nest. */ |
| 2661 | if (current->memcg_batch.do_batch == 1) { |
| 2662 | current->memcg_batch.memcg = NULL; |
| 2663 | current->memcg_batch.bytes = 0; |
| 2664 | current->memcg_batch.memsw_bytes = 0; |
| 2665 | } |
| 2666 | } |
| 2667 | |
| 2668 | void mem_cgroup_uncharge_end(void) |
| 2669 | { |
| 2670 | struct memcg_batch_info *batch = ¤t->memcg_batch; |
| 2671 | |
| 2672 | if (!batch->do_batch) |
| 2673 | return; |
| 2674 | |
| 2675 | batch->do_batch--; |
| 2676 | if (batch->do_batch) /* If stacked, do nothing. */ |
| 2677 | return; |
| 2678 | |
| 2679 | if (!batch->memcg) |
| 2680 | return; |
| 2681 | /* |
| 2682 | * This "batch->memcg" is valid without any css_get/put etc... |
| 2683 | * bacause we hide charges behind us. |
| 2684 | */ |
| 2685 | if (batch->bytes) |
| 2686 | res_counter_uncharge(&batch->memcg->res, batch->bytes); |
| 2687 | if (batch->memsw_bytes) |
| 2688 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2689 | memcg_oom_recover(batch->memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2690 | /* forget this pointer (for sanity check) */ |
| 2691 | batch->memcg = NULL; |
| 2692 | } |
| 2693 | |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2694 | #ifdef CONFIG_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2695 | /* |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2696 | * called after __delete_from_swap_cache() and drop "page" account. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2697 | * memcg information is recorded to swap_cgroup of "ent" |
| 2698 | */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2699 | void |
| 2700 | 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] | 2701 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2702 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2703 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2704 | |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2705 | if (!swapout) /* this was a swap cache but the swap is unused ! */ |
| 2706 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; |
| 2707 | |
| 2708 | memcg = __mem_cgroup_uncharge_common(page, ctype); |
| 2709 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2710 | /* |
| 2711 | * record memcg information, if swapout && memcg != NULL, |
| 2712 | * mem_cgroup_get() was called in uncharge(). |
| 2713 | */ |
| 2714 | if (do_swap_account && swapout && memcg) |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2715 | swap_cgroup_record(ent, css_id(&memcg->css)); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2716 | } |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2717 | #endif |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2718 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2719 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 2720 | /* |
| 2721 | * called from swap_entry_free(). remove record in swap_cgroup and |
| 2722 | * uncharge "memsw" account. |
| 2723 | */ |
| 2724 | void mem_cgroup_uncharge_swap(swp_entry_t ent) |
| 2725 | { |
| 2726 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2727 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2728 | |
| 2729 | if (!do_swap_account) |
| 2730 | return; |
| 2731 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2732 | id = swap_cgroup_record(ent, 0); |
| 2733 | rcu_read_lock(); |
| 2734 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2735 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2736 | /* |
| 2737 | * We uncharge this because swap is freed. |
| 2738 | * This memcg can be obsolete one. We avoid calling css_tryget |
| 2739 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2740 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 2741 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2742 | mem_cgroup_swap_statistics(memcg, false); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2743 | mem_cgroup_put(memcg); |
| 2744 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2745 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2746 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2747 | |
| 2748 | /** |
| 2749 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2750 | * @entry: swap entry to be moved |
| 2751 | * @from: mem_cgroup which the entry is moved from |
| 2752 | * @to: mem_cgroup which the entry is moved to |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2753 | * @need_fixup: whether we should fixup res_counters and refcounts. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2754 | * |
| 2755 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2756 | * as the mem_cgroup's id of @from. |
| 2757 | * |
| 2758 | * Returns 0 on success, -EINVAL on failure. |
| 2759 | * |
| 2760 | * The caller must have charged to @to, IOW, called res_counter_charge() about |
| 2761 | * both res and memsw, and called css_get(). |
| 2762 | */ |
| 2763 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2764 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2765 | { |
| 2766 | unsigned short old_id, new_id; |
| 2767 | |
| 2768 | old_id = css_id(&from->css); |
| 2769 | new_id = css_id(&to->css); |
| 2770 | |
| 2771 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2772 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2773 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2774 | /* |
| 2775 | * This function is only called from task migration context now. |
| 2776 | * It postpones res_counter and refcount handling till the end |
| 2777 | * of task migration(mem_cgroup_clear_mc()) for performance |
| 2778 | * improvement. But we cannot postpone mem_cgroup_get(to) |
| 2779 | * because if the process that has been moved to @to does |
| 2780 | * swap-in, the refcount of @to might be decreased to 0. |
| 2781 | */ |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2782 | mem_cgroup_get(to); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2783 | if (need_fixup) { |
| 2784 | if (!mem_cgroup_is_root(from)) |
| 2785 | res_counter_uncharge(&from->memsw, PAGE_SIZE); |
| 2786 | mem_cgroup_put(from); |
| 2787 | /* |
| 2788 | * we charged both to->res and to->memsw, so we should |
| 2789 | * uncharge to->res. |
| 2790 | */ |
| 2791 | if (!mem_cgroup_is_root(to)) |
| 2792 | res_counter_uncharge(&to->res, PAGE_SIZE); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2793 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2794 | return 0; |
| 2795 | } |
| 2796 | return -EINVAL; |
| 2797 | } |
| 2798 | #else |
| 2799 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2800 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2801 | { |
| 2802 | return -EINVAL; |
| 2803 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2804 | #endif |
| 2805 | |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2806 | /* |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2807 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
| 2808 | * page belongs to. |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2809 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2810 | int mem_cgroup_prepare_migration(struct page *page, |
| 2811 | struct page *newpage, struct mem_cgroup **ptr) |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2812 | { |
| 2813 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2814 | struct mem_cgroup *mem = NULL; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2815 | enum charge_type ctype; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2816 | int ret = 0; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 2817 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2818 | VM_BUG_ON(PageTransHuge(page)); |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2819 | if (mem_cgroup_disabled()) |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 2820 | return 0; |
| 2821 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2822 | pc = lookup_page_cgroup(page); |
| 2823 | lock_page_cgroup(pc); |
| 2824 | if (PageCgroupUsed(pc)) { |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2825 | mem = pc->mem_cgroup; |
| 2826 | css_get(&mem->css); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2827 | /* |
| 2828 | * At migrating an anonymous page, its mapcount goes down |
| 2829 | * to 0 and uncharge() will be called. But, even if it's fully |
| 2830 | * unmapped, migration may fail and this page has to be |
| 2831 | * charged again. We set MIGRATION flag here and delay uncharge |
| 2832 | * until end_migration() is called |
| 2833 | * |
| 2834 | * Corner Case Thinking |
| 2835 | * A) |
| 2836 | * When the old page was mapped as Anon and it's unmap-and-freed |
| 2837 | * while migration was ongoing. |
| 2838 | * If unmap finds the old page, uncharge() of it will be delayed |
| 2839 | * until end_migration(). If unmap finds a new page, it's |
| 2840 | * uncharged when it make mapcount to be 1->0. If unmap code |
| 2841 | * finds swap_migration_entry, the new page will not be mapped |
| 2842 | * and end_migration() will find it(mapcount==0). |
| 2843 | * |
| 2844 | * B) |
| 2845 | * When the old page was mapped but migraion fails, the kernel |
| 2846 | * remaps it. A charge for it is kept by MIGRATION flag even |
| 2847 | * if mapcount goes down to 0. We can do remap successfully |
| 2848 | * without charging it again. |
| 2849 | * |
| 2850 | * C) |
| 2851 | * The "old" page is under lock_page() until the end of |
| 2852 | * migration, so, the old page itself will not be swapped-out. |
| 2853 | * If the new page is swapped out before end_migraton, our |
| 2854 | * hook to usual swap-out path will catch the event. |
| 2855 | */ |
| 2856 | if (PageAnon(page)) |
| 2857 | SetPageCgroupMigration(pc); |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 2858 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2859 | unlock_page_cgroup(pc); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2860 | /* |
| 2861 | * If the page is not charged at this point, |
| 2862 | * we return here. |
| 2863 | */ |
| 2864 | if (!mem) |
| 2865 | return 0; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2866 | |
Andrea Arcangeli | 93d5c9b | 2010-04-23 13:17:39 -0400 | [diff] [blame] | 2867 | *ptr = mem; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2868 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false, PAGE_SIZE); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2869 | css_put(&mem->css);/* drop extra refcnt */ |
| 2870 | if (ret || *ptr == NULL) { |
| 2871 | if (PageAnon(page)) { |
| 2872 | lock_page_cgroup(pc); |
| 2873 | ClearPageCgroupMigration(pc); |
| 2874 | unlock_page_cgroup(pc); |
| 2875 | /* |
| 2876 | * The old page may be fully unmapped while we kept it. |
| 2877 | */ |
| 2878 | mem_cgroup_uncharge_page(page); |
| 2879 | } |
| 2880 | return -ENOMEM; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2881 | } |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2882 | /* |
| 2883 | * We charge new page before it's used/mapped. So, even if unlock_page() |
| 2884 | * is called before end_migration, we can catch all events on this new |
| 2885 | * page. In the case new page is migrated but not remapped, new page's |
| 2886 | * mapcount will be finally 0 and we call uncharge in end_migration(). |
| 2887 | */ |
| 2888 | pc = lookup_page_cgroup(newpage); |
| 2889 | if (PageAnon(page)) |
| 2890 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; |
| 2891 | else if (page_is_file_cache(page)) |
| 2892 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
| 2893 | else |
| 2894 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2895 | __mem_cgroup_commit_charge(mem, pc, ctype, PAGE_SIZE); |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2896 | return ret; |
| 2897 | } |
Hugh Dickins | fb59e9f | 2008-03-04 14:29:16 -0800 | [diff] [blame] | 2898 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2899 | /* remove redundant charge if migration failed*/ |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2900 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2901 | struct page *oldpage, struct page *newpage) |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2902 | { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2903 | struct page *used, *unused; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2904 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2905 | |
| 2906 | if (!mem) |
| 2907 | return; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2908 | /* blocks rmdir() */ |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2909 | cgroup_exclude_rmdir(&mem->css); |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2910 | /* at migration success, oldpage->mapping is NULL. */ |
| 2911 | if (oldpage->mapping) { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2912 | used = oldpage; |
| 2913 | unused = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2914 | } else { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2915 | used = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2916 | unused = oldpage; |
| 2917 | } |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2918 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2919 | * We disallowed uncharge of pages under migration because mapcount |
| 2920 | * of the page goes down to zero, temporarly. |
| 2921 | * Clear the flag and check the page should be charged. |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2922 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2923 | pc = lookup_page_cgroup(oldpage); |
| 2924 | lock_page_cgroup(pc); |
| 2925 | ClearPageCgroupMigration(pc); |
| 2926 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2927 | |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2928 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
| 2929 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2930 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2931 | * If a page is a file cache, radix-tree replacement is very atomic |
| 2932 | * and we can skip this check. When it was an Anon page, its mapcount |
| 2933 | * goes down to 0. But because we added MIGRATION flage, it's not |
| 2934 | * uncharged yet. There are several case but page->mapcount check |
| 2935 | * and USED bit check in mem_cgroup_uncharge_page() will do enough |
| 2936 | * check. (see prepare_charge() also) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2937 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2938 | if (PageAnon(used)) |
| 2939 | mem_cgroup_uncharge_page(used); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2940 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2941 | * At migration, we may charge account against cgroup which has no |
| 2942 | * tasks. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2943 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 2944 | * In that case, we need to call pre_destroy() again. check it here. |
| 2945 | */ |
| 2946 | cgroup_release_and_wakeup_rmdir(&mem->css); |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2947 | } |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 2948 | |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 2949 | /* |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 2950 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
| 2951 | * Calling hierarchical_reclaim is not enough because we should update |
| 2952 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. |
| 2953 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, |
| 2954 | * not from the memcg which this page would be charged to. |
| 2955 | * try_charge_swapin does all of these works properly. |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2956 | */ |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 2957 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2958 | struct mm_struct *mm, |
| 2959 | gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2960 | { |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2961 | struct mem_cgroup *mem = NULL; |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 2962 | int ret; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2963 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2964 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2965 | return 0; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2966 | |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 2967 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
| 2968 | if (!ret) |
| 2969 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2970 | |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 2971 | return ret; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 2972 | } |
| 2973 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2974 | static DEFINE_MUTEX(set_limit_mutex); |
| 2975 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2976 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2977 | unsigned long long val) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2978 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2979 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2980 | u64 memswlimit, memlimit; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2981 | int ret = 0; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2982 | int children = mem_cgroup_count_children(memcg); |
| 2983 | u64 curusage, oldusage; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2984 | int enlarge; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2985 | |
| 2986 | /* |
| 2987 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2988 | * is depends on callers. We set our retry-count to be function |
| 2989 | * of # of children which we should visit in this loop. |
| 2990 | */ |
| 2991 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; |
| 2992 | |
| 2993 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2994 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2995 | enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2996 | while (retry_count) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2997 | if (signal_pending(current)) { |
| 2998 | ret = -EINTR; |
| 2999 | break; |
| 3000 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3001 | /* |
| 3002 | * Rather than hide all in some function, I do this in |
| 3003 | * open coded manner. You see what this really does. |
| 3004 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
| 3005 | */ |
| 3006 | mutex_lock(&set_limit_mutex); |
| 3007 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3008 | if (memswlimit < val) { |
| 3009 | ret = -EINVAL; |
| 3010 | mutex_unlock(&set_limit_mutex); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3011 | break; |
| 3012 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3013 | |
| 3014 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3015 | if (memlimit < val) |
| 3016 | enlarge = 1; |
| 3017 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3018 | ret = res_counter_set_limit(&memcg->res, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3019 | if (!ret) { |
| 3020 | if (memswlimit == val) |
| 3021 | memcg->memsw_is_minimum = true; |
| 3022 | else |
| 3023 | memcg->memsw_is_minimum = false; |
| 3024 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3025 | mutex_unlock(&set_limit_mutex); |
| 3026 | |
| 3027 | if (!ret) |
| 3028 | break; |
| 3029 | |
Bob Liu | aa20d48 | 2009-12-15 16:47:14 -0800 | [diff] [blame] | 3030 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3031 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3032 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
| 3033 | /* Usage is reduced ? */ |
| 3034 | if (curusage >= oldusage) |
| 3035 | retry_count--; |
| 3036 | else |
| 3037 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3038 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3039 | if (!ret && enlarge) |
| 3040 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 3041 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3042 | return ret; |
| 3043 | } |
| 3044 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 3045 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
| 3046 | unsigned long long val) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3047 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3048 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3049 | u64 memlimit, memswlimit, oldusage, curusage; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3050 | int children = mem_cgroup_count_children(memcg); |
| 3051 | int ret = -EBUSY; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3052 | int enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3053 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3054 | /* see mem_cgroup_resize_res_limit */ |
| 3055 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
| 3056 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3057 | while (retry_count) { |
| 3058 | if (signal_pending(current)) { |
| 3059 | ret = -EINTR; |
| 3060 | break; |
| 3061 | } |
| 3062 | /* |
| 3063 | * Rather than hide all in some function, I do this in |
| 3064 | * open coded manner. You see what this really does. |
| 3065 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
| 3066 | */ |
| 3067 | mutex_lock(&set_limit_mutex); |
| 3068 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3069 | if (memlimit > val) { |
| 3070 | ret = -EINVAL; |
| 3071 | mutex_unlock(&set_limit_mutex); |
| 3072 | break; |
| 3073 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3074 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3075 | if (memswlimit < val) |
| 3076 | enlarge = 1; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3077 | ret = res_counter_set_limit(&memcg->memsw, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3078 | if (!ret) { |
| 3079 | if (memlimit == val) |
| 3080 | memcg->memsw_is_minimum = true; |
| 3081 | else |
| 3082 | memcg->memsw_is_minimum = false; |
| 3083 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3084 | mutex_unlock(&set_limit_mutex); |
| 3085 | |
| 3086 | if (!ret) |
| 3087 | break; |
| 3088 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3089 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 3090 | MEM_CGROUP_RECLAIM_NOSWAP | |
| 3091 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3092 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3093 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3094 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3095 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3096 | else |
| 3097 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3098 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3099 | if (!ret && enlarge) |
| 3100 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3101 | return ret; |
| 3102 | } |
| 3103 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3104 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 3105 | gfp_t gfp_mask) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3106 | { |
| 3107 | unsigned long nr_reclaimed = 0; |
| 3108 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 3109 | unsigned long reclaimed; |
| 3110 | int loop = 0; |
| 3111 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3112 | unsigned long long excess; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3113 | |
| 3114 | if (order > 0) |
| 3115 | return 0; |
| 3116 | |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 3117 | 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] | 3118 | /* |
| 3119 | * This loop can run a while, specially if mem_cgroup's continuously |
| 3120 | * keep exceeding their soft limit and putting the system under |
| 3121 | * pressure |
| 3122 | */ |
| 3123 | do { |
| 3124 | if (next_mz) |
| 3125 | mz = next_mz; |
| 3126 | else |
| 3127 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 3128 | if (!mz) |
| 3129 | break; |
| 3130 | |
| 3131 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
| 3132 | gfp_mask, |
| 3133 | MEM_CGROUP_RECLAIM_SOFT); |
| 3134 | nr_reclaimed += reclaimed; |
| 3135 | spin_lock(&mctz->lock); |
| 3136 | |
| 3137 | /* |
| 3138 | * If we failed to reclaim anything from this memory cgroup |
| 3139 | * it is time to move on to the next cgroup |
| 3140 | */ |
| 3141 | next_mz = NULL; |
| 3142 | if (!reclaimed) { |
| 3143 | do { |
| 3144 | /* |
| 3145 | * Loop until we find yet another one. |
| 3146 | * |
| 3147 | * By the time we get the soft_limit lock |
| 3148 | * again, someone might have aded the |
| 3149 | * group back on the RB tree. Iterate to |
| 3150 | * make sure we get a different mem. |
| 3151 | * mem_cgroup_largest_soft_limit_node returns |
| 3152 | * NULL if no other cgroup is present on |
| 3153 | * the tree |
| 3154 | */ |
| 3155 | next_mz = |
| 3156 | __mem_cgroup_largest_soft_limit_node(mctz); |
| 3157 | if (next_mz == mz) { |
| 3158 | css_put(&next_mz->mem->css); |
| 3159 | next_mz = NULL; |
| 3160 | } else /* next_mz == NULL or other memcg */ |
| 3161 | break; |
| 3162 | } while (1); |
| 3163 | } |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3164 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3165 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3166 | /* |
| 3167 | * One school of thought says that we should not add |
| 3168 | * back the node to the tree if reclaim returns 0. |
| 3169 | * But our reclaim could return 0, simply because due |
| 3170 | * to priority we are exposing a smaller subset of |
| 3171 | * memory to reclaim from. Consider this as a longer |
| 3172 | * term TODO. |
| 3173 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3174 | /* If excess == 0, no tree ops */ |
| 3175 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3176 | spin_unlock(&mctz->lock); |
| 3177 | css_put(&mz->mem->css); |
| 3178 | loop++; |
| 3179 | /* |
| 3180 | * Could not reclaim anything and there are no more |
| 3181 | * mem cgroups to try or we seem to be looping without |
| 3182 | * reclaiming anything. |
| 3183 | */ |
| 3184 | if (!nr_reclaimed && |
| 3185 | (next_mz == NULL || |
| 3186 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3187 | break; |
| 3188 | } while (!nr_reclaimed); |
| 3189 | if (next_mz) |
| 3190 | css_put(&next_mz->mem->css); |
| 3191 | return nr_reclaimed; |
| 3192 | } |
| 3193 | |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3194 | /* |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3195 | * This routine traverse page_cgroup in given list and drop them all. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3196 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
| 3197 | */ |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3198 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3199 | int node, int zid, enum lru_list lru) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3200 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3201 | struct zone *zone; |
| 3202 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3203 | struct page_cgroup *pc, *busy; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3204 | unsigned long flags, loop; |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3205 | struct list_head *list; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3206 | int ret = 0; |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3207 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3208 | zone = &NODE_DATA(node)->node_zones[zid]; |
| 3209 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 3210 | list = &mz->lists[lru]; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3211 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3212 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
| 3213 | /* give some margin against EBUSY etc...*/ |
| 3214 | loop += 256; |
| 3215 | busy = NULL; |
| 3216 | while (loop--) { |
| 3217 | ret = 0; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3218 | spin_lock_irqsave(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3219 | if (list_empty(list)) { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3220 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3221 | break; |
| 3222 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3223 | pc = list_entry(list->prev, struct page_cgroup, lru); |
| 3224 | if (busy == pc) { |
| 3225 | list_move(&pc->lru, list); |
Thiago Farina | 648bcc7 | 2010-03-05 13:42:04 -0800 | [diff] [blame] | 3226 | busy = NULL; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3227 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3228 | continue; |
| 3229 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3230 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3231 | |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 3232 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3233 | if (ret == -ENOMEM) |
| 3234 | break; |
| 3235 | |
| 3236 | if (ret == -EBUSY || ret == -EINVAL) { |
| 3237 | /* found lock contention or "pc" is obsolete. */ |
| 3238 | busy = pc; |
| 3239 | cond_resched(); |
| 3240 | } else |
| 3241 | busy = NULL; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3242 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3243 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3244 | if (!ret && !list_empty(list)) |
| 3245 | return -EBUSY; |
| 3246 | return ret; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3247 | } |
| 3248 | |
| 3249 | /* |
| 3250 | * make mem_cgroup's charge to be 0 if there is no task. |
| 3251 | * This enables deleting this mem_cgroup. |
| 3252 | */ |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3253 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3254 | { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3255 | int ret; |
| 3256 | int node, zid, shrink; |
| 3257 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3258 | struct cgroup *cgrp = mem->css.cgroup; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 3259 | |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3260 | css_get(&mem->css); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3261 | |
| 3262 | shrink = 0; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3263 | /* should free all ? */ |
| 3264 | if (free_all) |
| 3265 | goto try_to_free; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3266 | move_account: |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3267 | do { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3268 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3269 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3270 | goto out; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3271 | ret = -EINTR; |
| 3272 | if (signal_pending(current)) |
| 3273 | goto out; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3274 | /* This is for making all *used* pages to be on LRU. */ |
| 3275 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 3276 | drain_all_stock_sync(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3277 | ret = 0; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 3278 | mem_cgroup_start_move(mem); |
KAMEZAWA Hiroyuki | 299b4ea | 2009-01-29 14:25:17 -0800 | [diff] [blame] | 3279 | for_each_node_state(node, N_HIGH_MEMORY) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3280 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 3281 | enum lru_list l; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3282 | for_each_lru(l) { |
| 3283 | ret = mem_cgroup_force_empty_list(mem, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3284 | node, zid, l); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3285 | if (ret) |
| 3286 | break; |
| 3287 | } |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 3288 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3289 | if (ret) |
| 3290 | break; |
| 3291 | } |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 3292 | mem_cgroup_end_move(mem); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3293 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3294 | /* it seems parent cgroup doesn't have enough mem */ |
| 3295 | if (ret == -ENOMEM) |
| 3296 | goto try_to_free; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3297 | cond_resched(); |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3298 | /* "ret" should also be checked to ensure all lists are empty. */ |
| 3299 | } while (mem->res.usage > 0 || ret); |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3300 | out: |
| 3301 | css_put(&mem->css); |
| 3302 | return ret; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3303 | |
| 3304 | try_to_free: |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3305 | /* returns EBUSY if there is a task or if we come here twice. */ |
| 3306 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3307 | ret = -EBUSY; |
| 3308 | goto out; |
| 3309 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3310 | /* we call try-to-free pages for make this cgroup empty */ |
| 3311 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3312 | /* try to free all pages in this cgroup */ |
| 3313 | shrink = 1; |
| 3314 | while (nr_retries && mem->res.usage > 0) { |
| 3315 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3316 | |
| 3317 | if (signal_pending(current)) { |
| 3318 | ret = -EINTR; |
| 3319 | goto out; |
| 3320 | } |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3321 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
| 3322 | false, get_swappiness(mem)); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3323 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3324 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3325 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3326 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3327 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3328 | |
| 3329 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3330 | lru_add_drain(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3331 | /* try move_account...there may be some *locked* pages. */ |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3332 | goto move_account; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3333 | } |
| 3334 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3335 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
| 3336 | { |
| 3337 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); |
| 3338 | } |
| 3339 | |
| 3340 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3341 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
| 3342 | { |
| 3343 | return mem_cgroup_from_cont(cont)->use_hierarchy; |
| 3344 | } |
| 3345 | |
| 3346 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, |
| 3347 | u64 val) |
| 3348 | { |
| 3349 | int retval = 0; |
| 3350 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
| 3351 | struct cgroup *parent = cont->parent; |
| 3352 | struct mem_cgroup *parent_mem = NULL; |
| 3353 | |
| 3354 | if (parent) |
| 3355 | parent_mem = mem_cgroup_from_cont(parent); |
| 3356 | |
| 3357 | cgroup_lock(); |
| 3358 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 3359 | * 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] | 3360 | * in the child subtrees. If it is unset, then the change can |
| 3361 | * occur, provided the current cgroup has no children. |
| 3362 | * |
| 3363 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 3364 | * set if there are no children. |
| 3365 | */ |
| 3366 | if ((!parent_mem || !parent_mem->use_hierarchy) && |
| 3367 | (val == 1 || val == 0)) { |
| 3368 | if (list_empty(&cont->children)) |
| 3369 | mem->use_hierarchy = val; |
| 3370 | else |
| 3371 | retval = -EBUSY; |
| 3372 | } else |
| 3373 | retval = -EINVAL; |
| 3374 | cgroup_unlock(); |
| 3375 | |
| 3376 | return retval; |
| 3377 | } |
| 3378 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3379 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3380 | static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, |
| 3381 | enum mem_cgroup_stat_index idx) |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3382 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3383 | struct mem_cgroup *iter; |
| 3384 | s64 val = 0; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3385 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3386 | /* each per cpu's value can be minus.Then, use s64 */ |
| 3387 | for_each_mem_cgroup_tree(iter, mem) |
| 3388 | val += mem_cgroup_read_stat(iter, idx); |
| 3389 | |
| 3390 | if (val < 0) /* race ? */ |
| 3391 | val = 0; |
| 3392 | return val; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3393 | } |
| 3394 | |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3395 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
| 3396 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3397 | u64 val; |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3398 | |
| 3399 | if (!mem_cgroup_is_root(mem)) { |
| 3400 | if (!swap) |
| 3401 | return res_counter_read_u64(&mem->res, RES_USAGE); |
| 3402 | else |
| 3403 | return res_counter_read_u64(&mem->memsw, RES_USAGE); |
| 3404 | } |
| 3405 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3406 | val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE); |
| 3407 | val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3408 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3409 | if (swap) |
| 3410 | val += mem_cgroup_get_recursive_idx_stat(mem, |
| 3411 | MEM_CGROUP_STAT_SWAPOUT); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3412 | |
| 3413 | return val << PAGE_SHIFT; |
| 3414 | } |
| 3415 | |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 3416 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3417 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3418 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3419 | u64 val; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3420 | int type, name; |
| 3421 | |
| 3422 | type = MEMFILE_TYPE(cft->private); |
| 3423 | name = MEMFILE_ATTR(cft->private); |
| 3424 | switch (type) { |
| 3425 | case _MEM: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3426 | if (name == RES_USAGE) |
| 3427 | val = mem_cgroup_usage(mem, false); |
| 3428 | else |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3429 | val = res_counter_read_u64(&mem->res, name); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3430 | break; |
| 3431 | case _MEMSWAP: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3432 | if (name == RES_USAGE) |
| 3433 | val = mem_cgroup_usage(mem, true); |
| 3434 | else |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3435 | val = res_counter_read_u64(&mem->memsw, name); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3436 | break; |
| 3437 | default: |
| 3438 | BUG(); |
| 3439 | break; |
| 3440 | } |
| 3441 | return val; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3442 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3443 | /* |
| 3444 | * The user of this function is... |
| 3445 | * RES_LIMIT. |
| 3446 | */ |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 3447 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
| 3448 | const char *buffer) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3449 | { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3450 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3451 | int type, name; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3452 | unsigned long long val; |
| 3453 | int ret; |
| 3454 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3455 | type = MEMFILE_TYPE(cft->private); |
| 3456 | name = MEMFILE_ATTR(cft->private); |
| 3457 | switch (name) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3458 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3459 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3460 | ret = -EINVAL; |
| 3461 | break; |
| 3462 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3463 | /* This function does all necessary parse...reuse it */ |
| 3464 | ret = res_counter_memparse_write_strategy(buffer, &val); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3465 | if (ret) |
| 3466 | break; |
| 3467 | if (type == _MEM) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3468 | ret = mem_cgroup_resize_limit(memcg, val); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3469 | else |
| 3470 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3471 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3472 | case RES_SOFT_LIMIT: |
| 3473 | ret = res_counter_memparse_write_strategy(buffer, &val); |
| 3474 | if (ret) |
| 3475 | break; |
| 3476 | /* |
| 3477 | * For memsw, soft limits are hard to implement in terms |
| 3478 | * of semantics, for now, we support soft limits for |
| 3479 | * control without swap |
| 3480 | */ |
| 3481 | if (type == _MEM) |
| 3482 | ret = res_counter_set_soft_limit(&memcg->res, val); |
| 3483 | else |
| 3484 | ret = -EINVAL; |
| 3485 | break; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3486 | default: |
| 3487 | ret = -EINVAL; /* should be BUG() ? */ |
| 3488 | break; |
| 3489 | } |
| 3490 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3491 | } |
| 3492 | |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3493 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
| 3494 | unsigned long long *mem_limit, unsigned long long *memsw_limit) |
| 3495 | { |
| 3496 | struct cgroup *cgroup; |
| 3497 | unsigned long long min_limit, min_memsw_limit, tmp; |
| 3498 | |
| 3499 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3500 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3501 | cgroup = memcg->css.cgroup; |
| 3502 | if (!memcg->use_hierarchy) |
| 3503 | goto out; |
| 3504 | |
| 3505 | while (cgroup->parent) { |
| 3506 | cgroup = cgroup->parent; |
| 3507 | memcg = mem_cgroup_from_cont(cgroup); |
| 3508 | if (!memcg->use_hierarchy) |
| 3509 | break; |
| 3510 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3511 | min_limit = min(min_limit, tmp); |
| 3512 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3513 | min_memsw_limit = min(min_memsw_limit, tmp); |
| 3514 | } |
| 3515 | out: |
| 3516 | *mem_limit = min_limit; |
| 3517 | *memsw_limit = min_memsw_limit; |
| 3518 | return; |
| 3519 | } |
| 3520 | |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3521 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3522 | { |
| 3523 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3524 | int type, name; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3525 | |
| 3526 | mem = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3527 | type = MEMFILE_TYPE(event); |
| 3528 | name = MEMFILE_ATTR(event); |
| 3529 | switch (name) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3530 | case RES_MAX_USAGE: |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3531 | if (type == _MEM) |
| 3532 | res_counter_reset_max(&mem->res); |
| 3533 | else |
| 3534 | res_counter_reset_max(&mem->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3535 | break; |
| 3536 | case RES_FAILCNT: |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3537 | if (type == _MEM) |
| 3538 | res_counter_reset_failcnt(&mem->res); |
| 3539 | else |
| 3540 | res_counter_reset_failcnt(&mem->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3541 | break; |
| 3542 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3543 | |
Pavel Emelyanov | 85cc59d | 2008-04-29 01:00:20 -0700 | [diff] [blame] | 3544 | return 0; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3545 | } |
| 3546 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3547 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
| 3548 | struct cftype *cft) |
| 3549 | { |
| 3550 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; |
| 3551 | } |
| 3552 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3553 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3554 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 3555 | struct cftype *cft, u64 val) |
| 3556 | { |
| 3557 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 3558 | |
| 3559 | if (val >= (1 << NR_MOVE_TYPE)) |
| 3560 | return -EINVAL; |
| 3561 | /* |
| 3562 | * We check this value several times in both in can_attach() and |
| 3563 | * attach(), so we need cgroup lock to prevent this value from being |
| 3564 | * inconsistent. |
| 3565 | */ |
| 3566 | cgroup_lock(); |
| 3567 | mem->move_charge_at_immigrate = val; |
| 3568 | cgroup_unlock(); |
| 3569 | |
| 3570 | return 0; |
| 3571 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3572 | #else |
| 3573 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 3574 | struct cftype *cft, u64 val) |
| 3575 | { |
| 3576 | return -ENOSYS; |
| 3577 | } |
| 3578 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3579 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3580 | |
| 3581 | /* For read statistics */ |
| 3582 | enum { |
| 3583 | MCS_CACHE, |
| 3584 | MCS_RSS, |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 3585 | MCS_FILE_MAPPED, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3586 | MCS_PGPGIN, |
| 3587 | MCS_PGPGOUT, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3588 | MCS_SWAP, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3589 | MCS_INACTIVE_ANON, |
| 3590 | MCS_ACTIVE_ANON, |
| 3591 | MCS_INACTIVE_FILE, |
| 3592 | MCS_ACTIVE_FILE, |
| 3593 | MCS_UNEVICTABLE, |
| 3594 | NR_MCS_STAT, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3595 | }; |
| 3596 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3597 | struct mcs_total_stat { |
| 3598 | s64 stat[NR_MCS_STAT]; |
| 3599 | }; |
| 3600 | |
| 3601 | struct { |
| 3602 | char *local_name; |
| 3603 | char *total_name; |
| 3604 | } memcg_stat_strings[NR_MCS_STAT] = { |
| 3605 | {"cache", "total_cache"}, |
| 3606 | {"rss", "total_rss"}, |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 3607 | {"mapped_file", "total_mapped_file"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3608 | {"pgpgin", "total_pgpgin"}, |
| 3609 | {"pgpgout", "total_pgpgout"}, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3610 | {"swap", "total_swap"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3611 | {"inactive_anon", "total_inactive_anon"}, |
| 3612 | {"active_anon", "total_active_anon"}, |
| 3613 | {"inactive_file", "total_inactive_file"}, |
| 3614 | {"active_file", "total_active_file"}, |
| 3615 | {"unevictable", "total_unevictable"} |
| 3616 | }; |
| 3617 | |
| 3618 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3619 | static void |
| 3620 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3621 | { |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3622 | s64 val; |
| 3623 | |
| 3624 | /* per cpu stat */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3625 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3626 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3627 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3628 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3629 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 3630 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3631 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3632 | s->stat[MCS_PGPGIN] += val; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3633 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3634 | s->stat[MCS_PGPGOUT] += val; |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3635 | if (do_swap_account) { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3636 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3637 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
| 3638 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3639 | |
| 3640 | /* per zone stat */ |
| 3641 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); |
| 3642 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
| 3643 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); |
| 3644 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
| 3645 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); |
| 3646 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
| 3647 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); |
| 3648 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
| 3649 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); |
| 3650 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3651 | } |
| 3652 | |
| 3653 | static void |
| 3654 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) |
| 3655 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3656 | struct mem_cgroup *iter; |
| 3657 | |
| 3658 | for_each_mem_cgroup_tree(iter, mem) |
| 3659 | mem_cgroup_get_local_stat(iter, s); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3660 | } |
| 3661 | |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 3662 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
| 3663 | struct cgroup_map_cb *cb) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3664 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3665 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3666 | struct mcs_total_stat mystat; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3667 | int i; |
| 3668 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3669 | memset(&mystat, 0, sizeof(mystat)); |
| 3670 | mem_cgroup_get_local_stat(mem_cont, &mystat); |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3671 | |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3672 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 3673 | if (i == MCS_SWAP && !do_swap_account) |
| 3674 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3675 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3676 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3677 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3678 | /* Hierarchical information */ |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3679 | { |
| 3680 | unsigned long long limit, memsw_limit; |
| 3681 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); |
| 3682 | cb->fill(cb, "hierarchical_memory_limit", limit); |
| 3683 | if (do_swap_account) |
| 3684 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); |
| 3685 | } |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3686 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3687 | memset(&mystat, 0, sizeof(mystat)); |
| 3688 | mem_cgroup_get_total_stat(mem_cont, &mystat); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3689 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 3690 | if (i == MCS_SWAP && !do_swap_account) |
| 3691 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3692 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3693 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3694 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3695 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 3696 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3697 | |
| 3698 | { |
| 3699 | int nid, zid; |
| 3700 | struct mem_cgroup_per_zone *mz; |
| 3701 | unsigned long recent_rotated[2] = {0, 0}; |
| 3702 | unsigned long recent_scanned[2] = {0, 0}; |
| 3703 | |
| 3704 | for_each_online_node(nid) |
| 3705 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 3706 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
| 3707 | |
| 3708 | recent_rotated[0] += |
| 3709 | mz->reclaim_stat.recent_rotated[0]; |
| 3710 | recent_rotated[1] += |
| 3711 | mz->reclaim_stat.recent_rotated[1]; |
| 3712 | recent_scanned[0] += |
| 3713 | mz->reclaim_stat.recent_scanned[0]; |
| 3714 | recent_scanned[1] += |
| 3715 | mz->reclaim_stat.recent_scanned[1]; |
| 3716 | } |
| 3717 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); |
| 3718 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); |
| 3719 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); |
| 3720 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); |
| 3721 | } |
| 3722 | #endif |
| 3723 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3724 | return 0; |
| 3725 | } |
| 3726 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3727 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
| 3728 | { |
| 3729 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 3730 | |
| 3731 | return get_swappiness(memcg); |
| 3732 | } |
| 3733 | |
| 3734 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, |
| 3735 | u64 val) |
| 3736 | { |
| 3737 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 3738 | struct mem_cgroup *parent; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3739 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3740 | if (val > 100) |
| 3741 | return -EINVAL; |
| 3742 | |
| 3743 | if (cgrp->parent == NULL) |
| 3744 | return -EINVAL; |
| 3745 | |
| 3746 | parent = mem_cgroup_from_cont(cgrp->parent); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3747 | |
| 3748 | cgroup_lock(); |
| 3749 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3750 | /* If under hierarchy, only empty-root can set this value */ |
| 3751 | if ((parent->use_hierarchy) || |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3752 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
| 3753 | cgroup_unlock(); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3754 | return -EINVAL; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3755 | } |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3756 | |
| 3757 | spin_lock(&memcg->reclaim_param_lock); |
| 3758 | memcg->swappiness = val; |
| 3759 | spin_unlock(&memcg->reclaim_param_lock); |
| 3760 | |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3761 | cgroup_unlock(); |
| 3762 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3763 | return 0; |
| 3764 | } |
| 3765 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3766 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3767 | { |
| 3768 | struct mem_cgroup_threshold_ary *t; |
| 3769 | u64 usage; |
| 3770 | int i; |
| 3771 | |
| 3772 | rcu_read_lock(); |
| 3773 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3774 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3775 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3776 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3777 | |
| 3778 | if (!t) |
| 3779 | goto unlock; |
| 3780 | |
| 3781 | usage = mem_cgroup_usage(memcg, swap); |
| 3782 | |
| 3783 | /* |
| 3784 | * current_threshold points to threshold just below usage. |
| 3785 | * If it's not true, a threshold was crossed after last |
| 3786 | * call of __mem_cgroup_threshold(). |
| 3787 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3788 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3789 | |
| 3790 | /* |
| 3791 | * Iterate backward over array of thresholds starting from |
| 3792 | * current_threshold and check if a threshold is crossed. |
| 3793 | * If none of thresholds below usage is crossed, we read |
| 3794 | * only one element of the array here. |
| 3795 | */ |
| 3796 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3797 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3798 | |
| 3799 | /* i = current_threshold + 1 */ |
| 3800 | i++; |
| 3801 | |
| 3802 | /* |
| 3803 | * Iterate forward over array of thresholds starting from |
| 3804 | * current_threshold+1 and check if a threshold is crossed. |
| 3805 | * If none of thresholds above usage is crossed, we read |
| 3806 | * only one element of the array here. |
| 3807 | */ |
| 3808 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3809 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3810 | |
| 3811 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3812 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3813 | unlock: |
| 3814 | rcu_read_unlock(); |
| 3815 | } |
| 3816 | |
| 3817 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3818 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3819 | while (memcg) { |
| 3820 | __mem_cgroup_threshold(memcg, false); |
| 3821 | if (do_swap_account) |
| 3822 | __mem_cgroup_threshold(memcg, true); |
| 3823 | |
| 3824 | memcg = parent_mem_cgroup(memcg); |
| 3825 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3826 | } |
| 3827 | |
| 3828 | static int compare_thresholds(const void *a, const void *b) |
| 3829 | { |
| 3830 | const struct mem_cgroup_threshold *_a = a; |
| 3831 | const struct mem_cgroup_threshold *_b = b; |
| 3832 | |
| 3833 | return _a->threshold - _b->threshold; |
| 3834 | } |
| 3835 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3836 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3837 | { |
| 3838 | struct mem_cgroup_eventfd_list *ev; |
| 3839 | |
| 3840 | list_for_each_entry(ev, &mem->oom_notify, list) |
| 3841 | eventfd_signal(ev->eventfd, 1); |
| 3842 | return 0; |
| 3843 | } |
| 3844 | |
| 3845 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) |
| 3846 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3847 | struct mem_cgroup *iter; |
| 3848 | |
| 3849 | for_each_mem_cgroup_tree(iter, mem) |
| 3850 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3851 | } |
| 3852 | |
| 3853 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, |
| 3854 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3855 | { |
| 3856 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3857 | struct mem_cgroup_thresholds *thresholds; |
| 3858 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3859 | int type = MEMFILE_TYPE(cft->private); |
| 3860 | u64 threshold, usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3861 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3862 | |
| 3863 | ret = res_counter_memparse_write_strategy(args, &threshold); |
| 3864 | if (ret) |
| 3865 | return ret; |
| 3866 | |
| 3867 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3868 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3869 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3870 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3871 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3872 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3873 | else |
| 3874 | BUG(); |
| 3875 | |
| 3876 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 3877 | |
| 3878 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3879 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3880 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3881 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3882 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3883 | |
| 3884 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3885 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3886 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3887 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3888 | ret = -ENOMEM; |
| 3889 | goto unlock; |
| 3890 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3891 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3892 | |
| 3893 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3894 | if (thresholds->primary) { |
| 3895 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3896 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3897 | } |
| 3898 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3899 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3900 | new->entries[size - 1].eventfd = eventfd; |
| 3901 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3902 | |
| 3903 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3904 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3905 | compare_thresholds, NULL); |
| 3906 | |
| 3907 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3908 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3909 | for (i = 0; i < size; i++) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3910 | if (new->entries[i].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3911 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3912 | * new->current_threshold will not be used until |
| 3913 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3914 | * it here. |
| 3915 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3916 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3917 | } |
| 3918 | } |
| 3919 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3920 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3921 | kfree(thresholds->spare); |
| 3922 | thresholds->spare = thresholds->primary; |
| 3923 | |
| 3924 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3925 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3926 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3927 | synchronize_rcu(); |
| 3928 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3929 | unlock: |
| 3930 | mutex_unlock(&memcg->thresholds_lock); |
| 3931 | |
| 3932 | return ret; |
| 3933 | } |
| 3934 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3935 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3936 | struct cftype *cft, struct eventfd_ctx *eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3937 | { |
| 3938 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3939 | struct mem_cgroup_thresholds *thresholds; |
| 3940 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3941 | int type = MEMFILE_TYPE(cft->private); |
| 3942 | u64 usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3943 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3944 | |
| 3945 | mutex_lock(&memcg->thresholds_lock); |
| 3946 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3947 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3948 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3949 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3950 | else |
| 3951 | BUG(); |
| 3952 | |
| 3953 | /* |
| 3954 | * Something went wrong if we trying to unregister a threshold |
| 3955 | * if we don't have thresholds |
| 3956 | */ |
| 3957 | BUG_ON(!thresholds); |
| 3958 | |
| 3959 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 3960 | |
| 3961 | /* Check if a threshold crossed before removing */ |
| 3962 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3963 | |
| 3964 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3965 | size = 0; |
| 3966 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3967 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3968 | size++; |
| 3969 | } |
| 3970 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3971 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3972 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3973 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3974 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3975 | kfree(new); |
| 3976 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3977 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3978 | } |
| 3979 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3980 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3981 | |
| 3982 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3983 | new->current_threshold = -1; |
| 3984 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3985 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3986 | continue; |
| 3987 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3988 | new->entries[j] = thresholds->primary->entries[i]; |
| 3989 | if (new->entries[j].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3990 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3991 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3992 | * until rcu_assign_pointer(), so it's safe to increment |
| 3993 | * it here. |
| 3994 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3995 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3996 | } |
| 3997 | j++; |
| 3998 | } |
| 3999 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4000 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4001 | /* Swap primary and spare array */ |
| 4002 | thresholds->spare = thresholds->primary; |
| 4003 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4004 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4005 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4006 | synchronize_rcu(); |
| 4007 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4008 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4009 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4010 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4011 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
| 4012 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
| 4013 | { |
| 4014 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 4015 | struct mem_cgroup_eventfd_list *event; |
| 4016 | int type = MEMFILE_TYPE(cft->private); |
| 4017 | |
| 4018 | BUG_ON(type != _OOM_TYPE); |
| 4019 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 4020 | if (!event) |
| 4021 | return -ENOMEM; |
| 4022 | |
| 4023 | mutex_lock(&memcg_oom_mutex); |
| 4024 | |
| 4025 | event->eventfd = eventfd; |
| 4026 | list_add(&event->list, &memcg->oom_notify); |
| 4027 | |
| 4028 | /* already in OOM ? */ |
| 4029 | if (atomic_read(&memcg->oom_lock)) |
| 4030 | eventfd_signal(eventfd, 1); |
| 4031 | mutex_unlock(&memcg_oom_mutex); |
| 4032 | |
| 4033 | return 0; |
| 4034 | } |
| 4035 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4036 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4037 | struct cftype *cft, struct eventfd_ctx *eventfd) |
| 4038 | { |
| 4039 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4040 | struct mem_cgroup_eventfd_list *ev, *tmp; |
| 4041 | int type = MEMFILE_TYPE(cft->private); |
| 4042 | |
| 4043 | BUG_ON(type != _OOM_TYPE); |
| 4044 | |
| 4045 | mutex_lock(&memcg_oom_mutex); |
| 4046 | |
| 4047 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { |
| 4048 | if (ev->eventfd == eventfd) { |
| 4049 | list_del(&ev->list); |
| 4050 | kfree(ev); |
| 4051 | } |
| 4052 | } |
| 4053 | |
| 4054 | mutex_unlock(&memcg_oom_mutex); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4055 | } |
| 4056 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4057 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
| 4058 | struct cftype *cft, struct cgroup_map_cb *cb) |
| 4059 | { |
| 4060 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4061 | |
| 4062 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); |
| 4063 | |
| 4064 | if (atomic_read(&mem->oom_lock)) |
| 4065 | cb->fill(cb, "under_oom", 1); |
| 4066 | else |
| 4067 | cb->fill(cb, "under_oom", 0); |
| 4068 | return 0; |
| 4069 | } |
| 4070 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4071 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
| 4072 | struct cftype *cft, u64 val) |
| 4073 | { |
| 4074 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4075 | struct mem_cgroup *parent; |
| 4076 | |
| 4077 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
| 4078 | if (!cgrp->parent || !((val == 0) || (val == 1))) |
| 4079 | return -EINVAL; |
| 4080 | |
| 4081 | parent = mem_cgroup_from_cont(cgrp->parent); |
| 4082 | |
| 4083 | cgroup_lock(); |
| 4084 | /* oom-kill-disable is a flag for subhierarchy. */ |
| 4085 | if ((parent->use_hierarchy) || |
| 4086 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { |
| 4087 | cgroup_unlock(); |
| 4088 | return -EINVAL; |
| 4089 | } |
| 4090 | mem->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 4091 | if (!val) |
| 4092 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4093 | cgroup_unlock(); |
| 4094 | return 0; |
| 4095 | } |
| 4096 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4097 | static struct cftype mem_cgroup_files[] = { |
| 4098 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4099 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4100 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4101 | .read_u64 = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4102 | .register_event = mem_cgroup_usage_register_event, |
| 4103 | .unregister_event = mem_cgroup_usage_unregister_event, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4104 | }, |
| 4105 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4106 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4107 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4108 | .trigger = mem_cgroup_reset, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4109 | .read_u64 = mem_cgroup_read, |
| 4110 | }, |
| 4111 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4112 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4113 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 4114 | .write_string = mem_cgroup_write, |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4115 | .read_u64 = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4116 | }, |
| 4117 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4118 | .name = "soft_limit_in_bytes", |
| 4119 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
| 4120 | .write_string = mem_cgroup_write, |
| 4121 | .read_u64 = mem_cgroup_read, |
| 4122 | }, |
| 4123 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4124 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4125 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4126 | .trigger = mem_cgroup_reset, |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4127 | .read_u64 = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4128 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4129 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4130 | .name = "stat", |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 4131 | .read_map = mem_control_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4132 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4133 | { |
| 4134 | .name = "force_empty", |
| 4135 | .trigger = mem_cgroup_force_empty_write, |
| 4136 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4137 | { |
| 4138 | .name = "use_hierarchy", |
| 4139 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4140 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4141 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4142 | { |
| 4143 | .name = "swappiness", |
| 4144 | .read_u64 = mem_cgroup_swappiness_read, |
| 4145 | .write_u64 = mem_cgroup_swappiness_write, |
| 4146 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4147 | { |
| 4148 | .name = "move_charge_at_immigrate", |
| 4149 | .read_u64 = mem_cgroup_move_charge_read, |
| 4150 | .write_u64 = mem_cgroup_move_charge_write, |
| 4151 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4152 | { |
| 4153 | .name = "oom_control", |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4154 | .read_map = mem_cgroup_oom_control_read, |
| 4155 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4156 | .register_event = mem_cgroup_oom_register_event, |
| 4157 | .unregister_event = mem_cgroup_oom_unregister_event, |
| 4158 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4159 | }, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4160 | }; |
| 4161 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4162 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 4163 | static struct cftype memsw_cgroup_files[] = { |
| 4164 | { |
| 4165 | .name = "memsw.usage_in_bytes", |
| 4166 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 4167 | .read_u64 = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4168 | .register_event = mem_cgroup_usage_register_event, |
| 4169 | .unregister_event = mem_cgroup_usage_unregister_event, |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4170 | }, |
| 4171 | { |
| 4172 | .name = "memsw.max_usage_in_bytes", |
| 4173 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 4174 | .trigger = mem_cgroup_reset, |
| 4175 | .read_u64 = mem_cgroup_read, |
| 4176 | }, |
| 4177 | { |
| 4178 | .name = "memsw.limit_in_bytes", |
| 4179 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 4180 | .write_string = mem_cgroup_write, |
| 4181 | .read_u64 = mem_cgroup_read, |
| 4182 | }, |
| 4183 | { |
| 4184 | .name = "memsw.failcnt", |
| 4185 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 4186 | .trigger = mem_cgroup_reset, |
| 4187 | .read_u64 = mem_cgroup_read, |
| 4188 | }, |
| 4189 | }; |
| 4190 | |
| 4191 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
| 4192 | { |
| 4193 | if (!do_swap_account) |
| 4194 | return 0; |
| 4195 | return cgroup_add_files(cont, ss, memsw_cgroup_files, |
| 4196 | ARRAY_SIZE(memsw_cgroup_files)); |
| 4197 | }; |
| 4198 | #else |
| 4199 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
| 4200 | { |
| 4201 | return 0; |
| 4202 | } |
| 4203 | #endif |
| 4204 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4205 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 4206 | { |
| 4207 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4208 | struct mem_cgroup_per_zone *mz; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4209 | enum lru_list l; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4210 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4211 | /* |
| 4212 | * This routine is called against possible nodes. |
| 4213 | * But it's BUG to call kmalloc() against offline node. |
| 4214 | * |
| 4215 | * TODO: this routine can waste much memory for nodes which will |
| 4216 | * never be onlined. It's better to use memory hotplug callback |
| 4217 | * function. |
| 4218 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4219 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4220 | tmp = -1; |
| 4221 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4222 | if (!pn) |
| 4223 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4224 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4225 | mem->info.nodeinfo[node] = pn; |
| 4226 | memset(pn, 0, sizeof(*pn)); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4227 | |
| 4228 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4229 | mz = &pn->zoneinfo[zone]; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4230 | for_each_lru(l) |
| 4231 | INIT_LIST_HEAD(&mz->lists[l]); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4232 | mz->usage_in_excess = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 4233 | mz->on_tree = false; |
| 4234 | mz->mem = mem; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4235 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4236 | return 0; |
| 4237 | } |
| 4238 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4239 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 4240 | { |
| 4241 | kfree(mem->info.nodeinfo[node]); |
| 4242 | } |
| 4243 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4244 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4245 | { |
| 4246 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4247 | int size = sizeof(struct mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4248 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4249 | /* Can be very big if MAX_NUMNODES is very big */ |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4250 | if (size < PAGE_SIZE) |
| 4251 | mem = kmalloc(size, GFP_KERNEL); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4252 | else |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4253 | mem = vmalloc(size); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4254 | |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4255 | if (!mem) |
| 4256 | return NULL; |
| 4257 | |
| 4258 | memset(mem, 0, size); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4259 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4260 | if (!mem->stat) |
| 4261 | goto out_free; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 4262 | spin_lock_init(&mem->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4263 | return mem; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4264 | |
| 4265 | out_free: |
| 4266 | if (size < PAGE_SIZE) |
| 4267 | kfree(mem); |
| 4268 | else |
| 4269 | vfree(mem); |
| 4270 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4271 | } |
| 4272 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4273 | /* |
| 4274 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4275 | * (scanning all at force_empty is too costly...) |
| 4276 | * |
| 4277 | * Instead of clearing all references at force_empty, we remember |
| 4278 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4279 | * it goes down to 0. |
| 4280 | * |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4281 | * Removal of cgroup itself succeeds regardless of refs from swap. |
| 4282 | */ |
| 4283 | |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4284 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4285 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4286 | int node; |
| 4287 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4288 | mem_cgroup_remove_from_trees(mem); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4289 | free_css_id(&mem_cgroup_subsys, &mem->css); |
| 4290 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4291 | for_each_node_state(node, N_POSSIBLE) |
| 4292 | free_mem_cgroup_per_zone_info(mem, node); |
| 4293 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4294 | free_percpu(mem->stat); |
| 4295 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4296 | kfree(mem); |
| 4297 | else |
| 4298 | vfree(mem); |
| 4299 | } |
| 4300 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4301 | static void mem_cgroup_get(struct mem_cgroup *mem) |
| 4302 | { |
| 4303 | atomic_inc(&mem->refcnt); |
| 4304 | } |
| 4305 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4306 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4307 | { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4308 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4309 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4310 | __mem_cgroup_free(mem); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4311 | if (parent) |
| 4312 | mem_cgroup_put(parent); |
| 4313 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4314 | } |
| 4315 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4316 | static void mem_cgroup_put(struct mem_cgroup *mem) |
| 4317 | { |
| 4318 | __mem_cgroup_put(mem, 1); |
| 4319 | } |
| 4320 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4321 | /* |
| 4322 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4323 | */ |
| 4324 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) |
| 4325 | { |
| 4326 | if (!mem->res.parent) |
| 4327 | return NULL; |
| 4328 | return mem_cgroup_from_res_counter(mem->res.parent, res); |
| 4329 | } |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4330 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4331 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 4332 | static void __init enable_swap_cgroup(void) |
| 4333 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 4334 | if (!mem_cgroup_disabled() && really_do_swap_account) |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4335 | do_swap_account = 1; |
| 4336 | } |
| 4337 | #else |
| 4338 | static void __init enable_swap_cgroup(void) |
| 4339 | { |
| 4340 | } |
| 4341 | #endif |
| 4342 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4343 | static int mem_cgroup_soft_limit_tree_init(void) |
| 4344 | { |
| 4345 | struct mem_cgroup_tree_per_node *rtpn; |
| 4346 | struct mem_cgroup_tree_per_zone *rtpz; |
| 4347 | int tmp, node, zone; |
| 4348 | |
| 4349 | for_each_node_state(node, N_POSSIBLE) { |
| 4350 | tmp = node; |
| 4351 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4352 | tmp = -1; |
| 4353 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); |
| 4354 | if (!rtpn) |
| 4355 | return 1; |
| 4356 | |
| 4357 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 4358 | |
| 4359 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4360 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 4361 | rtpz->rb_root = RB_ROOT; |
| 4362 | spin_lock_init(&rtpz->lock); |
| 4363 | } |
| 4364 | } |
| 4365 | return 0; |
| 4366 | } |
| 4367 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 4368 | static struct cgroup_subsys_state * __ref |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4369 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
| 4370 | { |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4371 | struct mem_cgroup *mem, *parent; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4372 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4373 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4374 | |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4375 | mem = mem_cgroup_alloc(); |
| 4376 | if (!mem) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4377 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4378 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4379 | for_each_node_state(node, N_POSSIBLE) |
| 4380 | if (alloc_mem_cgroup_per_zone_info(mem, node)) |
| 4381 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4382 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4383 | /* root ? */ |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4384 | if (cont->parent == NULL) { |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4385 | int cpu; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4386 | enable_swap_cgroup(); |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4387 | parent = NULL; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 4388 | root_mem_cgroup = mem; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4389 | if (mem_cgroup_soft_limit_tree_init()) |
| 4390 | goto free_out; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4391 | for_each_possible_cpu(cpu) { |
| 4392 | struct memcg_stock_pcp *stock = |
| 4393 | &per_cpu(memcg_stock, cpu); |
| 4394 | INIT_WORK(&stock->work, drain_local_stock); |
| 4395 | } |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 4396 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4397 | } else { |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4398 | parent = mem_cgroup_from_cont(cont->parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4399 | mem->use_hierarchy = parent->use_hierarchy; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4400 | mem->oom_kill_disable = parent->oom_kill_disable; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4401 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4402 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4403 | if (parent && parent->use_hierarchy) { |
| 4404 | res_counter_init(&mem->res, &parent->res); |
| 4405 | res_counter_init(&mem->memsw, &parent->memsw); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4406 | /* |
| 4407 | * We increment refcnt of the parent to ensure that we can |
| 4408 | * safely access it on res_counter_charge/uncharge. |
| 4409 | * This refcnt will be decremented when freeing this |
| 4410 | * mem_cgroup(see mem_cgroup_put). |
| 4411 | */ |
| 4412 | mem_cgroup_get(parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4413 | } else { |
| 4414 | res_counter_init(&mem->res, NULL); |
| 4415 | res_counter_init(&mem->memsw, NULL); |
| 4416 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4417 | mem->last_scanned_child = 0; |
KOSAKI Motohiro | 2733c06 | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 4418 | spin_lock_init(&mem->reclaim_param_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4419 | INIT_LIST_HEAD(&mem->oom_notify); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 4420 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4421 | if (parent) |
| 4422 | mem->swappiness = get_swappiness(parent); |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4423 | atomic_set(&mem->refcnt, 1); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4424 | mem->move_charge_at_immigrate = 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4425 | mutex_init(&mem->thresholds_lock); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4426 | return &mem->css; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4427 | free_out: |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4428 | __mem_cgroup_free(mem); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 4429 | root_mem_cgroup = NULL; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4430 | return ERR_PTR(error); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4431 | } |
| 4432 | |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4433 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4434 | struct cgroup *cont) |
| 4435 | { |
| 4436 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4437 | |
| 4438 | return mem_cgroup_force_empty(mem, false); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4439 | } |
| 4440 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4441 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
| 4442 | struct cgroup *cont) |
| 4443 | { |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4444 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4445 | |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4446 | mem_cgroup_put(mem); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4447 | } |
| 4448 | |
| 4449 | static int mem_cgroup_populate(struct cgroup_subsys *ss, |
| 4450 | struct cgroup *cont) |
| 4451 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4452 | int ret; |
| 4453 | |
| 4454 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, |
| 4455 | ARRAY_SIZE(mem_cgroup_files)); |
| 4456 | |
| 4457 | if (!ret) |
| 4458 | ret = register_memsw_files(cont, ss); |
| 4459 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4460 | } |
| 4461 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4462 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4463 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4464 | #define PRECHARGE_COUNT_AT_ONCE 256 |
| 4465 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4466 | { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4467 | int ret = 0; |
| 4468 | int batch_count = PRECHARGE_COUNT_AT_ONCE; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4469 | struct mem_cgroup *mem = mc.to; |
| 4470 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4471 | if (mem_cgroup_is_root(mem)) { |
| 4472 | mc.precharge += count; |
| 4473 | /* we don't need css_get for root */ |
| 4474 | return ret; |
| 4475 | } |
| 4476 | /* try to charge at once */ |
| 4477 | if (count > 1) { |
| 4478 | struct res_counter *dummy; |
| 4479 | /* |
| 4480 | * "mem" cannot be under rmdir() because we've already checked |
| 4481 | * by cgroup_lock_live_cgroup() that it is not removed and we |
| 4482 | * are still under the same cgroup_mutex. So we can postpone |
| 4483 | * css_get(). |
| 4484 | */ |
| 4485 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) |
| 4486 | goto one_by_one; |
| 4487 | if (do_swap_account && res_counter_charge(&mem->memsw, |
| 4488 | PAGE_SIZE * count, &dummy)) { |
| 4489 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
| 4490 | goto one_by_one; |
| 4491 | } |
| 4492 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4493 | return ret; |
| 4494 | } |
| 4495 | one_by_one: |
| 4496 | /* fall back to one by one charge */ |
| 4497 | while (count--) { |
| 4498 | if (signal_pending(current)) { |
| 4499 | ret = -EINTR; |
| 4500 | break; |
| 4501 | } |
| 4502 | if (!batch_count--) { |
| 4503 | batch_count = PRECHARGE_COUNT_AT_ONCE; |
| 4504 | cond_resched(); |
| 4505 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 4506 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
| 4507 | PAGE_SIZE); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4508 | if (ret || !mem) |
| 4509 | /* mem_cgroup_clear_mc() will do uncharge later */ |
| 4510 | return -ENOMEM; |
| 4511 | mc.precharge++; |
| 4512 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4513 | return ret; |
| 4514 | } |
| 4515 | |
| 4516 | /** |
| 4517 | * is_target_pte_for_mc - check a pte whether it is valid for move charge |
| 4518 | * @vma: the vma the pte to be checked belongs |
| 4519 | * @addr: the address corresponding to the pte to be checked |
| 4520 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4521 | * @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] | 4522 | * |
| 4523 | * Returns |
| 4524 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4525 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4526 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4527 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4528 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4529 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4530 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4531 | * |
| 4532 | * Called with pte lock held. |
| 4533 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4534 | union mc_target { |
| 4535 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4536 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4537 | }; |
| 4538 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4539 | enum mc_target_type { |
| 4540 | MC_TARGET_NONE, /* not used */ |
| 4541 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4542 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4543 | }; |
| 4544 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4545 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4546 | unsigned long addr, pte_t ptent) |
| 4547 | { |
| 4548 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4549 | |
| 4550 | if (!page || !page_mapped(page)) |
| 4551 | return NULL; |
| 4552 | if (PageAnon(page)) { |
| 4553 | /* we don't move shared anon */ |
| 4554 | if (!move_anon() || page_mapcount(page) > 2) |
| 4555 | return NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4556 | } else if (!move_file()) |
| 4557 | /* we ignore mapcount for file pages */ |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4558 | return NULL; |
| 4559 | if (!get_page_unless_zero(page)) |
| 4560 | return NULL; |
| 4561 | |
| 4562 | return page; |
| 4563 | } |
| 4564 | |
| 4565 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4566 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4567 | { |
| 4568 | int usage_count; |
| 4569 | struct page *page = NULL; |
| 4570 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4571 | |
| 4572 | if (!move_anon() || non_swap_entry(ent)) |
| 4573 | return NULL; |
| 4574 | usage_count = mem_cgroup_count_swap_user(ent, &page); |
| 4575 | if (usage_count > 1) { /* we don't move shared anon */ |
| 4576 | if (page) |
| 4577 | put_page(page); |
| 4578 | return NULL; |
| 4579 | } |
| 4580 | if (do_swap_account) |
| 4581 | entry->val = ent.val; |
| 4582 | |
| 4583 | return page; |
| 4584 | } |
| 4585 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4586 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4587 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4588 | { |
| 4589 | struct page *page = NULL; |
| 4590 | struct inode *inode; |
| 4591 | struct address_space *mapping; |
| 4592 | pgoff_t pgoff; |
| 4593 | |
| 4594 | if (!vma->vm_file) /* anonymous vma */ |
| 4595 | return NULL; |
| 4596 | if (!move_file()) |
| 4597 | return NULL; |
| 4598 | |
| 4599 | inode = vma->vm_file->f_path.dentry->d_inode; |
| 4600 | mapping = vma->vm_file->f_mapping; |
| 4601 | if (pte_none(ptent)) |
| 4602 | pgoff = linear_page_index(vma, addr); |
| 4603 | else /* pte_file(ptent) is true */ |
| 4604 | pgoff = pte_to_pgoff(ptent); |
| 4605 | |
| 4606 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
| 4607 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ |
| 4608 | page = find_get_page(mapping, pgoff); |
| 4609 | } else { /* shmem/tmpfs file. we should take account of swap too. */ |
| 4610 | swp_entry_t ent; |
| 4611 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); |
| 4612 | if (do_swap_account) |
| 4613 | entry->val = ent.val; |
| 4614 | } |
| 4615 | |
| 4616 | return page; |
| 4617 | } |
| 4618 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4619 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
| 4620 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4621 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4622 | struct page *page = NULL; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4623 | struct page_cgroup *pc; |
| 4624 | int ret = 0; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4625 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4626 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4627 | if (pte_present(ptent)) |
| 4628 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4629 | else if (is_swap_pte(ptent)) |
| 4630 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4631 | else if (pte_none(ptent) || pte_file(ptent)) |
| 4632 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4633 | |
| 4634 | if (!page && !ent.val) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4635 | return 0; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4636 | if (page) { |
| 4637 | pc = lookup_page_cgroup(page); |
| 4638 | /* |
| 4639 | * Do only loose check w/o page_cgroup lock. |
| 4640 | * mem_cgroup_move_account() checks the pc is valid or not under |
| 4641 | * the lock. |
| 4642 | */ |
| 4643 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 4644 | ret = MC_TARGET_PAGE; |
| 4645 | if (target) |
| 4646 | target->page = page; |
| 4647 | } |
| 4648 | if (!ret || !target) |
| 4649 | put_page(page); |
| 4650 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4651 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4652 | if (ent.val && !ret && |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4653 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
| 4654 | ret = MC_TARGET_SWAP; |
| 4655 | if (target) |
| 4656 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4657 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4658 | return ret; |
| 4659 | } |
| 4660 | |
| 4661 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4662 | unsigned long addr, unsigned long end, |
| 4663 | struct mm_walk *walk) |
| 4664 | { |
| 4665 | struct vm_area_struct *vma = walk->private; |
| 4666 | pte_t *pte; |
| 4667 | spinlock_t *ptl; |
| 4668 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 4669 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4670 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4671 | for (; addr != end; pte++, addr += PAGE_SIZE) |
| 4672 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) |
| 4673 | mc.precharge++; /* increment precharge temporarily */ |
| 4674 | pte_unmap_unlock(pte - 1, ptl); |
| 4675 | cond_resched(); |
| 4676 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4677 | return 0; |
| 4678 | } |
| 4679 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4680 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4681 | { |
| 4682 | unsigned long precharge; |
| 4683 | struct vm_area_struct *vma; |
| 4684 | |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4685 | /* We've already held the mmap_sem */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4686 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 4687 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4688 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4689 | .mm = mm, |
| 4690 | .private = vma, |
| 4691 | }; |
| 4692 | if (is_vm_hugetlb_page(vma)) |
| 4693 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4694 | walk_page_range(vma->vm_start, vma->vm_end, |
| 4695 | &mem_cgroup_count_precharge_walk); |
| 4696 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4697 | |
| 4698 | precharge = mc.precharge; |
| 4699 | mc.precharge = 0; |
| 4700 | |
| 4701 | return precharge; |
| 4702 | } |
| 4703 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4704 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4705 | { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4706 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4707 | } |
| 4708 | |
| 4709 | static void mem_cgroup_clear_mc(void) |
| 4710 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4711 | struct mem_cgroup *from = mc.from; |
| 4712 | struct mem_cgroup *to = mc.to; |
| 4713 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4714 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4715 | if (mc.precharge) { |
| 4716 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); |
| 4717 | mc.precharge = 0; |
| 4718 | } |
| 4719 | /* |
| 4720 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4721 | * we must uncharge here. |
| 4722 | */ |
| 4723 | if (mc.moved_charge) { |
| 4724 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
| 4725 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4726 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4727 | /* we must fixup refcnts and charges */ |
| 4728 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4729 | /* uncharge swap account from the old cgroup */ |
| 4730 | if (!mem_cgroup_is_root(mc.from)) |
| 4731 | res_counter_uncharge(&mc.from->memsw, |
| 4732 | PAGE_SIZE * mc.moved_swap); |
| 4733 | __mem_cgroup_put(mc.from, mc.moved_swap); |
| 4734 | |
| 4735 | if (!mem_cgroup_is_root(mc.to)) { |
| 4736 | /* |
| 4737 | * we charged both to->res and to->memsw, so we should |
| 4738 | * uncharge to->res. |
| 4739 | */ |
| 4740 | res_counter_uncharge(&mc.to->res, |
| 4741 | PAGE_SIZE * mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4742 | } |
| 4743 | /* we've already done mem_cgroup_get(mc.to) */ |
| 4744 | |
| 4745 | mc.moved_swap = 0; |
| 4746 | } |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4747 | if (mc.mm) { |
| 4748 | up_read(&mc.mm->mmap_sem); |
| 4749 | mmput(mc.mm); |
| 4750 | } |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4751 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4752 | mc.from = NULL; |
| 4753 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4754 | spin_unlock(&mc.lock); |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4755 | mc.moving_task = NULL; |
| 4756 | mc.mm = NULL; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 4757 | mem_cgroup_end_move(from); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4758 | memcg_oom_recover(from); |
| 4759 | memcg_oom_recover(to); |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 4760 | wake_up_all(&mc.waitq); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4761 | } |
| 4762 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4763 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
| 4764 | struct cgroup *cgroup, |
| 4765 | struct task_struct *p, |
| 4766 | bool threadgroup) |
| 4767 | { |
| 4768 | int ret = 0; |
| 4769 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); |
| 4770 | |
| 4771 | if (mem->move_charge_at_immigrate) { |
| 4772 | struct mm_struct *mm; |
| 4773 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 4774 | |
| 4775 | VM_BUG_ON(from == mem); |
| 4776 | |
| 4777 | mm = get_task_mm(p); |
| 4778 | if (!mm) |
| 4779 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4780 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4781 | if (mm->owner == p) { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4782 | /* |
| 4783 | * We do all the move charge works under one mmap_sem to |
| 4784 | * avoid deadlock with down_write(&mmap_sem) |
| 4785 | * -> try_charge() -> if (mc.moving_task) -> sleep. |
| 4786 | */ |
| 4787 | down_read(&mm->mmap_sem); |
| 4788 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4789 | VM_BUG_ON(mc.from); |
| 4790 | VM_BUG_ON(mc.to); |
| 4791 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4792 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4793 | VM_BUG_ON(mc.moved_swap); |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 4794 | VM_BUG_ON(mc.moving_task); |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4795 | VM_BUG_ON(mc.mm); |
| 4796 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 4797 | mem_cgroup_start_move(from); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4798 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4799 | mc.from = from; |
| 4800 | mc.to = mem; |
| 4801 | mc.precharge = 0; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4802 | mc.moved_charge = 0; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4803 | mc.moved_swap = 0; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4804 | spin_unlock(&mc.lock); |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4805 | mc.moving_task = current; |
| 4806 | mc.mm = mm; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4807 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4808 | ret = mem_cgroup_precharge_mc(mm); |
| 4809 | if (ret) |
| 4810 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4811 | /* We call up_read() and mmput() in clear_mc(). */ |
| 4812 | } else |
| 4813 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4814 | } |
| 4815 | return ret; |
| 4816 | } |
| 4817 | |
| 4818 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, |
| 4819 | struct cgroup *cgroup, |
| 4820 | struct task_struct *p, |
| 4821 | bool threadgroup) |
| 4822 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4823 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4824 | } |
| 4825 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4826 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 4827 | unsigned long addr, unsigned long end, |
| 4828 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4829 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4830 | int ret = 0; |
| 4831 | struct vm_area_struct *vma = walk->private; |
| 4832 | pte_t *pte; |
| 4833 | spinlock_t *ptl; |
| 4834 | |
| 4835 | retry: |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 4836 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4837 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4838 | for (; addr != end; addr += PAGE_SIZE) { |
| 4839 | pte_t ptent = *(pte++); |
| 4840 | union mc_target target; |
| 4841 | int type; |
| 4842 | struct page *page; |
| 4843 | struct page_cgroup *pc; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4844 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4845 | |
| 4846 | if (!mc.precharge) |
| 4847 | break; |
| 4848 | |
| 4849 | type = is_target_pte_for_mc(vma, addr, ptent, &target); |
| 4850 | switch (type) { |
| 4851 | case MC_TARGET_PAGE: |
| 4852 | page = target.page; |
| 4853 | if (isolate_lru_page(page)) |
| 4854 | goto put; |
| 4855 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4856 | if (!mem_cgroup_move_account(pc, |
| 4857 | mc.from, mc.to, false)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4858 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4859 | /* we uncharge from mc.from later. */ |
| 4860 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4861 | } |
| 4862 | putback_lru_page(page); |
| 4863 | put: /* is_target_pte_for_mc() gets the page */ |
| 4864 | put_page(page); |
| 4865 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4866 | case MC_TARGET_SWAP: |
| 4867 | ent = target.ent; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4868 | if (!mem_cgroup_move_swap_account(ent, |
| 4869 | mc.from, mc.to, false)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4870 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4871 | /* we fixup refcnts and charges later. */ |
| 4872 | mc.moved_swap++; |
| 4873 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4874 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4875 | default: |
| 4876 | break; |
| 4877 | } |
| 4878 | } |
| 4879 | pte_unmap_unlock(pte - 1, ptl); |
| 4880 | cond_resched(); |
| 4881 | |
| 4882 | if (addr != end) { |
| 4883 | /* |
| 4884 | * We have consumed all precharges we got in can_attach(). |
| 4885 | * We try charge one by one, but don't do any additional |
| 4886 | * charges to mc.to if we have failed in charge once in attach() |
| 4887 | * phase. |
| 4888 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4889 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4890 | if (!ret) |
| 4891 | goto retry; |
| 4892 | } |
| 4893 | |
| 4894 | return ret; |
| 4895 | } |
| 4896 | |
| 4897 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 4898 | { |
| 4899 | struct vm_area_struct *vma; |
| 4900 | |
| 4901 | lru_add_drain_all(); |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4902 | /* We've already held the mmap_sem */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4903 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 4904 | int ret; |
| 4905 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 4906 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 4907 | .mm = mm, |
| 4908 | .private = vma, |
| 4909 | }; |
| 4910 | if (is_vm_hugetlb_page(vma)) |
| 4911 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4912 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
| 4913 | &mem_cgroup_move_charge_walk); |
| 4914 | if (ret) |
| 4915 | /* |
| 4916 | * means we have consumed all precharges and failed in |
| 4917 | * doing additional charge. Just abandon here. |
| 4918 | */ |
| 4919 | break; |
| 4920 | } |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4921 | } |
| 4922 | |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4923 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 4924 | struct cgroup *cont, |
| 4925 | struct cgroup *old_cont, |
Ben Blum | be367d0 | 2009-09-23 15:56:31 -0700 | [diff] [blame] | 4926 | struct task_struct *p, |
| 4927 | bool threadgroup) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4928 | { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4929 | if (!mc.mm) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4930 | /* no need to move charge */ |
| 4931 | return; |
| 4932 | |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 4933 | mem_cgroup_move_charge(mc.mm); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4934 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4935 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 4936 | #else /* !CONFIG_MMU */ |
| 4937 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
| 4938 | struct cgroup *cgroup, |
| 4939 | struct task_struct *p, |
| 4940 | bool threadgroup) |
| 4941 | { |
| 4942 | return 0; |
| 4943 | } |
| 4944 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, |
| 4945 | struct cgroup *cgroup, |
| 4946 | struct task_struct *p, |
| 4947 | bool threadgroup) |
| 4948 | { |
| 4949 | } |
| 4950 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 4951 | struct cgroup *cont, |
| 4952 | struct cgroup *old_cont, |
| 4953 | struct task_struct *p, |
| 4954 | bool threadgroup) |
| 4955 | { |
| 4956 | } |
| 4957 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4958 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4959 | struct cgroup_subsys mem_cgroup_subsys = { |
| 4960 | .name = "memory", |
| 4961 | .subsys_id = mem_cgroup_subsys_id, |
| 4962 | .create = mem_cgroup_create, |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4963 | .pre_destroy = mem_cgroup_pre_destroy, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4964 | .destroy = mem_cgroup_destroy, |
| 4965 | .populate = mem_cgroup_populate, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4966 | .can_attach = mem_cgroup_can_attach, |
| 4967 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4968 | .attach = mem_cgroup_move_task, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4969 | .early_init = 0, |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4970 | .use_id = 1, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4971 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4972 | |
| 4973 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 4974 | static int __init enable_swap_account(char *s) |
| 4975 | { |
| 4976 | /* consider enabled if no parameter or 1 is given */ |
| 4977 | if (!s || !strcmp(s, "1")) |
| 4978 | really_do_swap_account = 1; |
| 4979 | else if (!strcmp(s, "0")) |
| 4980 | really_do_swap_account = 0; |
| 4981 | return 1; |
| 4982 | } |
| 4983 | __setup("swapaccount", enable_swap_account); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4984 | |
| 4985 | static int __init disable_swap_account(char *s) |
| 4986 | { |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 4987 | enable_swap_account("0"); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4988 | return 1; |
| 4989 | } |
| 4990 | __setup("noswapaccount", disable_swap_account); |
| 4991 | #endif |