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 */ |
| 295 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 296 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 297 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 298 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 299 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 300 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 301 | static bool move_anon(void) |
| 302 | { |
| 303 | return test_bit(MOVE_CHARGE_TYPE_ANON, |
| 304 | &mc.to->move_charge_at_immigrate); |
| 305 | } |
| 306 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 307 | static bool move_file(void) |
| 308 | { |
| 309 | return test_bit(MOVE_CHARGE_TYPE_FILE, |
| 310 | &mc.to->move_charge_at_immigrate); |
| 311 | } |
| 312 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 313 | /* |
| 314 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 315 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 316 | */ |
| 317 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) |
| 318 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) |
| 319 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 320 | enum charge_type { |
| 321 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
| 322 | MEM_CGROUP_CHARGE_TYPE_MAPPED, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 323 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 324 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 325 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 326 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 327 | NR_CHARGE_TYPE, |
| 328 | }; |
| 329 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 330 | /* for encoding cft->private value on file */ |
| 331 | #define _MEM (0) |
| 332 | #define _MEMSWAP (1) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 333 | #define _OOM_TYPE (2) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 334 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
| 335 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) |
| 336 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 337 | /* Used for OOM nofiier */ |
| 338 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 339 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 340 | /* |
| 341 | * Reclaim flags for mem_cgroup_hierarchical_reclaim |
| 342 | */ |
| 343 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 |
| 344 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) |
| 345 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 |
| 346 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 347 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
| 348 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 349 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 350 | static void mem_cgroup_get(struct mem_cgroup *mem); |
| 351 | static void mem_cgroup_put(struct mem_cgroup *mem); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 352 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 353 | static void drain_all_stock_async(void); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 354 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 355 | static struct mem_cgroup_per_zone * |
| 356 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
| 357 | { |
| 358 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
| 359 | } |
| 360 | |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 361 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
| 362 | { |
| 363 | return &mem->css; |
| 364 | } |
| 365 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 366 | static struct mem_cgroup_per_zone * |
| 367 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
| 368 | { |
| 369 | struct mem_cgroup *mem = pc->mem_cgroup; |
| 370 | int nid = page_cgroup_nid(pc); |
| 371 | int zid = page_cgroup_zid(pc); |
| 372 | |
| 373 | if (!mem) |
| 374 | return NULL; |
| 375 | |
| 376 | return mem_cgroup_zoneinfo(mem, nid, zid); |
| 377 | } |
| 378 | |
| 379 | static struct mem_cgroup_tree_per_zone * |
| 380 | soft_limit_tree_node_zone(int nid, int zid) |
| 381 | { |
| 382 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 383 | } |
| 384 | |
| 385 | static struct mem_cgroup_tree_per_zone * |
| 386 | soft_limit_tree_from_page(struct page *page) |
| 387 | { |
| 388 | int nid = page_to_nid(page); |
| 389 | int zid = page_zonenum(page); |
| 390 | |
| 391 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 392 | } |
| 393 | |
| 394 | static void |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 395 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 396 | struct mem_cgroup_per_zone *mz, |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 397 | struct mem_cgroup_tree_per_zone *mctz, |
| 398 | unsigned long long new_usage_in_excess) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 399 | { |
| 400 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 401 | struct rb_node *parent = NULL; |
| 402 | struct mem_cgroup_per_zone *mz_node; |
| 403 | |
| 404 | if (mz->on_tree) |
| 405 | return; |
| 406 | |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 407 | mz->usage_in_excess = new_usage_in_excess; |
| 408 | if (!mz->usage_in_excess) |
| 409 | return; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 410 | while (*p) { |
| 411 | parent = *p; |
| 412 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 413 | tree_node); |
| 414 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 415 | p = &(*p)->rb_left; |
| 416 | /* |
| 417 | * We can't avoid mem cgroups that are over their soft |
| 418 | * limit by the same amount |
| 419 | */ |
| 420 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 421 | p = &(*p)->rb_right; |
| 422 | } |
| 423 | rb_link_node(&mz->tree_node, parent, p); |
| 424 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 425 | mz->on_tree = true; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 426 | } |
| 427 | |
| 428 | static void |
| 429 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
| 430 | struct mem_cgroup_per_zone *mz, |
| 431 | struct mem_cgroup_tree_per_zone *mctz) |
| 432 | { |
| 433 | if (!mz->on_tree) |
| 434 | return; |
| 435 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 436 | mz->on_tree = false; |
| 437 | } |
| 438 | |
| 439 | static void |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 440 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
| 441 | struct mem_cgroup_per_zone *mz, |
| 442 | struct mem_cgroup_tree_per_zone *mctz) |
| 443 | { |
| 444 | spin_lock(&mctz->lock); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 445 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 446 | spin_unlock(&mctz->lock); |
| 447 | } |
| 448 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 449 | |
| 450 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) |
| 451 | { |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 452 | unsigned long long excess; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 453 | struct mem_cgroup_per_zone *mz; |
| 454 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 455 | int nid = page_to_nid(page); |
| 456 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 457 | mctz = soft_limit_tree_from_page(page); |
| 458 | |
| 459 | /* |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 460 | * Necessary to update all ancestors when hierarchy is used. |
| 461 | * because their event counter is not touched. |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 462 | */ |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 463 | for (; mem; mem = parent_mem_cgroup(mem)) { |
| 464 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 465 | excess = res_counter_soft_limit_excess(&mem->res); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 466 | /* |
| 467 | * We have to update the tree if mz is on RB-tree or |
| 468 | * mem is over its softlimit. |
| 469 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 470 | if (excess || mz->on_tree) { |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 471 | spin_lock(&mctz->lock); |
| 472 | /* if on-tree, remove it */ |
| 473 | if (mz->on_tree) |
| 474 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
| 475 | /* |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 476 | * Insert again. mz->usage_in_excess will be updated. |
| 477 | * If excess is 0, no tree ops. |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 478 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 479 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 480 | spin_unlock(&mctz->lock); |
| 481 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 482 | } |
| 483 | } |
| 484 | |
| 485 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) |
| 486 | { |
| 487 | int node, zone; |
| 488 | struct mem_cgroup_per_zone *mz; |
| 489 | struct mem_cgroup_tree_per_zone *mctz; |
| 490 | |
| 491 | for_each_node_state(node, N_POSSIBLE) { |
| 492 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 493 | mz = mem_cgroup_zoneinfo(mem, node, zone); |
| 494 | mctz = soft_limit_tree_node_zone(node, zone); |
| 495 | mem_cgroup_remove_exceeded(mem, mz, mctz); |
| 496 | } |
| 497 | } |
| 498 | } |
| 499 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 500 | static struct mem_cgroup_per_zone * |
| 501 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 502 | { |
| 503 | struct rb_node *rightmost = NULL; |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 504 | struct mem_cgroup_per_zone *mz; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 505 | |
| 506 | retry: |
KAMEZAWA Hiroyuki | 26251ea | 2009-10-01 15:44:08 -0700 | [diff] [blame] | 507 | mz = NULL; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 508 | rightmost = rb_last(&mctz->rb_root); |
| 509 | if (!rightmost) |
| 510 | goto done; /* Nothing to reclaim from */ |
| 511 | |
| 512 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 513 | /* |
| 514 | * Remove the node now but someone else can add it back, |
| 515 | * we will to add it back at the end of reclaim to its correct |
| 516 | * position in the tree. |
| 517 | */ |
| 518 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
| 519 | if (!res_counter_soft_limit_excess(&mz->mem->res) || |
| 520 | !css_tryget(&mz->mem->css)) |
| 521 | goto retry; |
| 522 | done: |
| 523 | return mz; |
| 524 | } |
| 525 | |
| 526 | static struct mem_cgroup_per_zone * |
| 527 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 528 | { |
| 529 | struct mem_cgroup_per_zone *mz; |
| 530 | |
| 531 | spin_lock(&mctz->lock); |
| 532 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 533 | spin_unlock(&mctz->lock); |
| 534 | return mz; |
| 535 | } |
| 536 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 537 | /* |
| 538 | * Implementation Note: reading percpu statistics for memcg. |
| 539 | * |
| 540 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 541 | * synchronization to implement "quick" read. There are trade-off between |
| 542 | * reading cost and precision of value. Then, we may have a chance to implement |
| 543 | * a periodic synchronizion of counter in memcg's counter. |
| 544 | * |
| 545 | * But this _read() function is used for user interface now. The user accounts |
| 546 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 547 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 548 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 549 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 550 | * |
| 551 | * If there are kernel internal actions which can make use of some not-exact |
| 552 | * value, and reading all cpu value can be performance bottleneck in some |
| 553 | * common workload, threashold and synchonization as vmstat[] should be |
| 554 | * implemented. |
| 555 | */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 556 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
| 557 | enum mem_cgroup_stat_index idx) |
| 558 | { |
| 559 | int cpu; |
| 560 | s64 val = 0; |
| 561 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 562 | get_online_cpus(); |
| 563 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 564 | val += per_cpu(mem->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 565 | #ifdef CONFIG_HOTPLUG_CPU |
| 566 | spin_lock(&mem->pcp_counter_lock); |
| 567 | val += mem->nocpu_base.count[idx]; |
| 568 | spin_unlock(&mem->pcp_counter_lock); |
| 569 | #endif |
| 570 | put_online_cpus(); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 571 | return val; |
| 572 | } |
| 573 | |
| 574 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) |
| 575 | { |
| 576 | s64 ret; |
| 577 | |
| 578 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
| 579 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
| 580 | return ret; |
| 581 | } |
| 582 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 583 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
| 584 | bool charge) |
| 585 | { |
| 586 | int val = (charge) ? 1 : -1; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 587 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 588 | } |
| 589 | |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 590 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 591 | bool file, int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 592 | { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 593 | preempt_disable(); |
| 594 | |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 595 | if (file) |
| 596 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 597 | else |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 598 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 599 | |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 600 | /* pagein of a big page is an event. So, ignore page size */ |
| 601 | if (nr_pages > 0) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 602 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 603 | else { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 604 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 605 | nr_pages = -nr_pages; /* for event */ |
| 606 | } |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 607 | |
| 608 | __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 609 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 610 | preempt_enable(); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 611 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 612 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 613 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 614 | enum lru_list idx) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 615 | { |
| 616 | int nid, zid; |
| 617 | struct mem_cgroup_per_zone *mz; |
| 618 | u64 total = 0; |
| 619 | |
| 620 | for_each_online_node(nid) |
| 621 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 622 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
| 623 | total += MEM_CGROUP_ZSTAT(mz, idx); |
| 624 | } |
| 625 | return total; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 626 | } |
| 627 | |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 628 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
| 629 | { |
| 630 | s64 val; |
| 631 | |
| 632 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); |
| 633 | |
| 634 | return !(val & ((1 << event_mask_shift) - 1)); |
| 635 | } |
| 636 | |
| 637 | /* |
| 638 | * Check events in order. |
| 639 | * |
| 640 | */ |
| 641 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) |
| 642 | { |
| 643 | /* threshold event is triggered in finer grain than soft limit */ |
| 644 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { |
| 645 | mem_cgroup_threshold(mem); |
| 646 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) |
| 647 | mem_cgroup_update_tree(mem, page); |
| 648 | } |
| 649 | } |
| 650 | |
Hugh Dickins | d5b69e3 | 2008-03-04 14:29:10 -0800 | [diff] [blame] | 651 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 652 | { |
| 653 | return container_of(cgroup_subsys_state(cont, |
| 654 | mem_cgroup_subsys_id), struct mem_cgroup, |
| 655 | css); |
| 656 | } |
| 657 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 658 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 659 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 660 | /* |
| 661 | * mm_update_next_owner() may clear mm->owner to NULL |
| 662 | * if it races with swapoff, page migration, etc. |
| 663 | * So this can be called with p == NULL. |
| 664 | */ |
| 665 | if (unlikely(!p)) |
| 666 | return NULL; |
| 667 | |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 668 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
| 669 | struct mem_cgroup, css); |
| 670 | } |
| 671 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 672 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
| 673 | { |
| 674 | struct mem_cgroup *mem = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 675 | |
| 676 | if (!mm) |
| 677 | return NULL; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 678 | /* |
| 679 | * Because we have no locks, mm->owner's may be being moved to other |
| 680 | * cgroup. We use css_tryget() here even if this looks |
| 681 | * pessimistic (rather than adding locks here). |
| 682 | */ |
| 683 | rcu_read_lock(); |
| 684 | do { |
| 685 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 686 | if (unlikely(!mem)) |
| 687 | break; |
| 688 | } while (!css_tryget(&mem->css)); |
| 689 | rcu_read_unlock(); |
| 690 | return mem; |
| 691 | } |
| 692 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 693 | /* The caller has to guarantee "mem" exists before calling this */ |
| 694 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 695 | { |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 696 | struct cgroup_subsys_state *css; |
| 697 | int found; |
| 698 | |
| 699 | if (!mem) /* ROOT cgroup has the smallest ID */ |
| 700 | return root_mem_cgroup; /*css_put/get against root is ignored*/ |
| 701 | if (!mem->use_hierarchy) { |
| 702 | if (css_tryget(&mem->css)) |
| 703 | return mem; |
| 704 | return NULL; |
| 705 | } |
| 706 | rcu_read_lock(); |
| 707 | /* |
| 708 | * searching a memory cgroup which has the smallest ID under given |
| 709 | * ROOT cgroup. (ID >= 1) |
| 710 | */ |
| 711 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); |
| 712 | if (css && css_tryget(css)) |
| 713 | mem = container_of(css, struct mem_cgroup, css); |
| 714 | else |
| 715 | mem = NULL; |
| 716 | rcu_read_unlock(); |
| 717 | return mem; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 718 | } |
| 719 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 720 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, |
| 721 | struct mem_cgroup *root, |
| 722 | bool cond) |
| 723 | { |
| 724 | int nextid = css_id(&iter->css) + 1; |
| 725 | int found; |
| 726 | int hierarchy_used; |
| 727 | struct cgroup_subsys_state *css; |
| 728 | |
| 729 | hierarchy_used = iter->use_hierarchy; |
| 730 | |
| 731 | css_put(&iter->css); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 732 | /* If no ROOT, walk all, ignore hierarchy */ |
| 733 | if (!cond || (root && !hierarchy_used)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 734 | return NULL; |
| 735 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 736 | if (!root) |
| 737 | root = root_mem_cgroup; |
| 738 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 739 | do { |
| 740 | iter = NULL; |
| 741 | rcu_read_lock(); |
| 742 | |
| 743 | css = css_get_next(&mem_cgroup_subsys, nextid, |
| 744 | &root->css, &found); |
| 745 | if (css && css_tryget(css)) |
| 746 | iter = container_of(css, struct mem_cgroup, css); |
| 747 | rcu_read_unlock(); |
| 748 | /* If css is NULL, no more cgroups will be found */ |
| 749 | nextid = found + 1; |
| 750 | } while (css && !iter); |
| 751 | |
| 752 | return iter; |
| 753 | } |
| 754 | /* |
| 755 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please |
| 756 | * be careful that "break" loop is not allowed. We have reference count. |
| 757 | * Instead of that modify "cond" to be false and "continue" to exit the loop. |
| 758 | */ |
| 759 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ |
| 760 | for (iter = mem_cgroup_start_loop(root);\ |
| 761 | iter != NULL;\ |
| 762 | iter = mem_cgroup_get_next(iter, root, cond)) |
| 763 | |
| 764 | #define for_each_mem_cgroup_tree(iter, root) \ |
| 765 | for_each_mem_cgroup_tree_cond(iter, root, true) |
| 766 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 767 | #define for_each_mem_cgroup_all(iter) \ |
| 768 | for_each_mem_cgroup_tree_cond(iter, NULL, true) |
| 769 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 770 | |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 771 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
| 772 | { |
| 773 | return (mem == root_mem_cgroup); |
| 774 | } |
| 775 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 776 | /* |
| 777 | * Following LRU functions are allowed to be used without PCG_LOCK. |
| 778 | * Operations are called by routine of global LRU independently from memcg. |
| 779 | * What we have to take care of here is validness of pc->mem_cgroup. |
| 780 | * |
| 781 | * Changes to pc->mem_cgroup happens when |
| 782 | * 1. charge |
| 783 | * 2. moving account |
| 784 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. |
| 785 | * It is added to LRU before charge. |
| 786 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. |
| 787 | * When moving account, the page is not on LRU. It's isolated. |
| 788 | */ |
| 789 | |
| 790 | 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] | 791 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 792 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 793 | struct mem_cgroup_per_zone *mz; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 794 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 795 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 796 | return; |
| 797 | pc = lookup_page_cgroup(page); |
| 798 | /* can happen while we handle swapcache. */ |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 799 | if (!TestClearPageCgroupAcctLRU(pc)) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 800 | return; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 801 | VM_BUG_ON(!pc->mem_cgroup); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 802 | /* |
| 803 | * We don't check PCG_USED bit. It's cleared when the "page" is finally |
| 804 | * removed from global LRU. |
| 805 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 806 | mz = page_cgroup_zoneinfo(pc); |
KAMEZAWA Hiroyuki | ece35ca | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 807 | /* huge page split is done under lru_lock. so, we have no races. */ |
| 808 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 809 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
| 810 | return; |
| 811 | VM_BUG_ON(list_empty(&pc->lru)); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 812 | list_del_init(&pc->lru); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 813 | } |
| 814 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 815 | void mem_cgroup_del_lru(struct page *page) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 816 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 817 | mem_cgroup_del_lru_list(page, page_lru(page)); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 818 | } |
| 819 | |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 820 | /* |
| 821 | * Writeback is about to end against a page which has been marked for immediate |
| 822 | * reclaim. If it still appears to be reclaimable, move it to the tail of the |
| 823 | * inactive list. |
| 824 | */ |
| 825 | void mem_cgroup_rotate_reclaimable_page(struct page *page) |
| 826 | { |
| 827 | struct mem_cgroup_per_zone *mz; |
| 828 | struct page_cgroup *pc; |
| 829 | enum lru_list lru = page_lru(page); |
| 830 | |
| 831 | if (mem_cgroup_disabled()) |
| 832 | return; |
| 833 | |
| 834 | pc = lookup_page_cgroup(page); |
| 835 | /* unused or root page is not rotated. */ |
| 836 | if (!PageCgroupUsed(pc)) |
| 837 | return; |
| 838 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
| 839 | smp_rmb(); |
| 840 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
| 841 | return; |
| 842 | mz = page_cgroup_zoneinfo(pc); |
| 843 | list_move_tail(&pc->lru, &mz->lists[lru]); |
| 844 | } |
| 845 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 846 | 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] | 847 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 848 | struct mem_cgroup_per_zone *mz; |
| 849 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 850 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 851 | if (mem_cgroup_disabled()) |
Lee Schermerhorn | 894bc31 | 2008-10-18 20:26:39 -0700 | [diff] [blame] | 852 | return; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 853 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 854 | pc = lookup_page_cgroup(page); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 855 | /* unused or root page is not rotated. */ |
Johannes Weiner | 713735b | 2011-01-20 14:44:31 -0800 | [diff] [blame] | 856 | if (!PageCgroupUsed(pc)) |
| 857 | return; |
| 858 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
| 859 | smp_rmb(); |
| 860 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 861 | return; |
| 862 | mz = page_cgroup_zoneinfo(pc); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 863 | list_move(&pc->lru, &mz->lists[lru]); |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 864 | } |
| 865 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 866 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
| 867 | { |
| 868 | struct page_cgroup *pc; |
| 869 | struct mem_cgroup_per_zone *mz; |
| 870 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 871 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 872 | return; |
| 873 | pc = lookup_page_cgroup(page); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 874 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 875 | if (!PageCgroupUsed(pc)) |
| 876 | return; |
Johannes Weiner | 713735b | 2011-01-20 14:44:31 -0800 | [diff] [blame] | 877 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
| 878 | smp_rmb(); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 879 | mz = page_cgroup_zoneinfo(pc); |
KAMEZAWA Hiroyuki | ece35ca | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 880 | /* huge page split is done under lru_lock. so, we have no races. */ |
| 881 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 882 | SetPageCgroupAcctLRU(pc); |
| 883 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
| 884 | return; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 885 | list_add(&pc->lru, &mz->lists[lru]); |
| 886 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 887 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 888 | /* |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 889 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
| 890 | * lru because the page may.be reused after it's fully uncharged (because of |
| 891 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge |
| 892 | * it again. This function is only used to charge SwapCache. It's done under |
| 893 | * lock_page and expected that zone->lru_lock is never held. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 894 | */ |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 895 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 896 | { |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 897 | unsigned long flags; |
| 898 | struct zone *zone = page_zone(page); |
| 899 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 900 | |
| 901 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 902 | /* |
| 903 | * Forget old LRU when this page_cgroup is *not* used. This Used bit |
| 904 | * is guarded by lock_page() because the page is SwapCache. |
| 905 | */ |
| 906 | if (!PageCgroupUsed(pc)) |
| 907 | mem_cgroup_del_lru_list(page, page_lru(page)); |
| 908 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 909 | } |
| 910 | |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 911 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
| 912 | { |
| 913 | unsigned long flags; |
| 914 | struct zone *zone = page_zone(page); |
| 915 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 916 | |
| 917 | spin_lock_irqsave(&zone->lru_lock, flags); |
| 918 | /* 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] | 919 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 920 | mem_cgroup_add_lru_list(page, page_lru(page)); |
| 921 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
| 922 | } |
| 923 | |
| 924 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 925 | void mem_cgroup_move_lists(struct page *page, |
| 926 | enum lru_list from, enum lru_list to) |
| 927 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 928 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 929 | return; |
| 930 | mem_cgroup_del_lru_list(page, from); |
| 931 | mem_cgroup_add_lru_list(page, to); |
| 932 | } |
| 933 | |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 934 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
| 935 | { |
| 936 | int ret; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 937 | struct mem_cgroup *curr = NULL; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 938 | struct task_struct *p; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 939 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 940 | p = find_lock_task_mm(task); |
| 941 | if (!p) |
| 942 | return 0; |
| 943 | curr = try_get_mem_cgroup_from_mm(p->mm); |
| 944 | task_unlock(p); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 945 | if (!curr) |
| 946 | return 0; |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 947 | /* |
| 948 | * We should check use_hierarchy of "mem" not "curr". Because checking |
| 949 | * use_hierarchy of "curr" here make this function true if hierarchy is |
| 950 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* |
| 951 | * hierarchy(even if use_hierarchy is disabled in "mem"). |
| 952 | */ |
| 953 | if (mem->use_hierarchy) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 954 | ret = css_is_ancestor(&curr->css, &mem->css); |
| 955 | else |
| 956 | ret = (curr == mem); |
| 957 | css_put(&curr->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 958 | return ret; |
| 959 | } |
| 960 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 961 | 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] | 962 | { |
| 963 | unsigned long active; |
| 964 | unsigned long inactive; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 965 | unsigned long gb; |
| 966 | unsigned long inactive_ratio; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 967 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 968 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
| 969 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 970 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 971 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 972 | if (gb) |
| 973 | inactive_ratio = int_sqrt(10 * gb); |
| 974 | else |
| 975 | inactive_ratio = 1; |
| 976 | |
| 977 | if (present_pages) { |
| 978 | present_pages[0] = inactive; |
| 979 | present_pages[1] = active; |
| 980 | } |
| 981 | |
| 982 | return inactive_ratio; |
| 983 | } |
| 984 | |
| 985 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) |
| 986 | { |
| 987 | unsigned long active; |
| 988 | unsigned long inactive; |
| 989 | unsigned long present_pages[2]; |
| 990 | unsigned long inactive_ratio; |
| 991 | |
| 992 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); |
| 993 | |
| 994 | inactive = present_pages[0]; |
| 995 | active = present_pages[1]; |
| 996 | |
| 997 | if (inactive * inactive_ratio < active) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 998 | return 1; |
| 999 | |
| 1000 | return 0; |
| 1001 | } |
| 1002 | |
Rik van Riel | 56e49d2 | 2009-06-16 15:32:28 -0700 | [diff] [blame] | 1003 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
| 1004 | { |
| 1005 | unsigned long active; |
| 1006 | unsigned long inactive; |
| 1007 | |
| 1008 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); |
| 1009 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); |
| 1010 | |
| 1011 | return (active > inactive); |
| 1012 | } |
| 1013 | |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -0800 | [diff] [blame] | 1014 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
| 1015 | struct zone *zone, |
| 1016 | enum lru_list lru) |
| 1017 | { |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1018 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | a3d8e05 | 2009-01-07 18:08:19 -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 MEM_CGROUP_ZSTAT(mz, lru); |
| 1023 | } |
| 1024 | |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1025 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
| 1026 | struct zone *zone) |
| 1027 | { |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1028 | int nid = zone_to_nid(zone); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1029 | int zid = zone_idx(zone); |
| 1030 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 1031 | |
| 1032 | return &mz->reclaim_stat; |
| 1033 | } |
| 1034 | |
| 1035 | struct zone_reclaim_stat * |
| 1036 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) |
| 1037 | { |
| 1038 | struct page_cgroup *pc; |
| 1039 | struct mem_cgroup_per_zone *mz; |
| 1040 | |
| 1041 | if (mem_cgroup_disabled()) |
| 1042 | return NULL; |
| 1043 | |
| 1044 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | bd112db | 2009-01-15 13:51:11 -0800 | [diff] [blame] | 1045 | if (!PageCgroupUsed(pc)) |
| 1046 | return NULL; |
Johannes Weiner | 713735b | 2011-01-20 14:44:31 -0800 | [diff] [blame] | 1047 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
| 1048 | smp_rmb(); |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 1049 | mz = page_cgroup_zoneinfo(pc); |
| 1050 | if (!mz) |
| 1051 | return NULL; |
| 1052 | |
| 1053 | return &mz->reclaim_stat; |
| 1054 | } |
| 1055 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1056 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
| 1057 | struct list_head *dst, |
| 1058 | unsigned long *scanned, int order, |
| 1059 | int mode, struct zone *z, |
| 1060 | struct mem_cgroup *mem_cont, |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 1061 | int active, int file) |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1062 | { |
| 1063 | unsigned long nr_taken = 0; |
| 1064 | struct page *page; |
| 1065 | unsigned long scan; |
| 1066 | LIST_HEAD(pc_list); |
| 1067 | struct list_head *src; |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1068 | struct page_cgroup *pc, *tmp; |
KOSAKI Motohiro | 13d7e3a | 2010-08-10 18:03:06 -0700 | [diff] [blame] | 1069 | int nid = zone_to_nid(z); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 1070 | int zid = zone_idx(z); |
| 1071 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | b7c46d1 | 2009-09-21 17:02:56 -0700 | [diff] [blame] | 1072 | int lru = LRU_FILE * file + active; |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1073 | int ret; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1074 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 1075 | BUG_ON(!mem_cont); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 1076 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1077 | src = &mz->lists[lru]; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1078 | |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1079 | scan = 0; |
| 1080 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1081 | if (scan >= nr_to_scan) |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1082 | break; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1083 | |
| 1084 | page = pc->page; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 1085 | if (unlikely(!PageCgroupUsed(pc))) |
| 1086 | continue; |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1087 | if (unlikely(!PageLRU(page))) |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1088 | continue; |
KAMEZAWA Hiroyuki | ff7283f | 2008-02-07 00:14:11 -0800 | [diff] [blame] | 1089 | |
Hugh Dickins | 436c6541 | 2008-02-07 00:14:12 -0800 | [diff] [blame] | 1090 | scan++; |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1091 | ret = __isolate_lru_page(page, mode, file); |
| 1092 | switch (ret) { |
| 1093 | case 0: |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1094 | list_move(&page->lru, dst); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1095 | mem_cgroup_del_lru(page); |
Rik van Riel | 2c888cf | 2011-01-13 15:47:13 -0800 | [diff] [blame] | 1096 | nr_taken += hpage_nr_pages(page); |
KAMEZAWA Hiroyuki | 2ffebca | 2009-06-17 16:27:21 -0700 | [diff] [blame] | 1097 | break; |
| 1098 | case -EBUSY: |
| 1099 | /* we don't affect global LRU but rotate in our LRU */ |
| 1100 | mem_cgroup_rotate_lru_list(page, page_lru(page)); |
| 1101 | break; |
| 1102 | default: |
| 1103 | break; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1104 | } |
| 1105 | } |
| 1106 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1107 | *scanned = scan; |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 1108 | |
| 1109 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, |
| 1110 | 0, 0, 0, mode); |
| 1111 | |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 1112 | return nr_taken; |
| 1113 | } |
| 1114 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1115 | #define mem_cgroup_from_res_counter(counter, member) \ |
| 1116 | container_of(counter, struct mem_cgroup, member) |
| 1117 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1118 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1119 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
| 1120 | * @mem: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1121 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1122 | * Returns the maximum amount of memory @mem can be charged with, in |
| 1123 | * bytes. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1124 | */ |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1125 | static unsigned long long mem_cgroup_margin(struct mem_cgroup *mem) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1126 | { |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1127 | unsigned long long margin; |
| 1128 | |
| 1129 | margin = res_counter_margin(&mem->res); |
| 1130 | if (do_swap_account) |
| 1131 | margin = min(margin, res_counter_margin(&mem->memsw)); |
| 1132 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1133 | } |
| 1134 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1135 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
| 1136 | { |
| 1137 | struct cgroup *cgrp = memcg->css.cgroup; |
| 1138 | unsigned int swappiness; |
| 1139 | |
| 1140 | /* root ? */ |
| 1141 | if (cgrp->parent == NULL) |
| 1142 | return vm_swappiness; |
| 1143 | |
| 1144 | spin_lock(&memcg->reclaim_param_lock); |
| 1145 | swappiness = memcg->swappiness; |
| 1146 | spin_unlock(&memcg->reclaim_param_lock); |
| 1147 | |
| 1148 | return swappiness; |
| 1149 | } |
| 1150 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1151 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
| 1152 | { |
| 1153 | int cpu; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1154 | |
| 1155 | get_online_cpus(); |
| 1156 | spin_lock(&mem->pcp_counter_lock); |
| 1157 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1158 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1159 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
| 1160 | spin_unlock(&mem->pcp_counter_lock); |
| 1161 | put_online_cpus(); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1162 | |
| 1163 | synchronize_rcu(); |
| 1164 | } |
| 1165 | |
| 1166 | static void mem_cgroup_end_move(struct mem_cgroup *mem) |
| 1167 | { |
| 1168 | int cpu; |
| 1169 | |
| 1170 | if (!mem) |
| 1171 | return; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1172 | get_online_cpus(); |
| 1173 | spin_lock(&mem->pcp_counter_lock); |
| 1174 | for_each_online_cpu(cpu) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1175 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1176 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
| 1177 | spin_unlock(&mem->pcp_counter_lock); |
| 1178 | put_online_cpus(); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1179 | } |
| 1180 | /* |
| 1181 | * 2 routines for checking "mem" is under move_account() or not. |
| 1182 | * |
| 1183 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used |
| 1184 | * for avoiding race in accounting. If true, |
| 1185 | * pc->mem_cgroup may be overwritten. |
| 1186 | * |
| 1187 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or |
| 1188 | * under hierarchy of moving cgroups. This is for |
| 1189 | * waiting at hith-memory prressure caused by "move". |
| 1190 | */ |
| 1191 | |
| 1192 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) |
| 1193 | { |
| 1194 | VM_BUG_ON(!rcu_read_lock_held()); |
| 1195 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
| 1196 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1197 | |
| 1198 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) |
| 1199 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1200 | struct mem_cgroup *from; |
| 1201 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1202 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1203 | /* |
| 1204 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1205 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1206 | */ |
| 1207 | spin_lock(&mc.lock); |
| 1208 | from = mc.from; |
| 1209 | to = mc.to; |
| 1210 | if (!from) |
| 1211 | goto unlock; |
| 1212 | if (from == mem || to == mem |
| 1213 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) |
| 1214 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) |
| 1215 | ret = true; |
| 1216 | unlock: |
| 1217 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1218 | return ret; |
| 1219 | } |
| 1220 | |
| 1221 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) |
| 1222 | { |
| 1223 | if (mc.moving_task && current != mc.moving_task) { |
| 1224 | if (mem_cgroup_under_move(mem)) { |
| 1225 | DEFINE_WAIT(wait); |
| 1226 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1227 | /* moving charge context might have finished. */ |
| 1228 | if (mc.moving_task) |
| 1229 | schedule(); |
| 1230 | finish_wait(&mc.waitq, &wait); |
| 1231 | return true; |
| 1232 | } |
| 1233 | } |
| 1234 | return false; |
| 1235 | } |
| 1236 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1237 | /** |
Kirill A. Shutemov | 6a6135b | 2010-03-10 15:22:25 -0800 | [diff] [blame] | 1238 | * 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] | 1239 | * @memcg: The memory cgroup that went over limit |
| 1240 | * @p: Task that is going to be killed |
| 1241 | * |
| 1242 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1243 | * enabled |
| 1244 | */ |
| 1245 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1246 | { |
| 1247 | struct cgroup *task_cgrp; |
| 1248 | struct cgroup *mem_cgrp; |
| 1249 | /* |
| 1250 | * Need a buffer in BSS, can't rely on allocations. The code relies |
| 1251 | * on the assumption that OOM is serialized for memory controller. |
| 1252 | * If this assumption is broken, revisit this code. |
| 1253 | */ |
| 1254 | static char memcg_name[PATH_MAX]; |
| 1255 | int ret; |
| 1256 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1257 | if (!memcg || !p) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1258 | return; |
| 1259 | |
| 1260 | |
| 1261 | rcu_read_lock(); |
| 1262 | |
| 1263 | mem_cgrp = memcg->css.cgroup; |
| 1264 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); |
| 1265 | |
| 1266 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); |
| 1267 | if (ret < 0) { |
| 1268 | /* |
| 1269 | * Unfortunately, we are unable to convert to a useful name |
| 1270 | * But we'll still print out the usage information |
| 1271 | */ |
| 1272 | rcu_read_unlock(); |
| 1273 | goto done; |
| 1274 | } |
| 1275 | rcu_read_unlock(); |
| 1276 | |
| 1277 | printk(KERN_INFO "Task in %s killed", memcg_name); |
| 1278 | |
| 1279 | rcu_read_lock(); |
| 1280 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); |
| 1281 | if (ret < 0) { |
| 1282 | rcu_read_unlock(); |
| 1283 | goto done; |
| 1284 | } |
| 1285 | rcu_read_unlock(); |
| 1286 | |
| 1287 | /* |
| 1288 | * Continues from above, so we don't need an KERN_ level |
| 1289 | */ |
| 1290 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); |
| 1291 | done: |
| 1292 | |
| 1293 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", |
| 1294 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
| 1295 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, |
| 1296 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); |
| 1297 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " |
| 1298 | "failcnt %llu\n", |
| 1299 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
| 1300 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, |
| 1301 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); |
| 1302 | } |
| 1303 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1304 | /* |
| 1305 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1306 | * 1(self count) if no children. |
| 1307 | */ |
| 1308 | static int mem_cgroup_count_children(struct mem_cgroup *mem) |
| 1309 | { |
| 1310 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1311 | struct mem_cgroup *iter; |
| 1312 | |
| 1313 | for_each_mem_cgroup_tree(iter, mem) |
| 1314 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1315 | return num; |
| 1316 | } |
| 1317 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1318 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1319 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1320 | */ |
| 1321 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
| 1322 | { |
| 1323 | u64 limit; |
| 1324 | u64 memsw; |
| 1325 | |
Johannes Weiner | f3e8eb7 | 2011-01-13 15:47:39 -0800 | [diff] [blame] | 1326 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 1327 | limit += total_swap_pages << PAGE_SHIFT; |
| 1328 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1329 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 1330 | /* |
| 1331 | * If memsw is finite and limits the amount of swap space available |
| 1332 | * to this memcg, return that limit. |
| 1333 | */ |
| 1334 | return min(limit, memsw); |
| 1335 | } |
| 1336 | |
| 1337 | /* |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1338 | * Visit the first child (need not be the first child as per the ordering |
| 1339 | * of the cgroup list, since we track last_scanned_child) of @mem and use |
| 1340 | * that to reclaim free pages from. |
| 1341 | */ |
| 1342 | static struct mem_cgroup * |
| 1343 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) |
| 1344 | { |
| 1345 | struct mem_cgroup *ret = NULL; |
| 1346 | struct cgroup_subsys_state *css; |
| 1347 | int nextid, found; |
| 1348 | |
| 1349 | if (!root_mem->use_hierarchy) { |
| 1350 | css_get(&root_mem->css); |
| 1351 | ret = root_mem; |
| 1352 | } |
| 1353 | |
| 1354 | while (!ret) { |
| 1355 | rcu_read_lock(); |
| 1356 | nextid = root_mem->last_scanned_child + 1; |
| 1357 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, |
| 1358 | &found); |
| 1359 | if (css && css_tryget(css)) |
| 1360 | ret = container_of(css, struct mem_cgroup, css); |
| 1361 | |
| 1362 | rcu_read_unlock(); |
| 1363 | /* Updates scanning parameter */ |
| 1364 | spin_lock(&root_mem->reclaim_param_lock); |
| 1365 | if (!css) { |
| 1366 | /* this means start scan from ID:1 */ |
| 1367 | root_mem->last_scanned_child = 0; |
| 1368 | } else |
| 1369 | root_mem->last_scanned_child = found; |
| 1370 | spin_unlock(&root_mem->reclaim_param_lock); |
| 1371 | } |
| 1372 | |
| 1373 | return ret; |
| 1374 | } |
| 1375 | |
| 1376 | /* |
| 1377 | * Scan the hierarchy if needed to reclaim memory. We remember the last child |
| 1378 | * we reclaimed from, so that we don't end up penalizing one child extensively |
| 1379 | * based on its position in the children list. |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1380 | * |
| 1381 | * root_mem is the original ancestor that we've been reclaim from. |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1382 | * |
| 1383 | * We give up and return to the caller when we visit root_mem twice. |
| 1384 | * (other groups can be removed while we're walking....) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1385 | * |
| 1386 | * If shrink==true, for avoiding to free too much, this returns immedieately. |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1387 | */ |
| 1388 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1389 | struct zone *zone, |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 1390 | gfp_t gfp_mask, |
| 1391 | unsigned long reclaim_options) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1392 | { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1393 | struct mem_cgroup *victim; |
| 1394 | int ret, total = 0; |
| 1395 | int loop = 0; |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 1396 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
| 1397 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1398 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1399 | unsigned long excess; |
| 1400 | |
| 1401 | excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1402 | |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 1403 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
| 1404 | if (root_mem->memsw_is_minimum) |
| 1405 | noswap = true; |
| 1406 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1407 | while (1) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1408 | victim = mem_cgroup_select_victim(root_mem); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1409 | if (victim == root_mem) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1410 | loop++; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1411 | if (loop >= 1) |
| 1412 | drain_all_stock_async(); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1413 | if (loop >= 2) { |
| 1414 | /* |
| 1415 | * If we have not been able to reclaim |
| 1416 | * anything, it might because there are |
| 1417 | * no reclaimable pages under this hierarchy |
| 1418 | */ |
| 1419 | if (!check_soft || !total) { |
| 1420 | css_put(&victim->css); |
| 1421 | break; |
| 1422 | } |
| 1423 | /* |
| 1424 | * We want to do more targetted reclaim. |
| 1425 | * excess >> 2 is not to excessive so as to |
| 1426 | * reclaim too much, nor too less that we keep |
| 1427 | * coming back to reclaim from this cgroup |
| 1428 | */ |
| 1429 | if (total >= (excess >> 2) || |
| 1430 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { |
| 1431 | css_put(&victim->css); |
| 1432 | break; |
| 1433 | } |
| 1434 | } |
| 1435 | } |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 1436 | if (!mem_cgroup_local_usage(victim)) { |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1437 | /* this cgroup's local usage == 0 */ |
| 1438 | css_put(&victim->css); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1439 | continue; |
| 1440 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1441 | /* we use swappiness of local cgroup */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1442 | if (check_soft) |
| 1443 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
KOSAKI Motohiro | 14fec79 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 1444 | noswap, get_swappiness(victim), zone); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1445 | else |
| 1446 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
| 1447 | noswap, get_swappiness(victim)); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1448 | css_put(&victim->css); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1449 | /* |
| 1450 | * At shrinking usage, we can't check we should stop here or |
| 1451 | * reclaim more. It's depends on callers. last_scanned_child |
| 1452 | * will work enough for keeping fairness under tree. |
| 1453 | */ |
| 1454 | if (shrink) |
| 1455 | return ret; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1456 | total += ret; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1457 | if (check_soft) { |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1458 | if (!res_counter_soft_limit_excess(&root_mem->res)) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 1459 | return total; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1460 | } else if (mem_cgroup_margin(root_mem)) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1461 | return 1 + total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1462 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 1463 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1464 | } |
| 1465 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1466 | /* |
| 1467 | * Check OOM-Killer is already running under our hierarchy. |
| 1468 | * If someone is running, return false. |
| 1469 | */ |
| 1470 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) |
| 1471 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1472 | int x, lock_count = 0; |
| 1473 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1474 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1475 | for_each_mem_cgroup_tree(iter, mem) { |
| 1476 | x = atomic_inc_return(&iter->oom_lock); |
| 1477 | lock_count = max(x, lock_count); |
| 1478 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1479 | |
| 1480 | if (lock_count == 1) |
| 1481 | return true; |
| 1482 | return false; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1483 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1484 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1485 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1486 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1487 | struct mem_cgroup *iter; |
| 1488 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1489 | /* |
| 1490 | * When a new child is created while the hierarchy is under oom, |
| 1491 | * mem_cgroup_oom_lock() may not be called. We have to use |
| 1492 | * atomic_add_unless() here. |
| 1493 | */ |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1494 | for_each_mem_cgroup_tree(iter, mem) |
| 1495 | atomic_add_unless(&iter->oom_lock, -1, 0); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1496 | return 0; |
| 1497 | } |
| 1498 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1499 | |
| 1500 | static DEFINE_MUTEX(memcg_oom_mutex); |
| 1501 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1502 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1503 | struct oom_wait_info { |
| 1504 | struct mem_cgroup *mem; |
| 1505 | wait_queue_t wait; |
| 1506 | }; |
| 1507 | |
| 1508 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1509 | unsigned mode, int sync, void *arg) |
| 1510 | { |
| 1511 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; |
| 1512 | struct oom_wait_info *oom_wait_info; |
| 1513 | |
| 1514 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
| 1515 | |
| 1516 | if (oom_wait_info->mem == wake_mem) |
| 1517 | goto wakeup; |
| 1518 | /* if no hierarchy, no match */ |
| 1519 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) |
| 1520 | return 0; |
| 1521 | /* |
| 1522 | * Both of oom_wait_info->mem and wake_mem are stable under us. |
| 1523 | * Then we can use css_is_ancestor without taking care of RCU. |
| 1524 | */ |
| 1525 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && |
| 1526 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) |
| 1527 | return 0; |
| 1528 | |
| 1529 | wakeup: |
| 1530 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1531 | } |
| 1532 | |
| 1533 | static void memcg_wakeup_oom(struct mem_cgroup *mem) |
| 1534 | { |
| 1535 | /* for filtering, pass "mem" as argument. */ |
| 1536 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); |
| 1537 | } |
| 1538 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1539 | static void memcg_oom_recover(struct mem_cgroup *mem) |
| 1540 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1541 | if (mem && atomic_read(&mem->oom_lock)) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1542 | memcg_wakeup_oom(mem); |
| 1543 | } |
| 1544 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1545 | /* |
| 1546 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. |
| 1547 | */ |
| 1548 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) |
| 1549 | { |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1550 | struct oom_wait_info owait; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1551 | bool locked, need_to_kill; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1552 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1553 | owait.mem = mem; |
| 1554 | owait.wait.flags = 0; |
| 1555 | owait.wait.func = memcg_oom_wake_function; |
| 1556 | owait.wait.private = current; |
| 1557 | INIT_LIST_HEAD(&owait.wait.task_list); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1558 | need_to_kill = true; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1559 | /* At first, try to OOM lock hierarchy under mem.*/ |
| 1560 | mutex_lock(&memcg_oom_mutex); |
| 1561 | locked = mem_cgroup_oom_lock(mem); |
| 1562 | /* |
| 1563 | * Even if signal_pending(), we can't quit charge() loop without |
| 1564 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL |
| 1565 | * under OOM is always welcomed, use TASK_KILLABLE here. |
| 1566 | */ |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1567 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
| 1568 | if (!locked || mem->oom_kill_disable) |
| 1569 | need_to_kill = false; |
| 1570 | if (locked) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1571 | mem_cgroup_oom_notify(mem); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1572 | mutex_unlock(&memcg_oom_mutex); |
| 1573 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1574 | if (need_to_kill) { |
| 1575 | finish_wait(&memcg_oom_waitq, &owait.wait); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1576 | mem_cgroup_out_of_memory(mem, mask); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1577 | } else { |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1578 | schedule(); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1579 | finish_wait(&memcg_oom_waitq, &owait.wait); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1580 | } |
| 1581 | mutex_lock(&memcg_oom_mutex); |
| 1582 | mem_cgroup_oom_unlock(mem); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1583 | memcg_wakeup_oom(mem); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1584 | mutex_unlock(&memcg_oom_mutex); |
| 1585 | |
| 1586 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
| 1587 | return false; |
| 1588 | /* Give chance to dying process */ |
| 1589 | schedule_timeout(1); |
| 1590 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1591 | } |
| 1592 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1593 | /* |
| 1594 | * Currently used to update mapped file statistics, but the routine can be |
| 1595 | * generalized to update other statistics as well. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1596 | * |
| 1597 | * Notes: Race condition |
| 1598 | * |
| 1599 | * We usually use page_cgroup_lock() for accessing page_cgroup member but |
| 1600 | * it tends to be costly. But considering some conditions, we doesn't need |
| 1601 | * to do so _always_. |
| 1602 | * |
| 1603 | * Considering "charge", lock_page_cgroup() is not required because all |
| 1604 | * file-stat operations happen after a page is attached to radix-tree. There |
| 1605 | * are no race with "charge". |
| 1606 | * |
| 1607 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup |
| 1608 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even |
| 1609 | * if there are race with "uncharge". Statistics itself is properly handled |
| 1610 | * by flags. |
| 1611 | * |
| 1612 | * Considering "move", this is an only case we see a race. To make the race |
| 1613 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are |
| 1614 | * possibility of race condition. If there is, we take a lock. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1615 | */ |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1616 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1617 | void mem_cgroup_update_page_stat(struct page *page, |
| 1618 | enum mem_cgroup_page_stat_item idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1619 | { |
| 1620 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1621 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 1622 | bool need_unlock = false; |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1623 | unsigned long uninitialized_var(flags); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1624 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1625 | if (unlikely(!pc)) |
| 1626 | return; |
| 1627 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1628 | rcu_read_lock(); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1629 | mem = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1630 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
| 1631 | goto out; |
| 1632 | /* pc->mem_cgroup is unstable ? */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 1633 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1634 | /* take a lock against to access pc->mem_cgroup */ |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1635 | move_lock_page_cgroup(pc, &flags); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1636 | need_unlock = true; |
| 1637 | mem = pc->mem_cgroup; |
| 1638 | if (!mem || !PageCgroupUsed(pc)) |
| 1639 | goto out; |
| 1640 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1641 | |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1642 | switch (idx) { |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1643 | case MEMCG_NR_FILE_MAPPED: |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1644 | if (val > 0) |
| 1645 | SetPageCgroupFileMapped(pc); |
| 1646 | else if (!page_mapped(page)) |
KAMEZAWA Hiroyuki | 0c270f8 | 2010-10-27 15:33:39 -0700 | [diff] [blame] | 1647 | ClearPageCgroupFileMapped(pc); |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1648 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1649 | break; |
| 1650 | default: |
| 1651 | BUG(); |
KAMEZAWA Hiroyuki | 8725d54 | 2010-04-06 14:35:05 -0700 | [diff] [blame] | 1652 | } |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1653 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1654 | this_cpu_add(mem->stat->count[idx], val); |
| 1655 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1656 | out: |
| 1657 | if (unlikely(need_unlock)) |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 1658 | move_unlock_page_cgroup(pc, &flags); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1659 | rcu_read_unlock(); |
| 1660 | return; |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1661 | } |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1662 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 1663 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1664 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1665 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1666 | * TODO: maybe necessary to use big numbers in big irons. |
| 1667 | */ |
| 1668 | #define CHARGE_SIZE (32 * PAGE_SIZE) |
| 1669 | struct memcg_stock_pcp { |
| 1670 | struct mem_cgroup *cached; /* this never be root cgroup */ |
| 1671 | int charge; |
| 1672 | struct work_struct work; |
| 1673 | }; |
| 1674 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
| 1675 | static atomic_t memcg_drain_count; |
| 1676 | |
| 1677 | /* |
| 1678 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed |
| 1679 | * from local stock and true is returned. If the stock is 0 or charges from a |
| 1680 | * cgroup which is not current target, returns false. This stock will be |
| 1681 | * refilled. |
| 1682 | */ |
| 1683 | static bool consume_stock(struct mem_cgroup *mem) |
| 1684 | { |
| 1685 | struct memcg_stock_pcp *stock; |
| 1686 | bool ret = true; |
| 1687 | |
| 1688 | stock = &get_cpu_var(memcg_stock); |
| 1689 | if (mem == stock->cached && stock->charge) |
| 1690 | stock->charge -= PAGE_SIZE; |
| 1691 | else /* need to call res_counter_charge */ |
| 1692 | ret = false; |
| 1693 | put_cpu_var(memcg_stock); |
| 1694 | return ret; |
| 1695 | } |
| 1696 | |
| 1697 | /* |
| 1698 | * Returns stocks cached in percpu to res_counter and reset cached information. |
| 1699 | */ |
| 1700 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 1701 | { |
| 1702 | struct mem_cgroup *old = stock->cached; |
| 1703 | |
| 1704 | if (stock->charge) { |
| 1705 | res_counter_uncharge(&old->res, stock->charge); |
| 1706 | if (do_swap_account) |
| 1707 | res_counter_uncharge(&old->memsw, stock->charge); |
| 1708 | } |
| 1709 | stock->cached = NULL; |
| 1710 | stock->charge = 0; |
| 1711 | } |
| 1712 | |
| 1713 | /* |
| 1714 | * This must be called under preempt disabled or must be called by |
| 1715 | * a thread which is pinned to local cpu. |
| 1716 | */ |
| 1717 | static void drain_local_stock(struct work_struct *dummy) |
| 1718 | { |
| 1719 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); |
| 1720 | drain_stock(stock); |
| 1721 | } |
| 1722 | |
| 1723 | /* |
| 1724 | * 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] | 1725 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1726 | */ |
| 1727 | static void refill_stock(struct mem_cgroup *mem, int val) |
| 1728 | { |
| 1729 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 1730 | |
| 1731 | if (stock->cached != mem) { /* reset if necessary */ |
| 1732 | drain_stock(stock); |
| 1733 | stock->cached = mem; |
| 1734 | } |
| 1735 | stock->charge += val; |
| 1736 | put_cpu_var(memcg_stock); |
| 1737 | } |
| 1738 | |
| 1739 | /* |
| 1740 | * Tries to drain stocked charges in other cpus. This function is asynchronous |
| 1741 | * and just put a work per cpu for draining localy on each cpu. Caller can |
| 1742 | * expects some charges will be back to res_counter later but cannot wait for |
| 1743 | * it. |
| 1744 | */ |
| 1745 | static void drain_all_stock_async(void) |
| 1746 | { |
| 1747 | int cpu; |
| 1748 | /* This function is for scheduling "drain" in asynchronous way. |
| 1749 | * The result of "drain" is not directly handled by callers. Then, |
| 1750 | * if someone is calling drain, we don't have to call drain more. |
| 1751 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if |
| 1752 | * there is a race. We just do loose check here. |
| 1753 | */ |
| 1754 | if (atomic_read(&memcg_drain_count)) |
| 1755 | return; |
| 1756 | /* Notify other cpus that system-wide "drain" is running */ |
| 1757 | atomic_inc(&memcg_drain_count); |
| 1758 | get_online_cpus(); |
| 1759 | for_each_online_cpu(cpu) { |
| 1760 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
| 1761 | schedule_work_on(cpu, &stock->work); |
| 1762 | } |
| 1763 | put_online_cpus(); |
| 1764 | atomic_dec(&memcg_drain_count); |
| 1765 | /* We don't wait for flush_work */ |
| 1766 | } |
| 1767 | |
| 1768 | /* This is a synchronous drain interface. */ |
| 1769 | static void drain_all_stock_sync(void) |
| 1770 | { |
| 1771 | /* called when force_empty is called */ |
| 1772 | atomic_inc(&memcg_drain_count); |
| 1773 | schedule_on_each_cpu(drain_local_stock); |
| 1774 | atomic_dec(&memcg_drain_count); |
| 1775 | } |
| 1776 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1777 | /* |
| 1778 | * This function drains percpu counter value from DEAD cpu and |
| 1779 | * move it to local cpu. Note that this function can be preempted. |
| 1780 | */ |
| 1781 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) |
| 1782 | { |
| 1783 | int i; |
| 1784 | |
| 1785 | spin_lock(&mem->pcp_counter_lock); |
| 1786 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
| 1787 | s64 x = per_cpu(mem->stat->count[i], cpu); |
| 1788 | |
| 1789 | per_cpu(mem->stat->count[i], cpu) = 0; |
| 1790 | mem->nocpu_base.count[i] += x; |
| 1791 | } |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1792 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
| 1793 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
| 1794 | spin_unlock(&mem->pcp_counter_lock); |
| 1795 | } |
| 1796 | |
| 1797 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) |
| 1798 | { |
| 1799 | int idx = MEM_CGROUP_ON_MOVE; |
| 1800 | |
| 1801 | spin_lock(&mem->pcp_counter_lock); |
| 1802 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1803 | spin_unlock(&mem->pcp_counter_lock); |
| 1804 | } |
| 1805 | |
| 1806 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1807 | unsigned long action, |
| 1808 | void *hcpu) |
| 1809 | { |
| 1810 | int cpu = (unsigned long)hcpu; |
| 1811 | struct memcg_stock_pcp *stock; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1812 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1813 | |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1814 | if ((action == CPU_ONLINE)) { |
| 1815 | for_each_mem_cgroup_all(iter) |
| 1816 | synchronize_mem_cgroup_on_move(iter, cpu); |
| 1817 | return NOTIFY_OK; |
| 1818 | } |
| 1819 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1820 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1821 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1822 | |
| 1823 | for_each_mem_cgroup_all(iter) |
| 1824 | mem_cgroup_drain_pcp_counter(iter, cpu); |
| 1825 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1826 | stock = &per_cpu(memcg_stock, cpu); |
| 1827 | drain_stock(stock); |
| 1828 | return NOTIFY_OK; |
| 1829 | } |
| 1830 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1831 | |
| 1832 | /* See __mem_cgroup_try_charge() for details */ |
| 1833 | enum { |
| 1834 | CHARGE_OK, /* success */ |
| 1835 | CHARGE_RETRY, /* need to retry but retry is not bad */ |
| 1836 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ |
| 1837 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ |
| 1838 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ |
| 1839 | }; |
| 1840 | |
| 1841 | static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, |
| 1842 | int csize, bool oom_check) |
| 1843 | { |
| 1844 | struct mem_cgroup *mem_over_limit; |
| 1845 | struct res_counter *fail_res; |
| 1846 | unsigned long flags = 0; |
| 1847 | int ret; |
| 1848 | |
| 1849 | ret = res_counter_charge(&mem->res, csize, &fail_res); |
| 1850 | |
| 1851 | if (likely(!ret)) { |
| 1852 | if (!do_swap_account) |
| 1853 | return CHARGE_OK; |
| 1854 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
| 1855 | if (likely(!ret)) |
| 1856 | return CHARGE_OK; |
| 1857 | |
KAMEZAWA Hiroyuki | 01c88e2 | 2011-01-25 15:07:27 -0800 | [diff] [blame] | 1858 | res_counter_uncharge(&mem->res, csize); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1859 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
| 1860 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
| 1861 | } else |
| 1862 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); |
Johannes Weiner | 9221edb | 2011-02-01 15:52:42 -0800 | [diff] [blame] | 1863 | /* |
| 1864 | * csize can be either a huge page (HPAGE_SIZE), a batch of |
| 1865 | * regular pages (CHARGE_SIZE), or a single regular page |
| 1866 | * (PAGE_SIZE). |
| 1867 | * |
| 1868 | * Never reclaim on behalf of optional batching, retry with a |
| 1869 | * single page instead. |
| 1870 | */ |
| 1871 | if (csize == CHARGE_SIZE) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1872 | return CHARGE_RETRY; |
| 1873 | |
| 1874 | if (!(gfp_mask & __GFP_WAIT)) |
| 1875 | return CHARGE_WOULDBLOCK; |
| 1876 | |
| 1877 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1878 | gfp_mask, flags); |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1879 | if (mem_cgroup_margin(mem_over_limit) >= csize) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1880 | return CHARGE_RETRY; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1881 | /* |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1882 | * Even though the limit is exceeded at this point, reclaim |
| 1883 | * may have been able to free some pages. Retry the charge |
| 1884 | * before killing the task. |
| 1885 | * |
| 1886 | * Only for regular pages, though: huge pages are rather |
| 1887 | * unlikely to succeed so close to the limit, and we fall back |
| 1888 | * to regular pages anyway in case of failure. |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1889 | */ |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1890 | if (csize == PAGE_SIZE && ret) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1891 | return CHARGE_RETRY; |
| 1892 | |
| 1893 | /* |
| 1894 | * At task move, charge accounts can be doubly counted. So, it's |
| 1895 | * better to wait until the end of task_move if something is going on. |
| 1896 | */ |
| 1897 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 1898 | return CHARGE_RETRY; |
| 1899 | |
| 1900 | /* If we don't need to call oom-killer at el, return immediately */ |
| 1901 | if (!oom_check) |
| 1902 | return CHARGE_NOMEM; |
| 1903 | /* check OOM */ |
| 1904 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) |
| 1905 | return CHARGE_OOM_DIE; |
| 1906 | |
| 1907 | return CHARGE_RETRY; |
| 1908 | } |
| 1909 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1910 | /* |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1911 | * Unlike exported interface, "oom" parameter is added. if oom==true, |
| 1912 | * oom-killer can be invoked. |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1913 | */ |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1914 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1915 | gfp_t gfp_mask, |
| 1916 | struct mem_cgroup **memcg, bool oom, |
| 1917 | int page_size) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1918 | { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1919 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 1920 | struct mem_cgroup *mem = NULL; |
| 1921 | int ret; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1922 | int csize = max(CHARGE_SIZE, (unsigned long) page_size); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1923 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1924 | /* |
| 1925 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage |
| 1926 | * in system level. So, allow to go ahead dying process in addition to |
| 1927 | * MEMDIE process. |
| 1928 | */ |
| 1929 | if (unlikely(test_thread_flag(TIF_MEMDIE) |
| 1930 | || fatal_signal_pending(current))) |
| 1931 | goto bypass; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1932 | |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1933 | /* |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 1934 | * We always charge the cgroup the mm_struct belongs to. |
| 1935 | * The mm_struct's mem_cgroup changes on task migration if the |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1936 | * thread group leader migrates. It's possible that mm is not |
| 1937 | * set, if so charge the init_mm (happens for pagecache usage). |
| 1938 | */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1939 | if (!*memcg && !mm) |
| 1940 | goto bypass; |
| 1941 | again: |
| 1942 | if (*memcg) { /* css should be a valid one */ |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1943 | mem = *memcg; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1944 | VM_BUG_ON(css_is_removed(&mem->css)); |
| 1945 | if (mem_cgroup_is_root(mem)) |
| 1946 | goto done; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1947 | if (page_size == PAGE_SIZE && consume_stock(mem)) |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1948 | goto done; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1949 | css_get(&mem->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1950 | } else { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1951 | struct task_struct *p; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1952 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1953 | rcu_read_lock(); |
| 1954 | p = rcu_dereference(mm->owner); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1955 | /* |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 1956 | * Because we don't have task_lock(), "p" can exit. |
| 1957 | * In that case, "mem" can point to root or p can be NULL with |
| 1958 | * race with swapoff. Then, we have small risk of mis-accouning. |
| 1959 | * But such kind of mis-account by race always happens because |
| 1960 | * we don't have cgroup_mutex(). It's overkill and we allo that |
| 1961 | * small race, here. |
| 1962 | * (*) swapoff at el will charge against mm-struct not against |
| 1963 | * task-struct. So, mm->owner can be NULL. |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1964 | */ |
| 1965 | mem = mem_cgroup_from_task(p); |
KAMEZAWA Hiroyuki | ebb76ce | 2010-12-29 14:07:11 -0800 | [diff] [blame] | 1966 | if (!mem || mem_cgroup_is_root(mem)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1967 | rcu_read_unlock(); |
| 1968 | goto done; |
| 1969 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 1970 | if (page_size == PAGE_SIZE && consume_stock(mem)) { |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1971 | /* |
| 1972 | * It seems dagerous to access memcg without css_get(). |
| 1973 | * But considering how consume_stok works, it's not |
| 1974 | * necessary. If consume_stock success, some charges |
| 1975 | * from this memcg are cached on this cpu. So, we |
| 1976 | * don't need to call css_get()/css_tryget() before |
| 1977 | * calling consume_stock(). |
| 1978 | */ |
| 1979 | rcu_read_unlock(); |
| 1980 | goto done; |
| 1981 | } |
| 1982 | /* after here, we may be blocked. we need to get refcnt */ |
| 1983 | if (!css_tryget(&mem->css)) { |
| 1984 | rcu_read_unlock(); |
| 1985 | goto again; |
| 1986 | } |
| 1987 | rcu_read_unlock(); |
| 1988 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1989 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1990 | do { |
| 1991 | bool oom_check; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1992 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1993 | /* If killed, bypass charge */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1994 | if (fatal_signal_pending(current)) { |
| 1995 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1996 | goto bypass; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 1997 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1998 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1999 | oom_check = false; |
| 2000 | if (oom && !nr_oom_retries) { |
| 2001 | oom_check = true; |
| 2002 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 2003 | } |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2004 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2005 | ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check); |
| 2006 | |
| 2007 | switch (ret) { |
| 2008 | case CHARGE_OK: |
| 2009 | break; |
| 2010 | case CHARGE_RETRY: /* not in OOM situation but retry */ |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2011 | csize = page_size; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2012 | css_put(&mem->css); |
| 2013 | mem = NULL; |
| 2014 | goto again; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2015 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2016 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2017 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2018 | case CHARGE_NOMEM: /* OOM routine works */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2019 | if (!oom) { |
| 2020 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2021 | goto nomem; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2022 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2023 | /* If oom, we never return -ENOMEM */ |
| 2024 | nr_oom_retries--; |
| 2025 | break; |
| 2026 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2027 | css_put(&mem->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2028 | goto bypass; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2029 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2030 | } while (ret != CHARGE_OK); |
| 2031 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2032 | if (csize > page_size) |
| 2033 | refill_stock(mem, csize - page_size); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2034 | css_put(&mem->css); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2035 | done: |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2036 | *memcg = mem; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2037 | return 0; |
| 2038 | nomem: |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2039 | *memcg = NULL; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2040 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2041 | bypass: |
| 2042 | *memcg = NULL; |
| 2043 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2044 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2045 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2046 | /* |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2047 | * Somemtimes we have to undo a charge we got by try_charge(). |
| 2048 | * This function is for that and do uncharge, put css's refcnt. |
| 2049 | * gotten by try_charge(). |
| 2050 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2051 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
| 2052 | unsigned long count) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2053 | { |
| 2054 | if (!mem_cgroup_is_root(mem)) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2055 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2056 | if (do_swap_account) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2057 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2058 | } |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2059 | } |
| 2060 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2061 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
| 2062 | int page_size) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2063 | { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2064 | __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2065 | } |
| 2066 | |
| 2067 | /* |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2068 | * A helper function to get mem_cgroup from ID. must be called under |
| 2069 | * rcu_read_lock(). The caller must check css_is_removed() or some if |
| 2070 | * it's concern. (dropping refcnt from swap can be called against removed |
| 2071 | * memcg.) |
| 2072 | */ |
| 2073 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
| 2074 | { |
| 2075 | struct cgroup_subsys_state *css; |
| 2076 | |
| 2077 | /* ID 0 is unused ID */ |
| 2078 | if (!id) |
| 2079 | return NULL; |
| 2080 | css = css_lookup(&mem_cgroup_subsys, id); |
| 2081 | if (!css) |
| 2082 | return NULL; |
| 2083 | return container_of(css, struct mem_cgroup, css); |
| 2084 | } |
| 2085 | |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2086 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2087 | { |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2088 | struct mem_cgroup *mem = NULL; |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2089 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2090 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2091 | swp_entry_t ent; |
| 2092 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2093 | VM_BUG_ON(!PageLocked(page)); |
| 2094 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2095 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2096 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2097 | if (PageCgroupUsed(pc)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2098 | mem = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2099 | if (mem && !css_tryget(&mem->css)) |
| 2100 | mem = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2101 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2102 | ent.val = page_private(page); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2103 | id = lookup_swap_cgroup(ent); |
| 2104 | rcu_read_lock(); |
| 2105 | mem = mem_cgroup_lookup(id); |
| 2106 | if (mem && !css_tryget(&mem->css)) |
| 2107 | mem = NULL; |
| 2108 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2109 | } |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2110 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2111 | return mem; |
| 2112 | } |
| 2113 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2114 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
| 2115 | struct page_cgroup *pc, |
| 2116 | enum charge_type ctype, |
| 2117 | int page_size) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2118 | { |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2119 | int nr_pages = page_size >> PAGE_SHIFT; |
| 2120 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2121 | lock_page_cgroup(pc); |
| 2122 | if (unlikely(PageCgroupUsed(pc))) { |
| 2123 | unlock_page_cgroup(pc); |
| 2124 | mem_cgroup_cancel_charge(mem, page_size); |
| 2125 | return; |
| 2126 | } |
| 2127 | /* |
| 2128 | * we don't need page_cgroup_lock about tail pages, becase they are not |
| 2129 | * accessed by any other context at this point. |
| 2130 | */ |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2131 | pc->mem_cgroup = mem; |
KAMEZAWA Hiroyuki | 261fb61 | 2009-09-23 15:56:33 -0700 | [diff] [blame] | 2132 | /* |
| 2133 | * We access a page_cgroup asynchronously without lock_page_cgroup(). |
| 2134 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup |
| 2135 | * is accessed after testing USED bit. To make pc->mem_cgroup visible |
| 2136 | * before USED bit, we need memory barrier here. |
| 2137 | * See mem_cgroup_add_lru_list(), etc. |
| 2138 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2139 | smp_wmb(); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 2140 | switch (ctype) { |
| 2141 | case MEM_CGROUP_CHARGE_TYPE_CACHE: |
| 2142 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: |
| 2143 | SetPageCgroupCache(pc); |
| 2144 | SetPageCgroupUsed(pc); |
| 2145 | break; |
| 2146 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
| 2147 | ClearPageCgroupCache(pc); |
| 2148 | SetPageCgroupUsed(pc); |
| 2149 | break; |
| 2150 | default: |
| 2151 | break; |
| 2152 | } |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2153 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2154 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2155 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 430e4863 | 2010-03-10 15:22:30 -0800 | [diff] [blame] | 2156 | /* |
| 2157 | * "charge_statistics" updated event counter. Then, check it. |
| 2158 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
| 2159 | * if they exceeds softlimit. |
| 2160 | */ |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2161 | memcg_check_events(mem, pc->page); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2162 | } |
| 2163 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2164 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2165 | |
| 2166 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ |
| 2167 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) |
| 2168 | /* |
| 2169 | * Because tail pages are not marked as "used", set it. We're under |
| 2170 | * zone->lru_lock, 'splitting on pmd' and compund_lock. |
| 2171 | */ |
| 2172 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) |
| 2173 | { |
| 2174 | struct page_cgroup *head_pc = lookup_page_cgroup(head); |
| 2175 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); |
| 2176 | unsigned long flags; |
| 2177 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2178 | if (mem_cgroup_disabled()) |
| 2179 | return; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2180 | /* |
KAMEZAWA Hiroyuki | ece35ca | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2181 | * We have no races with charge/uncharge but will have races with |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2182 | * page state accounting. |
| 2183 | */ |
| 2184 | move_lock_page_cgroup(head_pc, &flags); |
| 2185 | |
| 2186 | tail_pc->mem_cgroup = head_pc->mem_cgroup; |
| 2187 | smp_wmb(); /* see __commit_charge() */ |
KAMEZAWA Hiroyuki | ece35ca | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2188 | if (PageCgroupAcctLRU(head_pc)) { |
| 2189 | enum lru_list lru; |
| 2190 | struct mem_cgroup_per_zone *mz; |
| 2191 | |
| 2192 | /* |
| 2193 | * LRU flags cannot be copied because we need to add tail |
| 2194 | *.page to LRU by generic call and our hook will be called. |
| 2195 | * We hold lru_lock, then, reduce counter directly. |
| 2196 | */ |
| 2197 | lru = page_lru(head); |
| 2198 | mz = page_cgroup_zoneinfo(head_pc); |
| 2199 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
| 2200 | } |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2201 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
| 2202 | move_unlock_page_cgroup(head_pc, &flags); |
| 2203 | } |
| 2204 | #endif |
| 2205 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2206 | /** |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2207 | * __mem_cgroup_move_account - move account of the page |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2208 | * @pc: page_cgroup of the page. |
| 2209 | * @from: mem_cgroup which the page is moved from. |
| 2210 | * @to: mem_cgroup which the page is moved to. @from != @to. |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2211 | * @uncharge: whether we should call uncharge and css_put against @from. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2212 | * |
| 2213 | * The caller must confirm following. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2214 | * - page is not on LRU (isolate_page() is useful.) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2215 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2216 | * |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2217 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
| 2218 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is |
| 2219 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
| 2220 | * @uncharge is false, so a caller should do "uncharge". |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2221 | */ |
| 2222 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2223 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2224 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge, |
| 2225 | int charge_size) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2226 | { |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2227 | int nr_pages = charge_size >> PAGE_SHIFT; |
| 2228 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2229 | VM_BUG_ON(from == to); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2230 | VM_BUG_ON(PageLRU(pc->page)); |
Kirill A. Shutemov | 112bc2e | 2010-11-24 12:56:58 -0800 | [diff] [blame] | 2231 | VM_BUG_ON(!page_is_cgroup_locked(pc)); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2232 | VM_BUG_ON(!PageCgroupUsed(pc)); |
| 2233 | VM_BUG_ON(pc->mem_cgroup != from); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2234 | |
KAMEZAWA Hiroyuki | 8725d54 | 2010-04-06 14:35:05 -0700 | [diff] [blame] | 2235 | if (PageCgroupFileMapped(pc)) { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 2236 | /* Update mapped_file data for mem_cgroup */ |
| 2237 | preempt_disable(); |
| 2238 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2239 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
| 2240 | preempt_enable(); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2241 | } |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2242 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2243 | if (uncharge) |
| 2244 | /* This is not "cancel", but cancel_charge does all we need. */ |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2245 | mem_cgroup_cancel_charge(from, charge_size); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2246 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2247 | /* caller should have done css_get */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2248 | pc->mem_cgroup = to; |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2249 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2250 | /* |
| 2251 | * We charges against "to" which may not have any tasks. Then, "to" |
| 2252 | * can be under rmdir(). But in current implementation, caller of |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 2253 | * this function is just force_empty() and move charge, so it's |
| 2254 | * garanteed that "to" is never removed. So, we don't check rmdir |
| 2255 | * status here. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2256 | */ |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2257 | } |
| 2258 | |
| 2259 | /* |
| 2260 | * check whether the @pc is valid for moving account and call |
| 2261 | * __mem_cgroup_move_account() |
| 2262 | */ |
| 2263 | static int mem_cgroup_move_account(struct page_cgroup *pc, |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2264 | struct mem_cgroup *from, struct mem_cgroup *to, |
| 2265 | bool uncharge, int charge_size) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2266 | { |
| 2267 | int ret = -EINVAL; |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2268 | unsigned long flags; |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2269 | /* |
| 2270 | * The page is isolated from LRU. So, collapse function |
| 2271 | * will not handle this page. But page splitting can happen. |
| 2272 | * Do this check under compound_page_lock(). The caller should |
| 2273 | * hold it. |
| 2274 | */ |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2275 | if ((charge_size > PAGE_SIZE) && !PageTransHuge(pc->page)) |
| 2276 | return -EBUSY; |
| 2277 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2278 | lock_page_cgroup(pc); |
| 2279 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2280 | move_lock_page_cgroup(pc, &flags); |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2281 | __mem_cgroup_move_account(pc, from, to, uncharge, charge_size); |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2282 | move_unlock_page_cgroup(pc, &flags); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2283 | ret = 0; |
| 2284 | } |
| 2285 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2286 | /* |
| 2287 | * check events |
| 2288 | */ |
| 2289 | memcg_check_events(to, pc->page); |
| 2290 | memcg_check_events(from, pc->page); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2291 | return ret; |
| 2292 | } |
| 2293 | |
| 2294 | /* |
| 2295 | * move charges to its parent. |
| 2296 | */ |
| 2297 | |
| 2298 | static int mem_cgroup_move_parent(struct page_cgroup *pc, |
| 2299 | struct mem_cgroup *child, |
| 2300 | gfp_t gfp_mask) |
| 2301 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2302 | struct page *page = pc->page; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2303 | struct cgroup *cg = child->css.cgroup; |
| 2304 | struct cgroup *pcg = cg->parent; |
| 2305 | struct mem_cgroup *parent; |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2306 | int page_size = PAGE_SIZE; |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2307 | unsigned long flags; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2308 | int ret; |
| 2309 | |
| 2310 | /* Is ROOT ? */ |
| 2311 | if (!pcg) |
| 2312 | return -EINVAL; |
| 2313 | |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2314 | ret = -EBUSY; |
| 2315 | if (!get_page_unless_zero(page)) |
| 2316 | goto out; |
| 2317 | if (isolate_lru_page(page)) |
| 2318 | goto put; |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2319 | |
| 2320 | if (PageTransHuge(page)) |
| 2321 | page_size = HPAGE_SIZE; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2322 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2323 | parent = mem_cgroup_from_cont(pcg); |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2324 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, |
| 2325 | &parent, false, page_size); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2326 | if (ret || !parent) |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2327 | goto put_back; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2328 | |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2329 | if (page_size > PAGE_SIZE) |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2330 | flags = compound_lock_irqsave(page); |
| 2331 | |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2332 | ret = mem_cgroup_move_account(pc, child, parent, true, page_size); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2333 | if (ret) |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2334 | mem_cgroup_cancel_charge(parent, page_size); |
Jesper Juhl | 8dba474 | 2011-01-25 15:07:24 -0800 | [diff] [blame] | 2335 | |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 2336 | if (page_size > PAGE_SIZE) |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 2337 | compound_unlock_irqrestore(page, flags); |
Jesper Juhl | 8dba474 | 2011-01-25 15:07:24 -0800 | [diff] [blame] | 2338 | put_back: |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2339 | putback_lru_page(page); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2340 | put: |
Daisuke Nishimura | 40d5813 | 2009-01-15 13:51:12 -0800 | [diff] [blame] | 2341 | put_page(page); |
Daisuke Nishimura | 57f9fd7 | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 2342 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2343 | return ret; |
| 2344 | } |
| 2345 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2346 | /* |
| 2347 | * Charge the memory controller for page usage. |
| 2348 | * Return |
| 2349 | * 0 if the charge was successful |
| 2350 | * < 0 if the cgroup is over its limit |
| 2351 | */ |
| 2352 | 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] | 2353 | gfp_t gfp_mask, enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2354 | { |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2355 | struct mem_cgroup *mem = NULL; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2356 | int page_size = PAGE_SIZE; |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 2357 | struct page_cgroup *pc; |
| 2358 | bool oom = true; |
| 2359 | int ret; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2360 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2361 | if (PageTransHuge(page)) { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2362 | page_size <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2363 | VM_BUG_ON(!PageTransHuge(page)); |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 2364 | /* |
| 2365 | * Never OOM-kill a process for a huge page. The |
| 2366 | * fault handler will fall back to regular pages. |
| 2367 | */ |
| 2368 | oom = false; |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2369 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2370 | |
| 2371 | pc = lookup_page_cgroup(page); |
Johannes Weiner | af4a662 | 2011-03-23 16:42:24 -0700 | [diff] [blame] | 2372 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2373 | |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 2374 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, oom, page_size); |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2375 | if (ret || !mem) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2376 | return ret; |
| 2377 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2378 | __mem_cgroup_commit_charge(mem, pc, ctype, page_size); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2379 | return 0; |
| 2380 | } |
| 2381 | |
| 2382 | int mem_cgroup_newpage_charge(struct page *page, |
| 2383 | struct mm_struct *mm, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2384 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2385 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2386 | return 0; |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2387 | /* |
| 2388 | * If already mapped, we don't have to account. |
| 2389 | * If page cache, page->mapping has address_space. |
| 2390 | * But page->mapping may have out-of-use anon_vma pointer, |
| 2391 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping |
| 2392 | * is NULL. |
| 2393 | */ |
| 2394 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) |
| 2395 | return 0; |
| 2396 | if (unlikely(!mm)) |
| 2397 | mm = &init_mm; |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2398 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2399 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 2400 | } |
| 2401 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2402 | static void |
| 2403 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
| 2404 | enum charge_type ctype); |
| 2405 | |
Balbir Singh | e1a1cd5 | 2008-02-07 00:14:02 -0800 | [diff] [blame] | 2406 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
| 2407 | gfp_t gfp_mask) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2408 | { |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2409 | int ret; |
| 2410 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2411 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 2412 | return 0; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2413 | if (PageCompound(page)) |
| 2414 | return 0; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2415 | /* |
| 2416 | * Corner case handling. This is called from add_to_page_cache() |
| 2417 | * in usual. But some FS (shmem) precharges this page before calling it |
| 2418 | * and call add_to_page_cache() with GFP_NOWAIT. |
| 2419 | * |
| 2420 | * For GFP_NOWAIT case, the page may be pre-charged before calling |
| 2421 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call |
| 2422 | * charge twice. (It works but has to pay a bit larger cost.) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2423 | * And when the page is SwapCache, it should take swap information |
| 2424 | * into account. This is under lock_page() now. |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2425 | */ |
| 2426 | if (!(gfp_mask & __GFP_WAIT)) { |
| 2427 | struct page_cgroup *pc; |
| 2428 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2429 | pc = lookup_page_cgroup(page); |
| 2430 | if (!pc) |
| 2431 | return 0; |
| 2432 | lock_page_cgroup(pc); |
| 2433 | if (PageCgroupUsed(pc)) { |
| 2434 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2435 | return 0; |
| 2436 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2437 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2438 | } |
| 2439 | |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2440 | if (unlikely(!mm)) |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2441 | mm = &init_mm; |
KAMEZAWA Hiroyuki | accf163 | 2008-07-25 01:47:17 -0700 | [diff] [blame] | 2442 | |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 2443 | if (page_is_file_cache(page)) |
| 2444 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2445 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2446 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2447 | /* shmem */ |
| 2448 | if (PageSwapCache(page)) { |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 2449 | struct mem_cgroup *mem; |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2450 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2451 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
| 2452 | if (!ret) |
| 2453 | __mem_cgroup_commit_charge_swapin(page, mem, |
| 2454 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
| 2455 | } else |
| 2456 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
Daisuke Nishimura | 73045c4 | 2010-08-10 18:02:59 -0700 | [diff] [blame] | 2457 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2458 | |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2459 | return ret; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2460 | } |
| 2461 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2462 | /* |
| 2463 | * While swap-in, try_charge -> commit or cancel, the page is locked. |
| 2464 | * And when try_charge() successfully returns, one refcnt to memcg without |
Uwe Kleine-König | 21ae295 | 2009-10-07 15:21:09 +0200 | [diff] [blame] | 2465 | * struct page_cgroup is acquired. This refcnt will be consumed by |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2466 | * "commit()" or removed by "cancel()" |
| 2467 | */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2468 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
| 2469 | struct page *page, |
| 2470 | gfp_t mask, struct mem_cgroup **ptr) |
| 2471 | { |
| 2472 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2473 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2474 | |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 2475 | *ptr = NULL; |
| 2476 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2477 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2478 | return 0; |
| 2479 | |
| 2480 | if (!do_swap_account) |
| 2481 | goto charge_cur_mm; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2482 | /* |
| 2483 | * A racing thread's fault, or swapoff, may have already updated |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2484 | * the pte, and even removed page from swap cache: in those cases |
| 2485 | * do_swap_page()'s pte_same() test will fail; but there's also a |
| 2486 | * KSM case which does need to charge the page. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2487 | */ |
| 2488 | if (!PageSwapCache(page)) |
Hugh Dickins | 407f9c8 | 2009-12-14 17:59:30 -0800 | [diff] [blame] | 2489 | goto charge_cur_mm; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2490 | mem = try_get_mem_cgroup_from_page(page); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2491 | if (!mem) |
| 2492 | goto charge_cur_mm; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2493 | *ptr = mem; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2494 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 2495 | css_put(&mem->css); |
| 2496 | return ret; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2497 | charge_cur_mm: |
| 2498 | if (unlikely(!mm)) |
| 2499 | mm = &init_mm; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2500 | return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2501 | } |
| 2502 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2503 | static void |
| 2504 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
| 2505 | enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2506 | { |
| 2507 | struct page_cgroup *pc; |
| 2508 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2509 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2510 | return; |
| 2511 | if (!ptr) |
| 2512 | return; |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2513 | cgroup_exclude_rmdir(&ptr->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2514 | pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2515 | mem_cgroup_lru_del_before_commit_swapcache(page); |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2516 | __mem_cgroup_commit_charge(ptr, pc, ctype, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2517 | mem_cgroup_lru_add_after_commit_swapcache(page); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2518 | /* |
| 2519 | * Now swap is on-memory. This means this page may be |
| 2520 | * counted both as mem and swap....double count. |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2521 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
| 2522 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() |
| 2523 | * may call delete_from_swap_cache() before reach here. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2524 | */ |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 2525 | if (do_swap_account && PageSwapCache(page)) { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2526 | swp_entry_t ent = {.val = page_private(page)}; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2527 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2528 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2529 | |
| 2530 | id = swap_cgroup_record(ent, 0); |
| 2531 | rcu_read_lock(); |
| 2532 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2533 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2534 | /* |
| 2535 | * This recorded memcg can be obsolete one. So, avoid |
| 2536 | * calling css_tryget |
| 2537 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2538 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 2539 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2540 | mem_cgroup_swap_statistics(memcg, false); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2541 | mem_cgroup_put(memcg); |
| 2542 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2543 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2544 | } |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2545 | /* |
| 2546 | * At swapin, we may charge account against cgroup which has no tasks. |
| 2547 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 2548 | * In that case, we need to call pre_destroy() again. check it here. |
| 2549 | */ |
| 2550 | cgroup_release_and_wakeup_rmdir(&ptr->css); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2551 | } |
| 2552 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 2553 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
| 2554 | { |
| 2555 | __mem_cgroup_commit_charge_swapin(page, ptr, |
| 2556 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 2557 | } |
| 2558 | |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2559 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
| 2560 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2561 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2562 | return; |
| 2563 | if (!mem) |
| 2564 | return; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2565 | mem_cgroup_cancel_charge(mem, PAGE_SIZE); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2566 | } |
| 2567 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2568 | static void |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2569 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype, |
| 2570 | int page_size) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2571 | { |
| 2572 | struct memcg_batch_info *batch = NULL; |
| 2573 | bool uncharge_memsw = true; |
| 2574 | /* If swapout, usage of swap doesn't decrease */ |
| 2575 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
| 2576 | uncharge_memsw = false; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2577 | |
| 2578 | batch = ¤t->memcg_batch; |
| 2579 | /* |
| 2580 | * In usual, we do css_get() when we remember memcg pointer. |
| 2581 | * But in this case, we keep res->usage until end of a series of |
| 2582 | * uncharges. Then, it's ok to ignore memcg's refcnt. |
| 2583 | */ |
| 2584 | if (!batch->memcg) |
| 2585 | batch->memcg = mem; |
| 2586 | /* |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2587 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. |
| 2588 | * In those cases, all pages freed continously can be expected to be in |
| 2589 | * the same cgroup and we have chance to coalesce uncharges. |
| 2590 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) |
| 2591 | * because we want to do uncharge as soon as possible. |
| 2592 | */ |
| 2593 | |
| 2594 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) |
| 2595 | goto direct_uncharge; |
| 2596 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2597 | if (page_size != PAGE_SIZE) |
| 2598 | goto direct_uncharge; |
| 2599 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2600 | /* |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2601 | * In typical case, batch->memcg == mem. This means we can |
| 2602 | * merge a series of uncharges to an uncharge of res_counter. |
| 2603 | * If not, we uncharge res_counter ony by one. |
| 2604 | */ |
| 2605 | if (batch->memcg != mem) |
| 2606 | goto direct_uncharge; |
| 2607 | /* remember freed charge and uncharge it later */ |
| 2608 | batch->bytes += PAGE_SIZE; |
| 2609 | if (uncharge_memsw) |
| 2610 | batch->memsw_bytes += PAGE_SIZE; |
| 2611 | return; |
| 2612 | direct_uncharge: |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2613 | res_counter_uncharge(&mem->res, page_size); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2614 | if (uncharge_memsw) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2615 | res_counter_uncharge(&mem->memsw, page_size); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2616 | if (unlikely(batch->memcg != mem)) |
| 2617 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2618 | return; |
| 2619 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2620 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2621 | /* |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2622 | * uncharge if !page_mapped(page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2623 | */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2624 | static struct mem_cgroup * |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2625 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2626 | { |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2627 | int count; |
Hugh Dickins | 8289546 | 2008-03-04 14:29:08 -0800 | [diff] [blame] | 2628 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2629 | struct mem_cgroup *mem = NULL; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2630 | int page_size = PAGE_SIZE; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2631 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2632 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2633 | return NULL; |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 2634 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2635 | if (PageSwapCache(page)) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2636 | return NULL; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2637 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2638 | if (PageTransHuge(page)) { |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2639 | page_size <<= compound_order(page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 2640 | VM_BUG_ON(!PageTransHuge(page)); |
| 2641 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2642 | |
Daisuke Nishimura | 152c9cc | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2643 | count = page_size >> PAGE_SHIFT; |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2644 | /* |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 2645 | * Check if our page_cgroup is valid |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 2646 | */ |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2647 | pc = lookup_page_cgroup(page); |
| 2648 | if (unlikely(!pc || !PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2649 | return NULL; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2650 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2651 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2652 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2653 | mem = pc->mem_cgroup; |
| 2654 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2655 | if (!PageCgroupUsed(pc)) |
| 2656 | goto unlock_out; |
| 2657 | |
| 2658 | switch (ctype) { |
| 2659 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2660 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2661 | /* See mem_cgroup_prepare_migration() */ |
| 2662 | if (page_mapped(page) || PageCgroupMigration(pc)) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2663 | goto unlock_out; |
| 2664 | break; |
| 2665 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: |
| 2666 | if (!PageAnon(page)) { /* Shared memory */ |
| 2667 | if (page->mapping && !page_is_file_cache(page)) |
| 2668 | goto unlock_out; |
| 2669 | } else if (page_mapped(page)) /* Anon */ |
| 2670 | goto unlock_out; |
| 2671 | break; |
| 2672 | default: |
| 2673 | break; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2674 | } |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2675 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2676 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 2677 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2678 | ClearPageCgroupUsed(pc); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 2679 | /* |
| 2680 | * pc->mem_cgroup is not cleared here. It will be accessed when it's |
| 2681 | * freed from LRU. This is safe because uncharged page is expected not |
| 2682 | * to be reused (freed soon). Exception is SwapCache, it's handled by |
| 2683 | * special functions. |
| 2684 | */ |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 2685 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2686 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2687 | /* |
| 2688 | * even after unlock, we have mem->res.usage here and this memcg |
| 2689 | * will never be freed. |
| 2690 | */ |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 2691 | memcg_check_events(mem, page); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2692 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
| 2693 | mem_cgroup_swap_statistics(mem, true); |
| 2694 | mem_cgroup_get(mem); |
| 2695 | } |
| 2696 | if (!mem_cgroup_is_root(mem)) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2697 | __do_uncharge(mem, ctype, page_size); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 2698 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2699 | return mem; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2700 | |
| 2701 | unlock_out: |
| 2702 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2703 | return NULL; |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 2704 | } |
| 2705 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2706 | void mem_cgroup_uncharge_page(struct page *page) |
| 2707 | { |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2708 | /* early check. */ |
| 2709 | if (page_mapped(page)) |
| 2710 | return; |
| 2711 | if (page->mapping && !PageAnon(page)) |
| 2712 | return; |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2713 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 2714 | } |
| 2715 | |
| 2716 | void mem_cgroup_uncharge_cache_page(struct page *page) |
| 2717 | { |
| 2718 | VM_BUG_ON(page_mapped(page)); |
KAMEZAWA Hiroyuki | b7abea9 | 2008-10-18 20:28:09 -0700 | [diff] [blame] | 2719 | VM_BUG_ON(page->mapping); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2720 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
| 2721 | } |
| 2722 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2723 | /* |
| 2724 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. |
| 2725 | * In that cases, pages are freed continuously and we can expect pages |
| 2726 | * are in the same memcg. All these calls itself limits the number of |
| 2727 | * pages freed at once, then uncharge_start/end() is called properly. |
| 2728 | * This may be called prural(2) times in a context, |
| 2729 | */ |
| 2730 | |
| 2731 | void mem_cgroup_uncharge_start(void) |
| 2732 | { |
| 2733 | current->memcg_batch.do_batch++; |
| 2734 | /* We can do nest. */ |
| 2735 | if (current->memcg_batch.do_batch == 1) { |
| 2736 | current->memcg_batch.memcg = NULL; |
| 2737 | current->memcg_batch.bytes = 0; |
| 2738 | current->memcg_batch.memsw_bytes = 0; |
| 2739 | } |
| 2740 | } |
| 2741 | |
| 2742 | void mem_cgroup_uncharge_end(void) |
| 2743 | { |
| 2744 | struct memcg_batch_info *batch = ¤t->memcg_batch; |
| 2745 | |
| 2746 | if (!batch->do_batch) |
| 2747 | return; |
| 2748 | |
| 2749 | batch->do_batch--; |
| 2750 | if (batch->do_batch) /* If stacked, do nothing. */ |
| 2751 | return; |
| 2752 | |
| 2753 | if (!batch->memcg) |
| 2754 | return; |
| 2755 | /* |
| 2756 | * This "batch->memcg" is valid without any css_get/put etc... |
| 2757 | * bacause we hide charges behind us. |
| 2758 | */ |
| 2759 | if (batch->bytes) |
| 2760 | res_counter_uncharge(&batch->memcg->res, batch->bytes); |
| 2761 | if (batch->memsw_bytes) |
| 2762 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2763 | memcg_oom_recover(batch->memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 2764 | /* forget this pointer (for sanity check) */ |
| 2765 | batch->memcg = NULL; |
| 2766 | } |
| 2767 | |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2768 | #ifdef CONFIG_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2769 | /* |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2770 | * called after __delete_from_swap_cache() and drop "page" account. |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2771 | * memcg information is recorded to swap_cgroup of "ent" |
| 2772 | */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2773 | void |
| 2774 | 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] | 2775 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2776 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2777 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2778 | |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 2779 | if (!swapout) /* this was a swap cache but the swap is unused ! */ |
| 2780 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; |
| 2781 | |
| 2782 | memcg = __mem_cgroup_uncharge_common(page, ctype); |
| 2783 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2784 | /* |
| 2785 | * record memcg information, if swapout && memcg != NULL, |
| 2786 | * mem_cgroup_get() was called in uncharge(). |
| 2787 | */ |
| 2788 | if (do_swap_account && swapout && memcg) |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2789 | swap_cgroup_record(ent, css_id(&memcg->css)); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2790 | } |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 2791 | #endif |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 2792 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2793 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 2794 | /* |
| 2795 | * called from swap_entry_free(). remove record in swap_cgroup and |
| 2796 | * uncharge "memsw" account. |
| 2797 | */ |
| 2798 | void mem_cgroup_uncharge_swap(swp_entry_t ent) |
| 2799 | { |
| 2800 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2801 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2802 | |
| 2803 | if (!do_swap_account) |
| 2804 | return; |
| 2805 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2806 | id = swap_cgroup_record(ent, 0); |
| 2807 | rcu_read_lock(); |
| 2808 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2809 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2810 | /* |
| 2811 | * We uncharge this because swap is freed. |
| 2812 | * This memcg can be obsolete one. We avoid calling css_tryget |
| 2813 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2814 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 2815 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2816 | mem_cgroup_swap_statistics(memcg, false); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2817 | mem_cgroup_put(memcg); |
| 2818 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2819 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2820 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2821 | |
| 2822 | /** |
| 2823 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2824 | * @entry: swap entry to be moved |
| 2825 | * @from: mem_cgroup which the entry is moved from |
| 2826 | * @to: mem_cgroup which the entry is moved to |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2827 | * @need_fixup: whether we should fixup res_counters and refcounts. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2828 | * |
| 2829 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2830 | * as the mem_cgroup's id of @from. |
| 2831 | * |
| 2832 | * Returns 0 on success, -EINVAL on failure. |
| 2833 | * |
| 2834 | * The caller must have charged to @to, IOW, called res_counter_charge() about |
| 2835 | * both res and memsw, and called css_get(). |
| 2836 | */ |
| 2837 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2838 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2839 | { |
| 2840 | unsigned short old_id, new_id; |
| 2841 | |
| 2842 | old_id = css_id(&from->css); |
| 2843 | new_id = css_id(&to->css); |
| 2844 | |
| 2845 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2846 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2847 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2848 | /* |
| 2849 | * This function is only called from task migration context now. |
| 2850 | * It postpones res_counter and refcount handling till the end |
| 2851 | * of task migration(mem_cgroup_clear_mc()) for performance |
| 2852 | * improvement. But we cannot postpone mem_cgroup_get(to) |
| 2853 | * because if the process that has been moved to @to does |
| 2854 | * swap-in, the refcount of @to might be decreased to 0. |
| 2855 | */ |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2856 | mem_cgroup_get(to); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2857 | if (need_fixup) { |
| 2858 | if (!mem_cgroup_is_root(from)) |
| 2859 | res_counter_uncharge(&from->memsw, PAGE_SIZE); |
| 2860 | mem_cgroup_put(from); |
| 2861 | /* |
| 2862 | * we charged both to->res and to->memsw, so we should |
| 2863 | * uncharge to->res. |
| 2864 | */ |
| 2865 | if (!mem_cgroup_is_root(to)) |
| 2866 | res_counter_uncharge(&to->res, PAGE_SIZE); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2867 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2868 | return 0; |
| 2869 | } |
| 2870 | return -EINVAL; |
| 2871 | } |
| 2872 | #else |
| 2873 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 2874 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2875 | { |
| 2876 | return -EINVAL; |
| 2877 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2878 | #endif |
| 2879 | |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2880 | /* |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2881 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
| 2882 | * page belongs to. |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2883 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2884 | int mem_cgroup_prepare_migration(struct page *page, |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 2885 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 2886 | { |
| 2887 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2888 | struct mem_cgroup *mem = NULL; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2889 | enum charge_type ctype; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2890 | int ret = 0; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 2891 | |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 2892 | *ptr = NULL; |
| 2893 | |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2894 | VM_BUG_ON(PageTransHuge(page)); |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 2895 | if (mem_cgroup_disabled()) |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 2896 | return 0; |
| 2897 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2898 | pc = lookup_page_cgroup(page); |
| 2899 | lock_page_cgroup(pc); |
| 2900 | if (PageCgroupUsed(pc)) { |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2901 | mem = pc->mem_cgroup; |
| 2902 | css_get(&mem->css); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2903 | /* |
| 2904 | * At migrating an anonymous page, its mapcount goes down |
| 2905 | * to 0 and uncharge() will be called. But, even if it's fully |
| 2906 | * unmapped, migration may fail and this page has to be |
| 2907 | * charged again. We set MIGRATION flag here and delay uncharge |
| 2908 | * until end_migration() is called |
| 2909 | * |
| 2910 | * Corner Case Thinking |
| 2911 | * A) |
| 2912 | * When the old page was mapped as Anon and it's unmap-and-freed |
| 2913 | * while migration was ongoing. |
| 2914 | * If unmap finds the old page, uncharge() of it will be delayed |
| 2915 | * until end_migration(). If unmap finds a new page, it's |
| 2916 | * uncharged when it make mapcount to be 1->0. If unmap code |
| 2917 | * finds swap_migration_entry, the new page will not be mapped |
| 2918 | * and end_migration() will find it(mapcount==0). |
| 2919 | * |
| 2920 | * B) |
| 2921 | * When the old page was mapped but migraion fails, the kernel |
| 2922 | * remaps it. A charge for it is kept by MIGRATION flag even |
| 2923 | * if mapcount goes down to 0. We can do remap successfully |
| 2924 | * without charging it again. |
| 2925 | * |
| 2926 | * C) |
| 2927 | * The "old" page is under lock_page() until the end of |
| 2928 | * migration, so, the old page itself will not be swapped-out. |
| 2929 | * If the new page is swapped out before end_migraton, our |
| 2930 | * hook to usual swap-out path will catch the event. |
| 2931 | */ |
| 2932 | if (PageAnon(page)) |
| 2933 | SetPageCgroupMigration(pc); |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 2934 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2935 | unlock_page_cgroup(pc); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2936 | /* |
| 2937 | * If the page is not charged at this point, |
| 2938 | * we return here. |
| 2939 | */ |
| 2940 | if (!mem) |
| 2941 | return 0; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2942 | |
Andrea Arcangeli | 93d5c9b | 2010-04-23 13:17:39 -0400 | [diff] [blame] | 2943 | *ptr = mem; |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 2944 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, ptr, false, PAGE_SIZE); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2945 | css_put(&mem->css);/* drop extra refcnt */ |
| 2946 | if (ret || *ptr == NULL) { |
| 2947 | if (PageAnon(page)) { |
| 2948 | lock_page_cgroup(pc); |
| 2949 | ClearPageCgroupMigration(pc); |
| 2950 | unlock_page_cgroup(pc); |
| 2951 | /* |
| 2952 | * The old page may be fully unmapped while we kept it. |
| 2953 | */ |
| 2954 | mem_cgroup_uncharge_page(page); |
| 2955 | } |
| 2956 | return -ENOMEM; |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2957 | } |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2958 | /* |
| 2959 | * We charge new page before it's used/mapped. So, even if unlock_page() |
| 2960 | * is called before end_migration, we can catch all events on this new |
| 2961 | * page. In the case new page is migrated but not remapped, new page's |
| 2962 | * mapcount will be finally 0 and we call uncharge in end_migration(). |
| 2963 | */ |
| 2964 | pc = lookup_page_cgroup(newpage); |
| 2965 | if (PageAnon(page)) |
| 2966 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; |
| 2967 | else if (page_is_file_cache(page)) |
| 2968 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
| 2969 | else |
| 2970 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 2971 | __mem_cgroup_commit_charge(mem, pc, ctype, PAGE_SIZE); |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2972 | return ret; |
| 2973 | } |
Hugh Dickins | fb59e9f | 2008-03-04 14:29:16 -0800 | [diff] [blame] | 2974 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2975 | /* remove redundant charge if migration failed*/ |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2976 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 2977 | struct page *oldpage, struct page *newpage, bool migration_ok) |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 2978 | { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2979 | struct page *used, *unused; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2980 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2981 | |
| 2982 | if (!mem) |
| 2983 | return; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2984 | /* blocks rmdir() */ |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 2985 | cgroup_exclude_rmdir(&mem->css); |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 2986 | if (!migration_ok) { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2987 | used = oldpage; |
| 2988 | unused = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2989 | } else { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2990 | used = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 2991 | unused = oldpage; |
| 2992 | } |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2993 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2994 | * We disallowed uncharge of pages under migration because mapcount |
| 2995 | * of the page goes down to zero, temporarly. |
| 2996 | * Clear the flag and check the page should be charged. |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 2997 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 2998 | pc = lookup_page_cgroup(oldpage); |
| 2999 | lock_page_cgroup(pc); |
| 3000 | ClearPageCgroupMigration(pc); |
| 3001 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3002 | |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3003 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
| 3004 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3005 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3006 | * If a page is a file cache, radix-tree replacement is very atomic |
| 3007 | * and we can skip this check. When it was an Anon page, its mapcount |
| 3008 | * goes down to 0. But because we added MIGRATION flage, it's not |
| 3009 | * uncharged yet. There are several case but page->mapcount check |
| 3010 | * and USED bit check in mem_cgroup_uncharge_page() will do enough |
| 3011 | * check. (see prepare_charge() also) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 3012 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3013 | if (PageAnon(used)) |
| 3014 | mem_cgroup_uncharge_page(used); |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 3015 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3016 | * At migration, we may charge account against cgroup which has no |
| 3017 | * tasks. |
KAMEZAWA Hiroyuki | 8870326 | 2009-07-29 15:04:06 -0700 | [diff] [blame] | 3018 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
| 3019 | * In that case, we need to call pre_destroy() again. check it here. |
| 3020 | */ |
| 3021 | cgroup_release_and_wakeup_rmdir(&mem->css); |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 3022 | } |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 3023 | |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3024 | /* |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 3025 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
| 3026 | * Calling hierarchical_reclaim is not enough because we should update |
| 3027 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. |
| 3028 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, |
| 3029 | * not from the memcg which this page would be charged to. |
| 3030 | * try_charge_swapin does all of these works properly. |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3031 | */ |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 3032 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 3033 | struct mm_struct *mm, |
| 3034 | gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3035 | { |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 3036 | struct mem_cgroup *mem; |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 3037 | int ret; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3038 | |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 3039 | if (mem_cgroup_disabled()) |
Li Zefan | cede86a | 2008-07-25 01:47:18 -0700 | [diff] [blame] | 3040 | return 0; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3041 | |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 3042 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
| 3043 | if (!ret) |
| 3044 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3045 | |
Daisuke Nishimura | ae3abae | 2009-04-30 15:08:19 -0700 | [diff] [blame] | 3046 | return ret; |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3047 | } |
| 3048 | |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame^] | 3049 | #ifdef CONFIG_DEBUG_VM |
| 3050 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) |
| 3051 | { |
| 3052 | struct page_cgroup *pc; |
| 3053 | |
| 3054 | pc = lookup_page_cgroup(page); |
| 3055 | if (likely(pc) && PageCgroupUsed(pc)) |
| 3056 | return pc; |
| 3057 | return NULL; |
| 3058 | } |
| 3059 | |
| 3060 | bool mem_cgroup_bad_page_check(struct page *page) |
| 3061 | { |
| 3062 | if (mem_cgroup_disabled()) |
| 3063 | return false; |
| 3064 | |
| 3065 | return lookup_page_cgroup_used(page) != NULL; |
| 3066 | } |
| 3067 | |
| 3068 | void mem_cgroup_print_bad_page(struct page *page) |
| 3069 | { |
| 3070 | struct page_cgroup *pc; |
| 3071 | |
| 3072 | pc = lookup_page_cgroup_used(page); |
| 3073 | if (pc) { |
| 3074 | int ret = -1; |
| 3075 | char *path; |
| 3076 | |
| 3077 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", |
| 3078 | pc, pc->flags, pc->mem_cgroup); |
| 3079 | |
| 3080 | path = kmalloc(PATH_MAX, GFP_KERNEL); |
| 3081 | if (path) { |
| 3082 | rcu_read_lock(); |
| 3083 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, |
| 3084 | path, PATH_MAX); |
| 3085 | rcu_read_unlock(); |
| 3086 | } |
| 3087 | |
| 3088 | printk(KERN_CONT "(%s)\n", |
| 3089 | (ret < 0) ? "cannot get the path" : path); |
| 3090 | kfree(path); |
| 3091 | } |
| 3092 | } |
| 3093 | #endif |
| 3094 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3095 | static DEFINE_MUTEX(set_limit_mutex); |
| 3096 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 3097 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3098 | unsigned long long val) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3099 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3100 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3101 | u64 memswlimit, memlimit; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3102 | int ret = 0; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3103 | int children = mem_cgroup_count_children(memcg); |
| 3104 | u64 curusage, oldusage; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3105 | int enlarge; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3106 | |
| 3107 | /* |
| 3108 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 3109 | * is depends on callers. We set our retry-count to be function |
| 3110 | * of # of children which we should visit in this loop. |
| 3111 | */ |
| 3112 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; |
| 3113 | |
| 3114 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3115 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3116 | enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3117 | while (retry_count) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3118 | if (signal_pending(current)) { |
| 3119 | ret = -EINTR; |
| 3120 | break; |
| 3121 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3122 | /* |
| 3123 | * Rather than hide all in some function, I do this in |
| 3124 | * open coded manner. You see what this really does. |
| 3125 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
| 3126 | */ |
| 3127 | mutex_lock(&set_limit_mutex); |
| 3128 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3129 | if (memswlimit < val) { |
| 3130 | ret = -EINVAL; |
| 3131 | mutex_unlock(&set_limit_mutex); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3132 | break; |
| 3133 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3134 | |
| 3135 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3136 | if (memlimit < val) |
| 3137 | enlarge = 1; |
| 3138 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3139 | ret = res_counter_set_limit(&memcg->res, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3140 | if (!ret) { |
| 3141 | if (memswlimit == val) |
| 3142 | memcg->memsw_is_minimum = true; |
| 3143 | else |
| 3144 | memcg->memsw_is_minimum = false; |
| 3145 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3146 | mutex_unlock(&set_limit_mutex); |
| 3147 | |
| 3148 | if (!ret) |
| 3149 | break; |
| 3150 | |
Bob Liu | aa20d48 | 2009-12-15 16:47:14 -0800 | [diff] [blame] | 3151 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3152 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3153 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
| 3154 | /* Usage is reduced ? */ |
| 3155 | if (curusage >= oldusage) |
| 3156 | retry_count--; |
| 3157 | else |
| 3158 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3159 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3160 | if (!ret && enlarge) |
| 3161 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 3162 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3163 | return ret; |
| 3164 | } |
| 3165 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 3166 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
| 3167 | unsigned long long val) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3168 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3169 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3170 | u64 memlimit, memswlimit, oldusage, curusage; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3171 | int children = mem_cgroup_count_children(memcg); |
| 3172 | int ret = -EBUSY; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3173 | int enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3174 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3175 | /* see mem_cgroup_resize_res_limit */ |
| 3176 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
| 3177 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3178 | while (retry_count) { |
| 3179 | if (signal_pending(current)) { |
| 3180 | ret = -EINTR; |
| 3181 | break; |
| 3182 | } |
| 3183 | /* |
| 3184 | * Rather than hide all in some function, I do this in |
| 3185 | * open coded manner. You see what this really does. |
| 3186 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
| 3187 | */ |
| 3188 | mutex_lock(&set_limit_mutex); |
| 3189 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3190 | if (memlimit > val) { |
| 3191 | ret = -EINVAL; |
| 3192 | mutex_unlock(&set_limit_mutex); |
| 3193 | break; |
| 3194 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3195 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3196 | if (memswlimit < val) |
| 3197 | enlarge = 1; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3198 | ret = res_counter_set_limit(&memcg->memsw, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 3199 | if (!ret) { |
| 3200 | if (memlimit == val) |
| 3201 | memcg->memsw_is_minimum = true; |
| 3202 | else |
| 3203 | memcg->memsw_is_minimum = false; |
| 3204 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3205 | mutex_unlock(&set_limit_mutex); |
| 3206 | |
| 3207 | if (!ret) |
| 3208 | break; |
| 3209 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3210 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 3211 | MEM_CGROUP_RECLAIM_NOSWAP | |
| 3212 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3213 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3214 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3215 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3216 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3217 | else |
| 3218 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3219 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3220 | if (!ret && enlarge) |
| 3221 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3222 | return ret; |
| 3223 | } |
| 3224 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3225 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 3226 | gfp_t gfp_mask) |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3227 | { |
| 3228 | unsigned long nr_reclaimed = 0; |
| 3229 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 3230 | unsigned long reclaimed; |
| 3231 | int loop = 0; |
| 3232 | struct mem_cgroup_tree_per_zone *mctz; |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3233 | unsigned long long excess; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3234 | |
| 3235 | if (order > 0) |
| 3236 | return 0; |
| 3237 | |
KOSAKI Motohiro | 00918b6 | 2010-08-10 18:03:05 -0700 | [diff] [blame] | 3238 | 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] | 3239 | /* |
| 3240 | * This loop can run a while, specially if mem_cgroup's continuously |
| 3241 | * keep exceeding their soft limit and putting the system under |
| 3242 | * pressure |
| 3243 | */ |
| 3244 | do { |
| 3245 | if (next_mz) |
| 3246 | mz = next_mz; |
| 3247 | else |
| 3248 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 3249 | if (!mz) |
| 3250 | break; |
| 3251 | |
| 3252 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
| 3253 | gfp_mask, |
| 3254 | MEM_CGROUP_RECLAIM_SOFT); |
| 3255 | nr_reclaimed += reclaimed; |
| 3256 | spin_lock(&mctz->lock); |
| 3257 | |
| 3258 | /* |
| 3259 | * If we failed to reclaim anything from this memory cgroup |
| 3260 | * it is time to move on to the next cgroup |
| 3261 | */ |
| 3262 | next_mz = NULL; |
| 3263 | if (!reclaimed) { |
| 3264 | do { |
| 3265 | /* |
| 3266 | * Loop until we find yet another one. |
| 3267 | * |
| 3268 | * By the time we get the soft_limit lock |
| 3269 | * again, someone might have aded the |
| 3270 | * group back on the RB tree. Iterate to |
| 3271 | * make sure we get a different mem. |
| 3272 | * mem_cgroup_largest_soft_limit_node returns |
| 3273 | * NULL if no other cgroup is present on |
| 3274 | * the tree |
| 3275 | */ |
| 3276 | next_mz = |
| 3277 | __mem_cgroup_largest_soft_limit_node(mctz); |
| 3278 | if (next_mz == mz) { |
| 3279 | css_put(&next_mz->mem->css); |
| 3280 | next_mz = NULL; |
| 3281 | } else /* next_mz == NULL or other memcg */ |
| 3282 | break; |
| 3283 | } while (1); |
| 3284 | } |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3285 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3286 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3287 | /* |
| 3288 | * One school of thought says that we should not add |
| 3289 | * back the node to the tree if reclaim returns 0. |
| 3290 | * But our reclaim could return 0, simply because due |
| 3291 | * to priority we are exposing a smaller subset of |
| 3292 | * memory to reclaim from. Consider this as a longer |
| 3293 | * term TODO. |
| 3294 | */ |
KAMEZAWA Hiroyuki | ef8745c | 2009-10-01 15:44:12 -0700 | [diff] [blame] | 3295 | /* If excess == 0, no tree ops */ |
| 3296 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 3297 | spin_unlock(&mctz->lock); |
| 3298 | css_put(&mz->mem->css); |
| 3299 | loop++; |
| 3300 | /* |
| 3301 | * Could not reclaim anything and there are no more |
| 3302 | * mem cgroups to try or we seem to be looping without |
| 3303 | * reclaiming anything. |
| 3304 | */ |
| 3305 | if (!nr_reclaimed && |
| 3306 | (next_mz == NULL || |
| 3307 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3308 | break; |
| 3309 | } while (!nr_reclaimed); |
| 3310 | if (next_mz) |
| 3311 | css_put(&next_mz->mem->css); |
| 3312 | return nr_reclaimed; |
| 3313 | } |
| 3314 | |
KAMEZAWA Hiroyuki | c9b0ed5 | 2008-07-25 01:47:15 -0700 | [diff] [blame] | 3315 | /* |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3316 | * This routine traverse page_cgroup in given list and drop them all. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3317 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
| 3318 | */ |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3319 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3320 | int node, int zid, enum lru_list lru) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3321 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3322 | struct zone *zone; |
| 3323 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3324 | struct page_cgroup *pc, *busy; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3325 | unsigned long flags, loop; |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3326 | struct list_head *list; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3327 | int ret = 0; |
KAMEZAWA Hiroyuki | 072c56c | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 3328 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3329 | zone = &NODE_DATA(node)->node_zones[zid]; |
| 3330 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 3331 | list = &mz->lists[lru]; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3332 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3333 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
| 3334 | /* give some margin against EBUSY etc...*/ |
| 3335 | loop += 256; |
| 3336 | busy = NULL; |
| 3337 | while (loop--) { |
| 3338 | ret = 0; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3339 | spin_lock_irqsave(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3340 | if (list_empty(list)) { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3341 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3342 | break; |
| 3343 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3344 | pc = list_entry(list->prev, struct page_cgroup, lru); |
| 3345 | if (busy == pc) { |
| 3346 | list_move(&pc->lru, list); |
Thiago Farina | 648bcc7 | 2010-03-05 13:42:04 -0800 | [diff] [blame] | 3347 | busy = NULL; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3348 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3349 | continue; |
| 3350 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3351 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3352 | |
KAMEZAWA Hiroyuki | 2c26fdd | 2009-01-07 18:08:10 -0800 | [diff] [blame] | 3353 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3354 | if (ret == -ENOMEM) |
| 3355 | break; |
| 3356 | |
| 3357 | if (ret == -EBUSY || ret == -EINVAL) { |
| 3358 | /* found lock contention or "pc" is obsolete. */ |
| 3359 | busy = pc; |
| 3360 | cond_resched(); |
| 3361 | } else |
| 3362 | busy = NULL; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3363 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3364 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3365 | if (!ret && !list_empty(list)) |
| 3366 | return -EBUSY; |
| 3367 | return ret; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3368 | } |
| 3369 | |
| 3370 | /* |
| 3371 | * make mem_cgroup's charge to be 0 if there is no task. |
| 3372 | * This enables deleting this mem_cgroup. |
| 3373 | */ |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3374 | 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] | 3375 | { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3376 | int ret; |
| 3377 | int node, zid, shrink; |
| 3378 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3379 | struct cgroup *cgrp = mem->css.cgroup; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 3380 | |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3381 | css_get(&mem->css); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3382 | |
| 3383 | shrink = 0; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3384 | /* should free all ? */ |
| 3385 | if (free_all) |
| 3386 | goto try_to_free; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3387 | move_account: |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3388 | do { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3389 | ret = -EBUSY; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3390 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3391 | goto out; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3392 | ret = -EINTR; |
| 3393 | if (signal_pending(current)) |
| 3394 | goto out; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3395 | /* This is for making all *used* pages to be on LRU. */ |
| 3396 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 3397 | drain_all_stock_sync(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3398 | ret = 0; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 3399 | mem_cgroup_start_move(mem); |
KAMEZAWA Hiroyuki | 299b4ea | 2009-01-29 14:25:17 -0800 | [diff] [blame] | 3400 | for_each_node_state(node, N_HIGH_MEMORY) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3401 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 3402 | enum lru_list l; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3403 | for_each_lru(l) { |
| 3404 | ret = mem_cgroup_force_empty_list(mem, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3405 | node, zid, l); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3406 | if (ret) |
| 3407 | break; |
| 3408 | } |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 3409 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3410 | if (ret) |
| 3411 | break; |
| 3412 | } |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 3413 | mem_cgroup_end_move(mem); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3414 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3415 | /* it seems parent cgroup doesn't have enough mem */ |
| 3416 | if (ret == -ENOMEM) |
| 3417 | goto try_to_free; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3418 | cond_resched(); |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3419 | /* "ret" should also be checked to ensure all lists are empty. */ |
| 3420 | } while (mem->res.usage > 0 || ret); |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3421 | out: |
| 3422 | css_put(&mem->css); |
| 3423 | return ret; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3424 | |
| 3425 | try_to_free: |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3426 | /* returns EBUSY if there is a task or if we come here twice. */ |
| 3427 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3428 | ret = -EBUSY; |
| 3429 | goto out; |
| 3430 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3431 | /* we call try-to-free pages for make this cgroup empty */ |
| 3432 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3433 | /* try to free all pages in this cgroup */ |
| 3434 | shrink = 1; |
| 3435 | while (nr_retries && mem->res.usage > 0) { |
| 3436 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3437 | |
| 3438 | if (signal_pending(current)) { |
| 3439 | ret = -EINTR; |
| 3440 | goto out; |
| 3441 | } |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3442 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
| 3443 | false, get_swappiness(mem)); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3444 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3445 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3446 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3447 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3448 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3449 | |
| 3450 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3451 | lru_add_drain(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3452 | /* try move_account...there may be some *locked* pages. */ |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 3453 | goto move_account; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3454 | } |
| 3455 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3456 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
| 3457 | { |
| 3458 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); |
| 3459 | } |
| 3460 | |
| 3461 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3462 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
| 3463 | { |
| 3464 | return mem_cgroup_from_cont(cont)->use_hierarchy; |
| 3465 | } |
| 3466 | |
| 3467 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, |
| 3468 | u64 val) |
| 3469 | { |
| 3470 | int retval = 0; |
| 3471 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
| 3472 | struct cgroup *parent = cont->parent; |
| 3473 | struct mem_cgroup *parent_mem = NULL; |
| 3474 | |
| 3475 | if (parent) |
| 3476 | parent_mem = mem_cgroup_from_cont(parent); |
| 3477 | |
| 3478 | cgroup_lock(); |
| 3479 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 3480 | * 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] | 3481 | * in the child subtrees. If it is unset, then the change can |
| 3482 | * occur, provided the current cgroup has no children. |
| 3483 | * |
| 3484 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 3485 | * set if there are no children. |
| 3486 | */ |
| 3487 | if ((!parent_mem || !parent_mem->use_hierarchy) && |
| 3488 | (val == 1 || val == 0)) { |
| 3489 | if (list_empty(&cont->children)) |
| 3490 | mem->use_hierarchy = val; |
| 3491 | else |
| 3492 | retval = -EBUSY; |
| 3493 | } else |
| 3494 | retval = -EINVAL; |
| 3495 | cgroup_unlock(); |
| 3496 | |
| 3497 | return retval; |
| 3498 | } |
| 3499 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3500 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3501 | static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, |
| 3502 | enum mem_cgroup_stat_index idx) |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3503 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3504 | struct mem_cgroup *iter; |
| 3505 | s64 val = 0; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3506 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3507 | /* each per cpu's value can be minus.Then, use s64 */ |
| 3508 | for_each_mem_cgroup_tree(iter, mem) |
| 3509 | val += mem_cgroup_read_stat(iter, idx); |
| 3510 | |
| 3511 | if (val < 0) /* race ? */ |
| 3512 | val = 0; |
| 3513 | return val; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3514 | } |
| 3515 | |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3516 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
| 3517 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3518 | u64 val; |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3519 | |
| 3520 | if (!mem_cgroup_is_root(mem)) { |
| 3521 | if (!swap) |
| 3522 | return res_counter_read_u64(&mem->res, RES_USAGE); |
| 3523 | else |
| 3524 | return res_counter_read_u64(&mem->memsw, RES_USAGE); |
| 3525 | } |
| 3526 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3527 | val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE); |
| 3528 | 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] | 3529 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3530 | if (swap) |
| 3531 | val += mem_cgroup_get_recursive_idx_stat(mem, |
| 3532 | MEM_CGROUP_STAT_SWAPOUT); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3533 | |
| 3534 | return val << PAGE_SHIFT; |
| 3535 | } |
| 3536 | |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 3537 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3538 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3539 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3540 | u64 val; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3541 | int type, name; |
| 3542 | |
| 3543 | type = MEMFILE_TYPE(cft->private); |
| 3544 | name = MEMFILE_ATTR(cft->private); |
| 3545 | switch (type) { |
| 3546 | case _MEM: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3547 | if (name == RES_USAGE) |
| 3548 | val = mem_cgroup_usage(mem, false); |
| 3549 | else |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3550 | val = res_counter_read_u64(&mem->res, name); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3551 | break; |
| 3552 | case _MEMSWAP: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 3553 | if (name == RES_USAGE) |
| 3554 | val = mem_cgroup_usage(mem, true); |
| 3555 | else |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 3556 | val = res_counter_read_u64(&mem->memsw, name); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3557 | break; |
| 3558 | default: |
| 3559 | BUG(); |
| 3560 | break; |
| 3561 | } |
| 3562 | return val; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3563 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3564 | /* |
| 3565 | * The user of this function is... |
| 3566 | * RES_LIMIT. |
| 3567 | */ |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 3568 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
| 3569 | const char *buffer) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3570 | { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3571 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3572 | int type, name; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3573 | unsigned long long val; |
| 3574 | int ret; |
| 3575 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3576 | type = MEMFILE_TYPE(cft->private); |
| 3577 | name = MEMFILE_ATTR(cft->private); |
| 3578 | switch (name) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3579 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3580 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3581 | ret = -EINVAL; |
| 3582 | break; |
| 3583 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3584 | /* This function does all necessary parse...reuse it */ |
| 3585 | ret = res_counter_memparse_write_strategy(buffer, &val); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3586 | if (ret) |
| 3587 | break; |
| 3588 | if (type == _MEM) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3589 | ret = mem_cgroup_resize_limit(memcg, val); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3590 | else |
| 3591 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3592 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3593 | case RES_SOFT_LIMIT: |
| 3594 | ret = res_counter_memparse_write_strategy(buffer, &val); |
| 3595 | if (ret) |
| 3596 | break; |
| 3597 | /* |
| 3598 | * For memsw, soft limits are hard to implement in terms |
| 3599 | * of semantics, for now, we support soft limits for |
| 3600 | * control without swap |
| 3601 | */ |
| 3602 | if (type == _MEM) |
| 3603 | ret = res_counter_set_soft_limit(&memcg->res, val); |
| 3604 | else |
| 3605 | ret = -EINVAL; |
| 3606 | break; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3607 | default: |
| 3608 | ret = -EINVAL; /* should be BUG() ? */ |
| 3609 | break; |
| 3610 | } |
| 3611 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3612 | } |
| 3613 | |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3614 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
| 3615 | unsigned long long *mem_limit, unsigned long long *memsw_limit) |
| 3616 | { |
| 3617 | struct cgroup *cgroup; |
| 3618 | unsigned long long min_limit, min_memsw_limit, tmp; |
| 3619 | |
| 3620 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3621 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3622 | cgroup = memcg->css.cgroup; |
| 3623 | if (!memcg->use_hierarchy) |
| 3624 | goto out; |
| 3625 | |
| 3626 | while (cgroup->parent) { |
| 3627 | cgroup = cgroup->parent; |
| 3628 | memcg = mem_cgroup_from_cont(cgroup); |
| 3629 | if (!memcg->use_hierarchy) |
| 3630 | break; |
| 3631 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 3632 | min_limit = min(min_limit, tmp); |
| 3633 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 3634 | min_memsw_limit = min(min_memsw_limit, tmp); |
| 3635 | } |
| 3636 | out: |
| 3637 | *mem_limit = min_limit; |
| 3638 | *memsw_limit = min_memsw_limit; |
| 3639 | return; |
| 3640 | } |
| 3641 | |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3642 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3643 | { |
| 3644 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3645 | int type, name; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3646 | |
| 3647 | mem = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3648 | type = MEMFILE_TYPE(event); |
| 3649 | name = MEMFILE_ATTR(event); |
| 3650 | switch (name) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3651 | case RES_MAX_USAGE: |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3652 | if (type == _MEM) |
| 3653 | res_counter_reset_max(&mem->res); |
| 3654 | else |
| 3655 | res_counter_reset_max(&mem->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3656 | break; |
| 3657 | case RES_FAILCNT: |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3658 | if (type == _MEM) |
| 3659 | res_counter_reset_failcnt(&mem->res); |
| 3660 | else |
| 3661 | res_counter_reset_failcnt(&mem->memsw); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3662 | break; |
| 3663 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3664 | |
Pavel Emelyanov | 85cc59d | 2008-04-29 01:00:20 -0700 | [diff] [blame] | 3665 | return 0; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3666 | } |
| 3667 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3668 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
| 3669 | struct cftype *cft) |
| 3670 | { |
| 3671 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; |
| 3672 | } |
| 3673 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3674 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3675 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 3676 | struct cftype *cft, u64 val) |
| 3677 | { |
| 3678 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 3679 | |
| 3680 | if (val >= (1 << NR_MOVE_TYPE)) |
| 3681 | return -EINVAL; |
| 3682 | /* |
| 3683 | * We check this value several times in both in can_attach() and |
| 3684 | * attach(), so we need cgroup lock to prevent this value from being |
| 3685 | * inconsistent. |
| 3686 | */ |
| 3687 | cgroup_lock(); |
| 3688 | mem->move_charge_at_immigrate = val; |
| 3689 | cgroup_unlock(); |
| 3690 | |
| 3691 | return 0; |
| 3692 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3693 | #else |
| 3694 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
| 3695 | struct cftype *cft, u64 val) |
| 3696 | { |
| 3697 | return -ENOSYS; |
| 3698 | } |
| 3699 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3700 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3701 | |
| 3702 | /* For read statistics */ |
| 3703 | enum { |
| 3704 | MCS_CACHE, |
| 3705 | MCS_RSS, |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 3706 | MCS_FILE_MAPPED, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3707 | MCS_PGPGIN, |
| 3708 | MCS_PGPGOUT, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3709 | MCS_SWAP, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3710 | MCS_INACTIVE_ANON, |
| 3711 | MCS_ACTIVE_ANON, |
| 3712 | MCS_INACTIVE_FILE, |
| 3713 | MCS_ACTIVE_FILE, |
| 3714 | MCS_UNEVICTABLE, |
| 3715 | NR_MCS_STAT, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3716 | }; |
| 3717 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3718 | struct mcs_total_stat { |
| 3719 | s64 stat[NR_MCS_STAT]; |
| 3720 | }; |
| 3721 | |
| 3722 | struct { |
| 3723 | char *local_name; |
| 3724 | char *total_name; |
| 3725 | } memcg_stat_strings[NR_MCS_STAT] = { |
| 3726 | {"cache", "total_cache"}, |
| 3727 | {"rss", "total_rss"}, |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 3728 | {"mapped_file", "total_mapped_file"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3729 | {"pgpgin", "total_pgpgin"}, |
| 3730 | {"pgpgout", "total_pgpgout"}, |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3731 | {"swap", "total_swap"}, |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3732 | {"inactive_anon", "total_inactive_anon"}, |
| 3733 | {"active_anon", "total_active_anon"}, |
| 3734 | {"inactive_file", "total_inactive_file"}, |
| 3735 | {"active_file", "total_active_file"}, |
| 3736 | {"unevictable", "total_unevictable"} |
| 3737 | }; |
| 3738 | |
| 3739 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3740 | static void |
| 3741 | 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] | 3742 | { |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3743 | s64 val; |
| 3744 | |
| 3745 | /* per cpu stat */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3746 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3747 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3748 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3749 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3750 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
KAMEZAWA Hiroyuki | d804658 | 2009-12-15 16:47:09 -0800 | [diff] [blame] | 3751 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3752 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3753 | s->stat[MCS_PGPGIN] += val; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3754 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3755 | s->stat[MCS_PGPGOUT] += val; |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3756 | if (do_swap_account) { |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 3757 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3758 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
| 3759 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3760 | |
| 3761 | /* per zone stat */ |
| 3762 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); |
| 3763 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
| 3764 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); |
| 3765 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
| 3766 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); |
| 3767 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
| 3768 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); |
| 3769 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
| 3770 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); |
| 3771 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3772 | } |
| 3773 | |
| 3774 | static void |
| 3775 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) |
| 3776 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3777 | struct mem_cgroup *iter; |
| 3778 | |
| 3779 | for_each_mem_cgroup_tree(iter, mem) |
| 3780 | mem_cgroup_get_local_stat(iter, s); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3781 | } |
| 3782 | |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 3783 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
| 3784 | struct cgroup_map_cb *cb) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3785 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3786 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3787 | struct mcs_total_stat mystat; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3788 | int i; |
| 3789 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3790 | memset(&mystat, 0, sizeof(mystat)); |
| 3791 | mem_cgroup_get_local_stat(mem_cont, &mystat); |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3792 | |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3793 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 3794 | if (i == MCS_SWAP && !do_swap_account) |
| 3795 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3796 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3797 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3798 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3799 | /* Hierarchical information */ |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3800 | { |
| 3801 | unsigned long long limit, memsw_limit; |
| 3802 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); |
| 3803 | cb->fill(cb, "hierarchical_memory_limit", limit); |
| 3804 | if (do_swap_account) |
| 3805 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); |
| 3806 | } |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3807 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3808 | memset(&mystat, 0, sizeof(mystat)); |
| 3809 | mem_cgroup_get_total_stat(mem_cont, &mystat); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3810 | for (i = 0; i < NR_MCS_STAT; i++) { |
| 3811 | if (i == MCS_SWAP && !do_swap_account) |
| 3812 | continue; |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3813 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3814 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3815 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3816 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 3817 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3818 | |
| 3819 | { |
| 3820 | int nid, zid; |
| 3821 | struct mem_cgroup_per_zone *mz; |
| 3822 | unsigned long recent_rotated[2] = {0, 0}; |
| 3823 | unsigned long recent_scanned[2] = {0, 0}; |
| 3824 | |
| 3825 | for_each_online_node(nid) |
| 3826 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 3827 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
| 3828 | |
| 3829 | recent_rotated[0] += |
| 3830 | mz->reclaim_stat.recent_rotated[0]; |
| 3831 | recent_rotated[1] += |
| 3832 | mz->reclaim_stat.recent_rotated[1]; |
| 3833 | recent_scanned[0] += |
| 3834 | mz->reclaim_stat.recent_scanned[0]; |
| 3835 | recent_scanned[1] += |
| 3836 | mz->reclaim_stat.recent_scanned[1]; |
| 3837 | } |
| 3838 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); |
| 3839 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); |
| 3840 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); |
| 3841 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); |
| 3842 | } |
| 3843 | #endif |
| 3844 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3845 | return 0; |
| 3846 | } |
| 3847 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3848 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
| 3849 | { |
| 3850 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 3851 | |
| 3852 | return get_swappiness(memcg); |
| 3853 | } |
| 3854 | |
| 3855 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, |
| 3856 | u64 val) |
| 3857 | { |
| 3858 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 3859 | struct mem_cgroup *parent; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3860 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3861 | if (val > 100) |
| 3862 | return -EINVAL; |
| 3863 | |
| 3864 | if (cgrp->parent == NULL) |
| 3865 | return -EINVAL; |
| 3866 | |
| 3867 | parent = mem_cgroup_from_cont(cgrp->parent); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3868 | |
| 3869 | cgroup_lock(); |
| 3870 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3871 | /* If under hierarchy, only empty-root can set this value */ |
| 3872 | if ((parent->use_hierarchy) || |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3873 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
| 3874 | cgroup_unlock(); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3875 | return -EINVAL; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3876 | } |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3877 | |
| 3878 | spin_lock(&memcg->reclaim_param_lock); |
| 3879 | memcg->swappiness = val; |
| 3880 | spin_unlock(&memcg->reclaim_param_lock); |
| 3881 | |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3882 | cgroup_unlock(); |
| 3883 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3884 | return 0; |
| 3885 | } |
| 3886 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3887 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3888 | { |
| 3889 | struct mem_cgroup_threshold_ary *t; |
| 3890 | u64 usage; |
| 3891 | int i; |
| 3892 | |
| 3893 | rcu_read_lock(); |
| 3894 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3895 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3896 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3897 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3898 | |
| 3899 | if (!t) |
| 3900 | goto unlock; |
| 3901 | |
| 3902 | usage = mem_cgroup_usage(memcg, swap); |
| 3903 | |
| 3904 | /* |
| 3905 | * current_threshold points to threshold just below usage. |
| 3906 | * If it's not true, a threshold was crossed after last |
| 3907 | * call of __mem_cgroup_threshold(). |
| 3908 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3909 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3910 | |
| 3911 | /* |
| 3912 | * Iterate backward over array of thresholds starting from |
| 3913 | * current_threshold and check if a threshold is crossed. |
| 3914 | * If none of thresholds below usage is crossed, we read |
| 3915 | * only one element of the array here. |
| 3916 | */ |
| 3917 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3918 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3919 | |
| 3920 | /* i = current_threshold + 1 */ |
| 3921 | i++; |
| 3922 | |
| 3923 | /* |
| 3924 | * Iterate forward over array of thresholds starting from |
| 3925 | * current_threshold+1 and check if a threshold is crossed. |
| 3926 | * If none of thresholds above usage is crossed, we read |
| 3927 | * only one element of the array here. |
| 3928 | */ |
| 3929 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3930 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3931 | |
| 3932 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3933 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3934 | unlock: |
| 3935 | rcu_read_unlock(); |
| 3936 | } |
| 3937 | |
| 3938 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3939 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3940 | while (memcg) { |
| 3941 | __mem_cgroup_threshold(memcg, false); |
| 3942 | if (do_swap_account) |
| 3943 | __mem_cgroup_threshold(memcg, true); |
| 3944 | |
| 3945 | memcg = parent_mem_cgroup(memcg); |
| 3946 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3947 | } |
| 3948 | |
| 3949 | static int compare_thresholds(const void *a, const void *b) |
| 3950 | { |
| 3951 | const struct mem_cgroup_threshold *_a = a; |
| 3952 | const struct mem_cgroup_threshold *_b = b; |
| 3953 | |
| 3954 | return _a->threshold - _b->threshold; |
| 3955 | } |
| 3956 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3957 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3958 | { |
| 3959 | struct mem_cgroup_eventfd_list *ev; |
| 3960 | |
| 3961 | list_for_each_entry(ev, &mem->oom_notify, list) |
| 3962 | eventfd_signal(ev->eventfd, 1); |
| 3963 | return 0; |
| 3964 | } |
| 3965 | |
| 3966 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) |
| 3967 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3968 | struct mem_cgroup *iter; |
| 3969 | |
| 3970 | for_each_mem_cgroup_tree(iter, mem) |
| 3971 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3972 | } |
| 3973 | |
| 3974 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, |
| 3975 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3976 | { |
| 3977 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3978 | struct mem_cgroup_thresholds *thresholds; |
| 3979 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3980 | int type = MEMFILE_TYPE(cft->private); |
| 3981 | u64 threshold, usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3982 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3983 | |
| 3984 | ret = res_counter_memparse_write_strategy(args, &threshold); |
| 3985 | if (ret) |
| 3986 | return ret; |
| 3987 | |
| 3988 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3989 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3990 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3991 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3992 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3993 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3994 | else |
| 3995 | BUG(); |
| 3996 | |
| 3997 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 3998 | |
| 3999 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4000 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4001 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4002 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4003 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4004 | |
| 4005 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4006 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4007 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4008 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4009 | ret = -ENOMEM; |
| 4010 | goto unlock; |
| 4011 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4012 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4013 | |
| 4014 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4015 | if (thresholds->primary) { |
| 4016 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4017 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4018 | } |
| 4019 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4020 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4021 | new->entries[size - 1].eventfd = eventfd; |
| 4022 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4023 | |
| 4024 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4025 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4026 | compare_thresholds, NULL); |
| 4027 | |
| 4028 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4029 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4030 | for (i = 0; i < size; i++) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4031 | if (new->entries[i].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4032 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4033 | * new->current_threshold will not be used until |
| 4034 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4035 | * it here. |
| 4036 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4037 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4038 | } |
| 4039 | } |
| 4040 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4041 | /* Free old spare buffer and save old primary buffer as spare */ |
| 4042 | kfree(thresholds->spare); |
| 4043 | thresholds->spare = thresholds->primary; |
| 4044 | |
| 4045 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4046 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4047 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4048 | synchronize_rcu(); |
| 4049 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4050 | unlock: |
| 4051 | mutex_unlock(&memcg->thresholds_lock); |
| 4052 | |
| 4053 | return ret; |
| 4054 | } |
| 4055 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4056 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4057 | struct cftype *cft, struct eventfd_ctx *eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4058 | { |
| 4059 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4060 | struct mem_cgroup_thresholds *thresholds; |
| 4061 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4062 | int type = MEMFILE_TYPE(cft->private); |
| 4063 | u64 usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4064 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4065 | |
| 4066 | mutex_lock(&memcg->thresholds_lock); |
| 4067 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4068 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4069 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4070 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4071 | else |
| 4072 | BUG(); |
| 4073 | |
| 4074 | /* |
| 4075 | * Something went wrong if we trying to unregister a threshold |
| 4076 | * if we don't have thresholds |
| 4077 | */ |
| 4078 | BUG_ON(!thresholds); |
| 4079 | |
| 4080 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 4081 | |
| 4082 | /* Check if a threshold crossed before removing */ |
| 4083 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4084 | |
| 4085 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4086 | size = 0; |
| 4087 | for (i = 0; i < thresholds->primary->size; i++) { |
| 4088 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4089 | size++; |
| 4090 | } |
| 4091 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4092 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4093 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4094 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 4095 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4096 | kfree(new); |
| 4097 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4098 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4099 | } |
| 4100 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4101 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4102 | |
| 4103 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4104 | new->current_threshold = -1; |
| 4105 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 4106 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4107 | continue; |
| 4108 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4109 | new->entries[j] = thresholds->primary->entries[i]; |
| 4110 | if (new->entries[j].threshold < usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4111 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4112 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4113 | * until rcu_assign_pointer(), so it's safe to increment |
| 4114 | * it here. |
| 4115 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4116 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4117 | } |
| 4118 | j++; |
| 4119 | } |
| 4120 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4121 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4122 | /* Swap primary and spare array */ |
| 4123 | thresholds->spare = thresholds->primary; |
| 4124 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4125 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4126 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4127 | synchronize_rcu(); |
| 4128 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4129 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4130 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4131 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4132 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
| 4133 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
| 4134 | { |
| 4135 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
| 4136 | struct mem_cgroup_eventfd_list *event; |
| 4137 | int type = MEMFILE_TYPE(cft->private); |
| 4138 | |
| 4139 | BUG_ON(type != _OOM_TYPE); |
| 4140 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 4141 | if (!event) |
| 4142 | return -ENOMEM; |
| 4143 | |
| 4144 | mutex_lock(&memcg_oom_mutex); |
| 4145 | |
| 4146 | event->eventfd = eventfd; |
| 4147 | list_add(&event->list, &memcg->oom_notify); |
| 4148 | |
| 4149 | /* already in OOM ? */ |
| 4150 | if (atomic_read(&memcg->oom_lock)) |
| 4151 | eventfd_signal(eventfd, 1); |
| 4152 | mutex_unlock(&memcg_oom_mutex); |
| 4153 | |
| 4154 | return 0; |
| 4155 | } |
| 4156 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4157 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4158 | struct cftype *cft, struct eventfd_ctx *eventfd) |
| 4159 | { |
| 4160 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4161 | struct mem_cgroup_eventfd_list *ev, *tmp; |
| 4162 | int type = MEMFILE_TYPE(cft->private); |
| 4163 | |
| 4164 | BUG_ON(type != _OOM_TYPE); |
| 4165 | |
| 4166 | mutex_lock(&memcg_oom_mutex); |
| 4167 | |
| 4168 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { |
| 4169 | if (ev->eventfd == eventfd) { |
| 4170 | list_del(&ev->list); |
| 4171 | kfree(ev); |
| 4172 | } |
| 4173 | } |
| 4174 | |
| 4175 | mutex_unlock(&memcg_oom_mutex); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4176 | } |
| 4177 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4178 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
| 4179 | struct cftype *cft, struct cgroup_map_cb *cb) |
| 4180 | { |
| 4181 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4182 | |
| 4183 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); |
| 4184 | |
| 4185 | if (atomic_read(&mem->oom_lock)) |
| 4186 | cb->fill(cb, "under_oom", 1); |
| 4187 | else |
| 4188 | cb->fill(cb, "under_oom", 0); |
| 4189 | return 0; |
| 4190 | } |
| 4191 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4192 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
| 4193 | struct cftype *cft, u64 val) |
| 4194 | { |
| 4195 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); |
| 4196 | struct mem_cgroup *parent; |
| 4197 | |
| 4198 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
| 4199 | if (!cgrp->parent || !((val == 0) || (val == 1))) |
| 4200 | return -EINVAL; |
| 4201 | |
| 4202 | parent = mem_cgroup_from_cont(cgrp->parent); |
| 4203 | |
| 4204 | cgroup_lock(); |
| 4205 | /* oom-kill-disable is a flag for subhierarchy. */ |
| 4206 | if ((parent->use_hierarchy) || |
| 4207 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { |
| 4208 | cgroup_unlock(); |
| 4209 | return -EINVAL; |
| 4210 | } |
| 4211 | mem->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 4212 | if (!val) |
| 4213 | memcg_oom_recover(mem); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4214 | cgroup_unlock(); |
| 4215 | return 0; |
| 4216 | } |
| 4217 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4218 | static struct cftype mem_cgroup_files[] = { |
| 4219 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4220 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4221 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4222 | .read_u64 = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4223 | .register_event = mem_cgroup_usage_register_event, |
| 4224 | .unregister_event = mem_cgroup_usage_unregister_event, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4225 | }, |
| 4226 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4227 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4228 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4229 | .trigger = mem_cgroup_reset, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4230 | .read_u64 = mem_cgroup_read, |
| 4231 | }, |
| 4232 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4233 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4234 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 4235 | .write_string = mem_cgroup_write, |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4236 | .read_u64 = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4237 | }, |
| 4238 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4239 | .name = "soft_limit_in_bytes", |
| 4240 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
| 4241 | .write_string = mem_cgroup_write, |
| 4242 | .read_u64 = mem_cgroup_read, |
| 4243 | }, |
| 4244 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4245 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4246 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 4247 | .trigger = mem_cgroup_reset, |
Paul Menage | 2c3daa7 | 2008-04-29 00:59:58 -0700 | [diff] [blame] | 4248 | .read_u64 = mem_cgroup_read, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4249 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4250 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4251 | .name = "stat", |
Paul Menage | c64745c | 2008-04-29 01:00:02 -0700 | [diff] [blame] | 4252 | .read_map = mem_control_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4253 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4254 | { |
| 4255 | .name = "force_empty", |
| 4256 | .trigger = mem_cgroup_force_empty_write, |
| 4257 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4258 | { |
| 4259 | .name = "use_hierarchy", |
| 4260 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4261 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4262 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4263 | { |
| 4264 | .name = "swappiness", |
| 4265 | .read_u64 = mem_cgroup_swappiness_read, |
| 4266 | .write_u64 = mem_cgroup_swappiness_write, |
| 4267 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4268 | { |
| 4269 | .name = "move_charge_at_immigrate", |
| 4270 | .read_u64 = mem_cgroup_move_charge_read, |
| 4271 | .write_u64 = mem_cgroup_move_charge_write, |
| 4272 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4273 | { |
| 4274 | .name = "oom_control", |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4275 | .read_map = mem_cgroup_oom_control_read, |
| 4276 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4277 | .register_event = mem_cgroup_oom_register_event, |
| 4278 | .unregister_event = mem_cgroup_oom_unregister_event, |
| 4279 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4280 | }, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4281 | }; |
| 4282 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4283 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 4284 | static struct cftype memsw_cgroup_files[] = { |
| 4285 | { |
| 4286 | .name = "memsw.usage_in_bytes", |
| 4287 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 4288 | .read_u64 = mem_cgroup_read, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4289 | .register_event = mem_cgroup_usage_register_event, |
| 4290 | .unregister_event = mem_cgroup_usage_unregister_event, |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4291 | }, |
| 4292 | { |
| 4293 | .name = "memsw.max_usage_in_bytes", |
| 4294 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 4295 | .trigger = mem_cgroup_reset, |
| 4296 | .read_u64 = mem_cgroup_read, |
| 4297 | }, |
| 4298 | { |
| 4299 | .name = "memsw.limit_in_bytes", |
| 4300 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 4301 | .write_string = mem_cgroup_write, |
| 4302 | .read_u64 = mem_cgroup_read, |
| 4303 | }, |
| 4304 | { |
| 4305 | .name = "memsw.failcnt", |
| 4306 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 4307 | .trigger = mem_cgroup_reset, |
| 4308 | .read_u64 = mem_cgroup_read, |
| 4309 | }, |
| 4310 | }; |
| 4311 | |
| 4312 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
| 4313 | { |
| 4314 | if (!do_swap_account) |
| 4315 | return 0; |
| 4316 | return cgroup_add_files(cont, ss, memsw_cgroup_files, |
| 4317 | ARRAY_SIZE(memsw_cgroup_files)); |
| 4318 | }; |
| 4319 | #else |
| 4320 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
| 4321 | { |
| 4322 | return 0; |
| 4323 | } |
| 4324 | #endif |
| 4325 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4326 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 4327 | { |
| 4328 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4329 | struct mem_cgroup_per_zone *mz; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4330 | enum lru_list l; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4331 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4332 | /* |
| 4333 | * This routine is called against possible nodes. |
| 4334 | * But it's BUG to call kmalloc() against offline node. |
| 4335 | * |
| 4336 | * TODO: this routine can waste much memory for nodes which will |
| 4337 | * never be onlined. It's better to use memory hotplug callback |
| 4338 | * function. |
| 4339 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4340 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4341 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4342 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4343 | if (!pn) |
| 4344 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4345 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4346 | mem->info.nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4347 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4348 | mz = &pn->zoneinfo[zone]; |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 4349 | for_each_lru(l) |
| 4350 | INIT_LIST_HEAD(&mz->lists[l]); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4351 | mz->usage_in_excess = 0; |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 4352 | mz->on_tree = false; |
| 4353 | mz->mem = mem; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4354 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4355 | return 0; |
| 4356 | } |
| 4357 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4358 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 4359 | { |
| 4360 | kfree(mem->info.nodeinfo[node]); |
| 4361 | } |
| 4362 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4363 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4364 | { |
| 4365 | struct mem_cgroup *mem; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4366 | int size = sizeof(struct mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4367 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4368 | /* Can be very big if MAX_NUMNODES is very big */ |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4369 | if (size < PAGE_SIZE) |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4370 | mem = kzalloc(size, GFP_KERNEL); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4371 | else |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4372 | mem = vzalloc(size); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4373 | |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4374 | if (!mem) |
| 4375 | return NULL; |
| 4376 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4377 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4378 | if (!mem->stat) |
| 4379 | goto out_free; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 4380 | spin_lock_init(&mem->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4381 | return mem; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4382 | |
| 4383 | out_free: |
| 4384 | if (size < PAGE_SIZE) |
| 4385 | kfree(mem); |
| 4386 | else |
| 4387 | vfree(mem); |
| 4388 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4389 | } |
| 4390 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4391 | /* |
| 4392 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4393 | * (scanning all at force_empty is too costly...) |
| 4394 | * |
| 4395 | * Instead of clearing all references at force_empty, we remember |
| 4396 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4397 | * it goes down to 0. |
| 4398 | * |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4399 | * Removal of cgroup itself succeeds regardless of refs from swap. |
| 4400 | */ |
| 4401 | |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4402 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4403 | { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4404 | int node; |
| 4405 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4406 | mem_cgroup_remove_from_trees(mem); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4407 | free_css_id(&mem_cgroup_subsys, &mem->css); |
| 4408 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4409 | for_each_node_state(node, N_POSSIBLE) |
| 4410 | free_mem_cgroup_per_zone_info(mem, node); |
| 4411 | |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 4412 | free_percpu(mem->stat); |
| 4413 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4414 | kfree(mem); |
| 4415 | else |
| 4416 | vfree(mem); |
| 4417 | } |
| 4418 | |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4419 | static void mem_cgroup_get(struct mem_cgroup *mem) |
| 4420 | { |
| 4421 | atomic_inc(&mem->refcnt); |
| 4422 | } |
| 4423 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4424 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4425 | { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4426 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4427 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4428 | __mem_cgroup_free(mem); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4429 | if (parent) |
| 4430 | mem_cgroup_put(parent); |
| 4431 | } |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4432 | } |
| 4433 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4434 | static void mem_cgroup_put(struct mem_cgroup *mem) |
| 4435 | { |
| 4436 | __mem_cgroup_put(mem, 1); |
| 4437 | } |
| 4438 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4439 | /* |
| 4440 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4441 | */ |
| 4442 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) |
| 4443 | { |
| 4444 | if (!mem->res.parent) |
| 4445 | return NULL; |
| 4446 | return mem_cgroup_from_res_counter(mem->res.parent, res); |
| 4447 | } |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4448 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4449 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
| 4450 | static void __init enable_swap_cgroup(void) |
| 4451 | { |
Hirokazu Takahashi | f8d6654 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 4452 | if (!mem_cgroup_disabled() && really_do_swap_account) |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4453 | do_swap_account = 1; |
| 4454 | } |
| 4455 | #else |
| 4456 | static void __init enable_swap_cgroup(void) |
| 4457 | { |
| 4458 | } |
| 4459 | #endif |
| 4460 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4461 | static int mem_cgroup_soft_limit_tree_init(void) |
| 4462 | { |
| 4463 | struct mem_cgroup_tree_per_node *rtpn; |
| 4464 | struct mem_cgroup_tree_per_zone *rtpz; |
| 4465 | int tmp, node, zone; |
| 4466 | |
| 4467 | for_each_node_state(node, N_POSSIBLE) { |
| 4468 | tmp = node; |
| 4469 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4470 | tmp = -1; |
| 4471 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); |
| 4472 | if (!rtpn) |
| 4473 | return 1; |
| 4474 | |
| 4475 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 4476 | |
| 4477 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4478 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 4479 | rtpz->rb_root = RB_ROOT; |
| 4480 | spin_lock_init(&rtpz->lock); |
| 4481 | } |
| 4482 | } |
| 4483 | return 0; |
| 4484 | } |
| 4485 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 4486 | static struct cgroup_subsys_state * __ref |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4487 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
| 4488 | { |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4489 | struct mem_cgroup *mem, *parent; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4490 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4491 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4492 | |
Jan Blunck | c8dad2b | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4493 | mem = mem_cgroup_alloc(); |
| 4494 | if (!mem) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4495 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4496 | |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4497 | for_each_node_state(node, N_POSSIBLE) |
| 4498 | if (alloc_mem_cgroup_per_zone_info(mem, node)) |
| 4499 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4500 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4501 | /* root ? */ |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4502 | if (cont->parent == NULL) { |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4503 | int cpu; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4504 | enable_swap_cgroup(); |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4505 | parent = NULL; |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 4506 | root_mem_cgroup = mem; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4507 | if (mem_cgroup_soft_limit_tree_init()) |
| 4508 | goto free_out; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 4509 | for_each_possible_cpu(cpu) { |
| 4510 | struct memcg_stock_pcp *stock = |
| 4511 | &per_cpu(memcg_stock, cpu); |
| 4512 | INIT_WORK(&stock->work, drain_local_stock); |
| 4513 | } |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 4514 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4515 | } else { |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4516 | parent = mem_cgroup_from_cont(cont->parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4517 | mem->use_hierarchy = parent->use_hierarchy; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4518 | mem->oom_kill_disable = parent->oom_kill_disable; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4519 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4520 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4521 | if (parent && parent->use_hierarchy) { |
| 4522 | res_counter_init(&mem->res, &parent->res); |
| 4523 | res_counter_init(&mem->memsw, &parent->memsw); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4524 | /* |
| 4525 | * We increment refcnt of the parent to ensure that we can |
| 4526 | * safely access it on res_counter_charge/uncharge. |
| 4527 | * This refcnt will be decremented when freeing this |
| 4528 | * mem_cgroup(see mem_cgroup_put). |
| 4529 | */ |
| 4530 | mem_cgroup_get(parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4531 | } else { |
| 4532 | res_counter_init(&mem->res, NULL); |
| 4533 | res_counter_init(&mem->memsw, NULL); |
| 4534 | } |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4535 | mem->last_scanned_child = 0; |
KOSAKI Motohiro | 2733c06 | 2009-01-07 18:08:23 -0800 | [diff] [blame] | 4536 | spin_lock_init(&mem->reclaim_param_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4537 | INIT_LIST_HEAD(&mem->oom_notify); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 4538 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4539 | if (parent) |
| 4540 | mem->swappiness = get_swappiness(parent); |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4541 | atomic_set(&mem->refcnt, 1); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4542 | mem->move_charge_at_immigrate = 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4543 | mutex_init(&mem->thresholds_lock); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4544 | return &mem->css; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4545 | free_out: |
KAMEZAWA Hiroyuki | a7ba0ee | 2009-01-07 18:08:32 -0800 | [diff] [blame] | 4546 | __mem_cgroup_free(mem); |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 4547 | root_mem_cgroup = NULL; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4548 | return ERR_PTR(error); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4549 | } |
| 4550 | |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4551 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4552 | struct cgroup *cont) |
| 4553 | { |
| 4554 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4555 | |
| 4556 | return mem_cgroup_force_empty(mem, false); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4557 | } |
| 4558 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4559 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
| 4560 | struct cgroup *cont) |
| 4561 | { |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4562 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4563 | |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4564 | mem_cgroup_put(mem); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4565 | } |
| 4566 | |
| 4567 | static int mem_cgroup_populate(struct cgroup_subsys *ss, |
| 4568 | struct cgroup *cont) |
| 4569 | { |
KAMEZAWA Hiroyuki | 8c7c6e3 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4570 | int ret; |
| 4571 | |
| 4572 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, |
| 4573 | ARRAY_SIZE(mem_cgroup_files)); |
| 4574 | |
| 4575 | if (!ret) |
| 4576 | ret = register_memsw_files(cont, ss); |
| 4577 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4578 | } |
| 4579 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4580 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4581 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4582 | #define PRECHARGE_COUNT_AT_ONCE 256 |
| 4583 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4584 | { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4585 | int ret = 0; |
| 4586 | int batch_count = PRECHARGE_COUNT_AT_ONCE; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4587 | struct mem_cgroup *mem = mc.to; |
| 4588 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4589 | if (mem_cgroup_is_root(mem)) { |
| 4590 | mc.precharge += count; |
| 4591 | /* we don't need css_get for root */ |
| 4592 | return ret; |
| 4593 | } |
| 4594 | /* try to charge at once */ |
| 4595 | if (count > 1) { |
| 4596 | struct res_counter *dummy; |
| 4597 | /* |
| 4598 | * "mem" cannot be under rmdir() because we've already checked |
| 4599 | * by cgroup_lock_live_cgroup() that it is not removed and we |
| 4600 | * are still under the same cgroup_mutex. So we can postpone |
| 4601 | * css_get(). |
| 4602 | */ |
| 4603 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) |
| 4604 | goto one_by_one; |
| 4605 | if (do_swap_account && res_counter_charge(&mem->memsw, |
| 4606 | PAGE_SIZE * count, &dummy)) { |
| 4607 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
| 4608 | goto one_by_one; |
| 4609 | } |
| 4610 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4611 | return ret; |
| 4612 | } |
| 4613 | one_by_one: |
| 4614 | /* fall back to one by one charge */ |
| 4615 | while (count--) { |
| 4616 | if (signal_pending(current)) { |
| 4617 | ret = -EINTR; |
| 4618 | break; |
| 4619 | } |
| 4620 | if (!batch_count--) { |
| 4621 | batch_count = PRECHARGE_COUNT_AT_ONCE; |
| 4622 | cond_resched(); |
| 4623 | } |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 4624 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
| 4625 | PAGE_SIZE); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4626 | if (ret || !mem) |
| 4627 | /* mem_cgroup_clear_mc() will do uncharge later */ |
| 4628 | return -ENOMEM; |
| 4629 | mc.precharge++; |
| 4630 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4631 | return ret; |
| 4632 | } |
| 4633 | |
| 4634 | /** |
| 4635 | * is_target_pte_for_mc - check a pte whether it is valid for move charge |
| 4636 | * @vma: the vma the pte to be checked belongs |
| 4637 | * @addr: the address corresponding to the pte to be checked |
| 4638 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4639 | * @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] | 4640 | * |
| 4641 | * Returns |
| 4642 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4643 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4644 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4645 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4646 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4647 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4648 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4649 | * |
| 4650 | * Called with pte lock held. |
| 4651 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4652 | union mc_target { |
| 4653 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4654 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4655 | }; |
| 4656 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4657 | enum mc_target_type { |
| 4658 | MC_TARGET_NONE, /* not used */ |
| 4659 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4660 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4661 | }; |
| 4662 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4663 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4664 | unsigned long addr, pte_t ptent) |
| 4665 | { |
| 4666 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4667 | |
| 4668 | if (!page || !page_mapped(page)) |
| 4669 | return NULL; |
| 4670 | if (PageAnon(page)) { |
| 4671 | /* we don't move shared anon */ |
| 4672 | if (!move_anon() || page_mapcount(page) > 2) |
| 4673 | return NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4674 | } else if (!move_file()) |
| 4675 | /* we ignore mapcount for file pages */ |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4676 | return NULL; |
| 4677 | if (!get_page_unless_zero(page)) |
| 4678 | return NULL; |
| 4679 | |
| 4680 | return page; |
| 4681 | } |
| 4682 | |
| 4683 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4684 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4685 | { |
| 4686 | int usage_count; |
| 4687 | struct page *page = NULL; |
| 4688 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4689 | |
| 4690 | if (!move_anon() || non_swap_entry(ent)) |
| 4691 | return NULL; |
| 4692 | usage_count = mem_cgroup_count_swap_user(ent, &page); |
| 4693 | if (usage_count > 1) { /* we don't move shared anon */ |
| 4694 | if (page) |
| 4695 | put_page(page); |
| 4696 | return NULL; |
| 4697 | } |
| 4698 | if (do_swap_account) |
| 4699 | entry->val = ent.val; |
| 4700 | |
| 4701 | return page; |
| 4702 | } |
| 4703 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4704 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4705 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4706 | { |
| 4707 | struct page *page = NULL; |
| 4708 | struct inode *inode; |
| 4709 | struct address_space *mapping; |
| 4710 | pgoff_t pgoff; |
| 4711 | |
| 4712 | if (!vma->vm_file) /* anonymous vma */ |
| 4713 | return NULL; |
| 4714 | if (!move_file()) |
| 4715 | return NULL; |
| 4716 | |
| 4717 | inode = vma->vm_file->f_path.dentry->d_inode; |
| 4718 | mapping = vma->vm_file->f_mapping; |
| 4719 | if (pte_none(ptent)) |
| 4720 | pgoff = linear_page_index(vma, addr); |
| 4721 | else /* pte_file(ptent) is true */ |
| 4722 | pgoff = pte_to_pgoff(ptent); |
| 4723 | |
| 4724 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
| 4725 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ |
| 4726 | page = find_get_page(mapping, pgoff); |
| 4727 | } else { /* shmem/tmpfs file. we should take account of swap too. */ |
| 4728 | swp_entry_t ent; |
| 4729 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); |
| 4730 | if (do_swap_account) |
| 4731 | entry->val = ent.val; |
| 4732 | } |
| 4733 | |
| 4734 | return page; |
| 4735 | } |
| 4736 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4737 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
| 4738 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4739 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4740 | struct page *page = NULL; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4741 | struct page_cgroup *pc; |
| 4742 | int ret = 0; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4743 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4744 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4745 | if (pte_present(ptent)) |
| 4746 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4747 | else if (is_swap_pte(ptent)) |
| 4748 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4749 | else if (pte_none(ptent) || pte_file(ptent)) |
| 4750 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4751 | |
| 4752 | if (!page && !ent.val) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4753 | return 0; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4754 | if (page) { |
| 4755 | pc = lookup_page_cgroup(page); |
| 4756 | /* |
| 4757 | * Do only loose check w/o page_cgroup lock. |
| 4758 | * mem_cgroup_move_account() checks the pc is valid or not under |
| 4759 | * the lock. |
| 4760 | */ |
| 4761 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 4762 | ret = MC_TARGET_PAGE; |
| 4763 | if (target) |
| 4764 | target->page = page; |
| 4765 | } |
| 4766 | if (!ret || !target) |
| 4767 | put_page(page); |
| 4768 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4769 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4770 | if (ent.val && !ret && |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4771 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
| 4772 | ret = MC_TARGET_SWAP; |
| 4773 | if (target) |
| 4774 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4775 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4776 | return ret; |
| 4777 | } |
| 4778 | |
| 4779 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4780 | unsigned long addr, unsigned long end, |
| 4781 | struct mm_walk *walk) |
| 4782 | { |
| 4783 | struct vm_area_struct *vma = walk->private; |
| 4784 | pte_t *pte; |
| 4785 | spinlock_t *ptl; |
| 4786 | |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 4787 | split_huge_page_pmd(walk->mm, pmd); |
| 4788 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4789 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4790 | for (; addr != end; pte++, addr += PAGE_SIZE) |
| 4791 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) |
| 4792 | mc.precharge++; /* increment precharge temporarily */ |
| 4793 | pte_unmap_unlock(pte - 1, ptl); |
| 4794 | cond_resched(); |
| 4795 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4796 | return 0; |
| 4797 | } |
| 4798 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4799 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4800 | { |
| 4801 | unsigned long precharge; |
| 4802 | struct vm_area_struct *vma; |
| 4803 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4804 | down_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4805 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 4806 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4807 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4808 | .mm = mm, |
| 4809 | .private = vma, |
| 4810 | }; |
| 4811 | if (is_vm_hugetlb_page(vma)) |
| 4812 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4813 | walk_page_range(vma->vm_start, vma->vm_end, |
| 4814 | &mem_cgroup_count_precharge_walk); |
| 4815 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4816 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4817 | |
| 4818 | precharge = mc.precharge; |
| 4819 | mc.precharge = 0; |
| 4820 | |
| 4821 | return precharge; |
| 4822 | } |
| 4823 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4824 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4825 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4826 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 4827 | |
| 4828 | VM_BUG_ON(mc.moving_task); |
| 4829 | mc.moving_task = current; |
| 4830 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4831 | } |
| 4832 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4833 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 4834 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4835 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4836 | struct mem_cgroup *from = mc.from; |
| 4837 | struct mem_cgroup *to = mc.to; |
| 4838 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4839 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4840 | if (mc.precharge) { |
| 4841 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); |
| 4842 | mc.precharge = 0; |
| 4843 | } |
| 4844 | /* |
| 4845 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4846 | * we must uncharge here. |
| 4847 | */ |
| 4848 | if (mc.moved_charge) { |
| 4849 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
| 4850 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4851 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4852 | /* we must fixup refcnts and charges */ |
| 4853 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4854 | /* uncharge swap account from the old cgroup */ |
| 4855 | if (!mem_cgroup_is_root(mc.from)) |
| 4856 | res_counter_uncharge(&mc.from->memsw, |
| 4857 | PAGE_SIZE * mc.moved_swap); |
| 4858 | __mem_cgroup_put(mc.from, mc.moved_swap); |
| 4859 | |
| 4860 | if (!mem_cgroup_is_root(mc.to)) { |
| 4861 | /* |
| 4862 | * we charged both to->res and to->memsw, so we should |
| 4863 | * uncharge to->res. |
| 4864 | */ |
| 4865 | res_counter_uncharge(&mc.to->res, |
| 4866 | PAGE_SIZE * mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4867 | } |
| 4868 | /* we've already done mem_cgroup_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4869 | mc.moved_swap = 0; |
| 4870 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4871 | memcg_oom_recover(from); |
| 4872 | memcg_oom_recover(to); |
| 4873 | wake_up_all(&mc.waitq); |
| 4874 | } |
| 4875 | |
| 4876 | static void mem_cgroup_clear_mc(void) |
| 4877 | { |
| 4878 | struct mem_cgroup *from = mc.from; |
| 4879 | |
| 4880 | /* |
| 4881 | * we must clear moving_task before waking up waiters at the end of |
| 4882 | * task migration. |
| 4883 | */ |
| 4884 | mc.moving_task = NULL; |
| 4885 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4886 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4887 | mc.from = NULL; |
| 4888 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4889 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 4890 | mem_cgroup_end_move(from); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4891 | } |
| 4892 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4893 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
| 4894 | struct cgroup *cgroup, |
| 4895 | struct task_struct *p, |
| 4896 | bool threadgroup) |
| 4897 | { |
| 4898 | int ret = 0; |
| 4899 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); |
| 4900 | |
| 4901 | if (mem->move_charge_at_immigrate) { |
| 4902 | struct mm_struct *mm; |
| 4903 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 4904 | |
| 4905 | VM_BUG_ON(from == mem); |
| 4906 | |
| 4907 | mm = get_task_mm(p); |
| 4908 | if (!mm) |
| 4909 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4910 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4911 | if (mm->owner == p) { |
| 4912 | VM_BUG_ON(mc.from); |
| 4913 | VM_BUG_ON(mc.to); |
| 4914 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4915 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4916 | VM_BUG_ON(mc.moved_swap); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 4917 | mem_cgroup_start_move(from); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4918 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4919 | mc.from = from; |
| 4920 | mc.to = mem; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4921 | spin_unlock(&mc.lock); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4922 | /* We set mc.moving_task later */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4923 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4924 | ret = mem_cgroup_precharge_mc(mm); |
| 4925 | if (ret) |
| 4926 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4927 | } |
| 4928 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4929 | } |
| 4930 | return ret; |
| 4931 | } |
| 4932 | |
| 4933 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, |
| 4934 | struct cgroup *cgroup, |
| 4935 | struct task_struct *p, |
| 4936 | bool threadgroup) |
| 4937 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4938 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4939 | } |
| 4940 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4941 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 4942 | unsigned long addr, unsigned long end, |
| 4943 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4944 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4945 | int ret = 0; |
| 4946 | struct vm_area_struct *vma = walk->private; |
| 4947 | pte_t *pte; |
| 4948 | spinlock_t *ptl; |
| 4949 | |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 4950 | split_huge_page_pmd(walk->mm, pmd); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4951 | retry: |
| 4952 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4953 | for (; addr != end; addr += PAGE_SIZE) { |
| 4954 | pte_t ptent = *(pte++); |
| 4955 | union mc_target target; |
| 4956 | int type; |
| 4957 | struct page *page; |
| 4958 | struct page_cgroup *pc; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4959 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4960 | |
| 4961 | if (!mc.precharge) |
| 4962 | break; |
| 4963 | |
| 4964 | type = is_target_pte_for_mc(vma, addr, ptent, &target); |
| 4965 | switch (type) { |
| 4966 | case MC_TARGET_PAGE: |
| 4967 | page = target.page; |
| 4968 | if (isolate_lru_page(page)) |
| 4969 | goto put; |
| 4970 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4971 | if (!mem_cgroup_move_account(pc, |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 4972 | mc.from, mc.to, false, PAGE_SIZE)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4973 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4974 | /* we uncharge from mc.from later. */ |
| 4975 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4976 | } |
| 4977 | putback_lru_page(page); |
| 4978 | put: /* is_target_pte_for_mc() gets the page */ |
| 4979 | put_page(page); |
| 4980 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4981 | case MC_TARGET_SWAP: |
| 4982 | ent = target.ent; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4983 | if (!mem_cgroup_move_swap_account(ent, |
| 4984 | mc.from, mc.to, false)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4985 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4986 | /* we fixup refcnts and charges later. */ |
| 4987 | mc.moved_swap++; |
| 4988 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4989 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4990 | default: |
| 4991 | break; |
| 4992 | } |
| 4993 | } |
| 4994 | pte_unmap_unlock(pte - 1, ptl); |
| 4995 | cond_resched(); |
| 4996 | |
| 4997 | if (addr != end) { |
| 4998 | /* |
| 4999 | * We have consumed all precharges we got in can_attach(). |
| 5000 | * We try charge one by one, but don't do any additional |
| 5001 | * charges to mc.to if we have failed in charge once in attach() |
| 5002 | * phase. |
| 5003 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5004 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5005 | if (!ret) |
| 5006 | goto retry; |
| 5007 | } |
| 5008 | |
| 5009 | return ret; |
| 5010 | } |
| 5011 | |
| 5012 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 5013 | { |
| 5014 | struct vm_area_struct *vma; |
| 5015 | |
| 5016 | lru_add_drain_all(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5017 | retry: |
| 5018 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
| 5019 | /* |
| 5020 | * Someone who are holding the mmap_sem might be waiting in |
| 5021 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 5022 | * and retry. Because we cancel precharges, we might not be able |
| 5023 | * to move enough charges, but moving charge is a best-effort |
| 5024 | * feature anyway, so it wouldn't be a big problem. |
| 5025 | */ |
| 5026 | __mem_cgroup_clear_mc(); |
| 5027 | cond_resched(); |
| 5028 | goto retry; |
| 5029 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5030 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 5031 | int ret; |
| 5032 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 5033 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 5034 | .mm = mm, |
| 5035 | .private = vma, |
| 5036 | }; |
| 5037 | if (is_vm_hugetlb_page(vma)) |
| 5038 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5039 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
| 5040 | &mem_cgroup_move_charge_walk); |
| 5041 | if (ret) |
| 5042 | /* |
| 5043 | * means we have consumed all precharges and failed in |
| 5044 | * doing additional charge. Just abandon here. |
| 5045 | */ |
| 5046 | break; |
| 5047 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5048 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5049 | } |
| 5050 | |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5051 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 5052 | struct cgroup *cont, |
| 5053 | struct cgroup *old_cont, |
Ben Blum | be367d0 | 2009-09-23 15:56:31 -0700 | [diff] [blame] | 5054 | struct task_struct *p, |
| 5055 | bool threadgroup) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5056 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5057 | struct mm_struct *mm; |
| 5058 | |
| 5059 | if (!mc.to) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5060 | /* no need to move charge */ |
| 5061 | return; |
| 5062 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5063 | mm = get_task_mm(p); |
| 5064 | if (mm) { |
| 5065 | mem_cgroup_move_charge(mm); |
| 5066 | mmput(mm); |
| 5067 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5068 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5069 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5070 | #else /* !CONFIG_MMU */ |
| 5071 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
| 5072 | struct cgroup *cgroup, |
| 5073 | struct task_struct *p, |
| 5074 | bool threadgroup) |
| 5075 | { |
| 5076 | return 0; |
| 5077 | } |
| 5078 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, |
| 5079 | struct cgroup *cgroup, |
| 5080 | struct task_struct *p, |
| 5081 | bool threadgroup) |
| 5082 | { |
| 5083 | } |
| 5084 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 5085 | struct cgroup *cont, |
| 5086 | struct cgroup *old_cont, |
| 5087 | struct task_struct *p, |
| 5088 | bool threadgroup) |
| 5089 | { |
| 5090 | } |
| 5091 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5092 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5093 | struct cgroup_subsys mem_cgroup_subsys = { |
| 5094 | .name = "memory", |
| 5095 | .subsys_id = mem_cgroup_subsys_id, |
| 5096 | .create = mem_cgroup_create, |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5097 | .pre_destroy = mem_cgroup_pre_destroy, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5098 | .destroy = mem_cgroup_destroy, |
| 5099 | .populate = mem_cgroup_populate, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5100 | .can_attach = mem_cgroup_can_attach, |
| 5101 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5102 | .attach = mem_cgroup_move_task, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5103 | .early_init = 0, |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 5104 | .use_id = 1, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5105 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5106 | |
| 5107 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5108 | static int __init enable_swap_account(char *s) |
| 5109 | { |
| 5110 | /* consider enabled if no parameter or 1 is given */ |
Michal Hocko | fceda1b | 2011-02-01 15:52:30 -0800 | [diff] [blame] | 5111 | if (!(*s) || !strcmp(s, "=1")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5112 | really_do_swap_account = 1; |
Michal Hocko | fceda1b | 2011-02-01 15:52:30 -0800 | [diff] [blame] | 5113 | else if (!strcmp(s, "=0")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 5114 | really_do_swap_account = 0; |
| 5115 | return 1; |
| 5116 | } |
| 5117 | __setup("swapaccount", enable_swap_account); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5118 | |
| 5119 | static int __init disable_swap_account(char *s) |
| 5120 | { |
Michal Hocko | 552b372 | 2011-02-01 15:52:31 -0800 | [diff] [blame] | 5121 | printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n"); |
Michal Hocko | fceda1b | 2011-02-01 15:52:30 -0800 | [diff] [blame] | 5122 | enable_swap_account("=0"); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5123 | return 1; |
| 5124 | } |
| 5125 | __setup("noswapaccount", disable_swap_account); |
| 5126 | #endif |