blob: d436634ac54866dccd2d7bb02be14302c9569bcd [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080013 * 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 Emelianov78fb7462008-02-07 00:13:51 -080027#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080028#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080029#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080030#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070035#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040036#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080037#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070038#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080041#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080043#include <linux/eventfd.h>
44#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080045#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080046#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070047#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070048#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080050#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070051#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080052#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000053#include <net/sock.h>
54#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080055
Balbir Singh8697d332008-02-07 00:13:59 -080056#include <asm/uaccess.h>
57
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070058#include <trace/events/vmscan.h>
59
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070060struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070061#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070062struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070065/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080066int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080067
68/* for remember boot option*/
69#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70static int really_do_swap_account __initdata = 1;
71#else
72static int really_do_swap_account __initdata = 0;
73#endif
74
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075#else
76#define do_swap_account (0)
77#endif
78
79
Balbir Singh8cdea7c2008-02-07 00:13:50 -080080/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080081 * Statistics for memory cgroup.
82 */
83enum mem_cgroup_stat_index {
84 /*
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 */
87 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070088 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080089 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070090 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -070091 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080092 MEM_CGROUP_STAT_NSTATS,
93};
94
Johannes Weinere9f89742011-03-23 16:42:37 -070095enum mem_cgroup_events_index {
96 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
97 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
98 MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
Ying Han456f9982011-05-26 16:25:38 -070099 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
100 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700101 MEM_CGROUP_EVENTS_NSTATS,
102};
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700103/*
104 * Per memcg event counter is incremented at every pagein/pageout. With THP,
105 * it will be incremated by the number of pages. This counter is used for
106 * for trigger some periodic events. This is straightforward and better
107 * than using jiffies etc. to handle periodic memcg event.
108 */
109enum mem_cgroup_events_target {
110 MEM_CGROUP_TARGET_THRESH,
111 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700112 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700113 MEM_CGROUP_NTARGETS,
114};
115#define THRESHOLDS_EVENTS_TARGET (128)
116#define SOFTLIMIT_EVENTS_TARGET (1024)
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700117#define NUMAINFO_EVENTS_TARGET (1024)
Johannes Weinere9f89742011-03-23 16:42:37 -0700118
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800119struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700120 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700121 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700122 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800123};
124
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800125struct mem_cgroup_reclaim_iter {
126 /* css_id of the last scanned hierarchy member */
127 int position;
128 /* scan generation, increased every round-trip */
129 unsigned int generation;
130};
131
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800132/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800133 * per-zone information in memory controller.
134 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800135struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800136 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700137 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800138
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800139 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
140
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800141 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700142 struct rb_node tree_node; /* RB tree node */
143 unsigned long long usage_in_excess;/* Set to the value by which */
144 /* the soft limit is exceeded*/
145 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700146 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700147 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800148};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800149
150struct mem_cgroup_per_node {
151 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
152};
153
154struct mem_cgroup_lru_info {
155 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
156};
157
158/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700159 * Cgroups above their limits are maintained in a RB-Tree, independent of
160 * their hierarchy representation
161 */
162
163struct mem_cgroup_tree_per_zone {
164 struct rb_root rb_root;
165 spinlock_t lock;
166};
167
168struct mem_cgroup_tree_per_node {
169 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
170};
171
172struct mem_cgroup_tree {
173 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
174};
175
176static struct mem_cgroup_tree soft_limit_tree __read_mostly;
177
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800178struct mem_cgroup_threshold {
179 struct eventfd_ctx *eventfd;
180 u64 threshold;
181};
182
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700183/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800184struct mem_cgroup_threshold_ary {
185 /* An array index points to threshold just below usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700186 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800187 /* Size of entries[] */
188 unsigned int size;
189 /* Array of thresholds */
190 struct mem_cgroup_threshold entries[0];
191};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700192
193struct mem_cgroup_thresholds {
194 /* Primary thresholds array */
195 struct mem_cgroup_threshold_ary *primary;
196 /*
197 * Spare threshold array.
198 * This is needed to make mem_cgroup_unregister_event() "never fail".
199 * It must be able to store at least primary->size - 1 entries.
200 */
201 struct mem_cgroup_threshold_ary *spare;
202};
203
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700204/* for OOM */
205struct mem_cgroup_eventfd_list {
206 struct list_head list;
207 struct eventfd_ctx *eventfd;
208};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700210static void mem_cgroup_threshold(struct mem_cgroup *memcg);
211static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800212
Balbir Singhf64c3f52009-09-23 15:56:37 -0700213/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800214 * The memory controller data structure. The memory controller controls both
215 * page cache and RSS per cgroup. We would eventually like to provide
216 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
217 * to help the administrator determine what knobs to tune.
218 *
219 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800220 * we hit the water mark. May be even add a low water mark, such that
221 * no reclaim occurs from a cgroup at it's low water mark, this is
222 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800223 */
224struct mem_cgroup {
225 struct cgroup_subsys_state css;
226 /*
227 * the counter to account for memory usage
228 */
229 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700230
231 union {
232 /*
233 * the counter to account for mem+swap usage.
234 */
235 struct res_counter memsw;
236
237 /*
238 * rcu_freeing is used only when freeing struct mem_cgroup,
239 * so put it into a union to avoid wasting more memory.
240 * It must be disjoint from the css field. It could be
241 * in a union with the res field, but res plays a much
242 * larger part in mem_cgroup life than memsw, and might
243 * be of interest, even at time of free, when debugging.
244 * So share rcu_head with the less interesting memsw.
245 */
246 struct rcu_head rcu_freeing;
247 /*
248 * But when using vfree(), that cannot be done at
249 * interrupt time, so we must then queue the work.
250 */
251 struct work_struct work_freeing;
252 };
253
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800254 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800255 * Per cgroup active and inactive list, similar to the
256 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800257 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800258 struct mem_cgroup_lru_info info;
Ying Han889976d2011-05-26 16:25:33 -0700259 int last_scanned_node;
260#if MAX_NUMNODES > 1
261 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700262 atomic_t numainfo_events;
263 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700264#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800265 /*
266 * Should the accounting and control be hierarchical, per subtree?
267 */
268 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700269
270 bool oom_lock;
271 atomic_t under_oom;
272
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800273 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800274
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700275 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700276 /* OOM-Killer disable */
277 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800278
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700279 /* set when res.limit == memsw.limit */
280 bool memsw_is_minimum;
281
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800282 /* protect arrays of thresholds */
283 struct mutex thresholds_lock;
284
285 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700286 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700287
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800288 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700289 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700290
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700291 /* For oom notifier event fd */
292 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700293
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800294 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800295 * Should we move charges of a task when a task is moved into this
296 * mem_cgroup ? And what type of charges should we move ?
297 */
298 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800299 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700300 * set > 0 if pages under this cgroup are moving to other cgroup.
301 */
302 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700303 /* taken only while moving_account > 0 */
304 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700305 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800306 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800307 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800308 struct mem_cgroup_stat_cpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700309 /*
310 * used when a cpu is offlined or other synchronizations
311 * See mem_cgroup_read_stat().
312 */
313 struct mem_cgroup_stat_cpu nocpu_base;
314 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000315
316#ifdef CONFIG_INET
317 struct tcp_memcontrol tcp_mem;
318#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800319};
320
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800321/* Stuffs for move charges at task migration. */
322/*
323 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
324 * left-shifted bitmap of these types.
325 */
326enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800327 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700328 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800329 NR_MOVE_TYPE,
330};
331
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800332/* "mc" and its members are protected by cgroup_mutex */
333static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800334 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800335 struct mem_cgroup *from;
336 struct mem_cgroup *to;
337 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800338 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800339 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800340 struct task_struct *moving_task; /* a task moving charges */
341 wait_queue_head_t waitq; /* a waitq for other context */
342} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700343 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800344 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
345};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800346
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700347static bool move_anon(void)
348{
349 return test_bit(MOVE_CHARGE_TYPE_ANON,
350 &mc.to->move_charge_at_immigrate);
351}
352
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700353static bool move_file(void)
354{
355 return test_bit(MOVE_CHARGE_TYPE_FILE,
356 &mc.to->move_charge_at_immigrate);
357}
358
Balbir Singh4e416952009-09-23 15:56:39 -0700359/*
360 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
361 * limit reclaim to prevent infinite loops, if they ever occur.
362 */
363#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
364#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
365
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800366enum charge_type {
367 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
368 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700369 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700370 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800371 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700372 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700373 NR_CHARGE_TYPE,
374};
375
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800376/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000377#define _MEM (0)
378#define _MEMSWAP (1)
379#define _OOM_TYPE (2)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800380#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
381#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
382#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700383/* Used for OOM nofiier */
384#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800385
Balbir Singh75822b42009-09-23 15:56:38 -0700386/*
387 * Reclaim flags for mem_cgroup_hierarchical_reclaim
388 */
389#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
390#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
391#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
392#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
393
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700394static void mem_cgroup_get(struct mem_cgroup *memcg);
395static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000396
397/* Writing them here to avoid exposing memcg's inner layout */
398#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae1aab162011-12-11 21:47:03 +0000399#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000400#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000401
402static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
403void sock_update_memcg(struct sock *sk)
404{
Glauber Costa376be5f2012-01-20 04:57:14 +0000405 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000406 struct mem_cgroup *memcg;
407
408 BUG_ON(!sk->sk_prot->proto_cgroup);
409
Glauber Costaf3f511e2012-01-05 20:16:39 +0000410 /* Socket cloning can throw us here with sk_cgrp already
411 * filled. It won't however, necessarily happen from
412 * process context. So the test for root memcg given
413 * the current task's memcg won't help us in this case.
414 *
415 * Respecting the original socket's memcg is a better
416 * decision in this case.
417 */
418 if (sk->sk_cgrp) {
419 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
420 mem_cgroup_get(sk->sk_cgrp->memcg);
421 return;
422 }
423
Glauber Costae1aab162011-12-11 21:47:03 +0000424 rcu_read_lock();
425 memcg = mem_cgroup_from_task(current);
426 if (!mem_cgroup_is_root(memcg)) {
427 mem_cgroup_get(memcg);
428 sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg);
429 }
430 rcu_read_unlock();
431 }
432}
433EXPORT_SYMBOL(sock_update_memcg);
434
435void sock_release_memcg(struct sock *sk)
436{
Glauber Costa376be5f2012-01-20 04:57:14 +0000437 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000438 struct mem_cgroup *memcg;
439 WARN_ON(!sk->sk_cgrp->memcg);
440 memcg = sk->sk_cgrp->memcg;
441 mem_cgroup_put(memcg);
442 }
443}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000444
Glauber Costa319d3b92012-01-15 22:04:39 +0000445#ifdef CONFIG_INET
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000446struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
447{
448 if (!memcg || mem_cgroup_is_root(memcg))
449 return NULL;
450
451 return &memcg->tcp_mem.cg_proto;
452}
453EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000454#endif /* CONFIG_INET */
455#endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
456
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700457static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800458
Balbir Singhf64c3f52009-09-23 15:56:37 -0700459static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700460mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700461{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700462 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700463}
464
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700465struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100466{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700467 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100468}
469
Balbir Singhf64c3f52009-09-23 15:56:37 -0700470static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700471page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700472{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700473 int nid = page_to_nid(page);
474 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700475
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700476 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700477}
478
479static struct mem_cgroup_tree_per_zone *
480soft_limit_tree_node_zone(int nid, int zid)
481{
482 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
483}
484
485static struct mem_cgroup_tree_per_zone *
486soft_limit_tree_from_page(struct page *page)
487{
488 int nid = page_to_nid(page);
489 int zid = page_zonenum(page);
490
491 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
492}
493
494static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700495__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700496 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700497 struct mem_cgroup_tree_per_zone *mctz,
498 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700499{
500 struct rb_node **p = &mctz->rb_root.rb_node;
501 struct rb_node *parent = NULL;
502 struct mem_cgroup_per_zone *mz_node;
503
504 if (mz->on_tree)
505 return;
506
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700507 mz->usage_in_excess = new_usage_in_excess;
508 if (!mz->usage_in_excess)
509 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700510 while (*p) {
511 parent = *p;
512 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
513 tree_node);
514 if (mz->usage_in_excess < mz_node->usage_in_excess)
515 p = &(*p)->rb_left;
516 /*
517 * We can't avoid mem cgroups that are over their soft
518 * limit by the same amount
519 */
520 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
521 p = &(*p)->rb_right;
522 }
523 rb_link_node(&mz->tree_node, parent, p);
524 rb_insert_color(&mz->tree_node, &mctz->rb_root);
525 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700526}
527
528static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700529__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700530 struct mem_cgroup_per_zone *mz,
531 struct mem_cgroup_tree_per_zone *mctz)
532{
533 if (!mz->on_tree)
534 return;
535 rb_erase(&mz->tree_node, &mctz->rb_root);
536 mz->on_tree = false;
537}
538
539static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700540mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700541 struct mem_cgroup_per_zone *mz,
542 struct mem_cgroup_tree_per_zone *mctz)
543{
544 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700545 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700546 spin_unlock(&mctz->lock);
547}
548
Balbir Singhf64c3f52009-09-23 15:56:37 -0700549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700550static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700551{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700552 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700553 struct mem_cgroup_per_zone *mz;
554 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700555 int nid = page_to_nid(page);
556 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700557 mctz = soft_limit_tree_from_page(page);
558
559 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700560 * Necessary to update all ancestors when hierarchy is used.
561 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700562 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700563 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
564 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
565 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700566 /*
567 * We have to update the tree if mz is on RB-tree or
568 * mem is over its softlimit.
569 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700570 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700571 spin_lock(&mctz->lock);
572 /* if on-tree, remove it */
573 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700574 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700575 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700576 * Insert again. mz->usage_in_excess will be updated.
577 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700578 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700579 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700580 spin_unlock(&mctz->lock);
581 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700582 }
583}
584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700585static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700586{
587 int node, zone;
588 struct mem_cgroup_per_zone *mz;
589 struct mem_cgroup_tree_per_zone *mctz;
590
Bob Liu3ed28fa2012-01-12 17:19:04 -0800591 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700592 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700594 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700595 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700596 }
597 }
598}
599
Balbir Singh4e416952009-09-23 15:56:39 -0700600static struct mem_cgroup_per_zone *
601__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
602{
603 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700604 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700605
606retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700607 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700608 rightmost = rb_last(&mctz->rb_root);
609 if (!rightmost)
610 goto done; /* Nothing to reclaim from */
611
612 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
613 /*
614 * Remove the node now but someone else can add it back,
615 * we will to add it back at the end of reclaim to its correct
616 * position in the tree.
617 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700618 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
619 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
620 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700621 goto retry;
622done:
623 return mz;
624}
625
626static struct mem_cgroup_per_zone *
627mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
628{
629 struct mem_cgroup_per_zone *mz;
630
631 spin_lock(&mctz->lock);
632 mz = __mem_cgroup_largest_soft_limit_node(mctz);
633 spin_unlock(&mctz->lock);
634 return mz;
635}
636
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700637/*
638 * Implementation Note: reading percpu statistics for memcg.
639 *
640 * Both of vmstat[] and percpu_counter has threshold and do periodic
641 * synchronization to implement "quick" read. There are trade-off between
642 * reading cost and precision of value. Then, we may have a chance to implement
643 * a periodic synchronizion of counter in memcg's counter.
644 *
645 * But this _read() function is used for user interface now. The user accounts
646 * memory usage by memory cgroup and he _always_ requires exact value because
647 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
648 * have to visit all online cpus and make sum. So, for now, unnecessary
649 * synchronization is not implemented. (just implemented for cpu hotplug)
650 *
651 * If there are kernel internal actions which can make use of some not-exact
652 * value, and reading all cpu value can be performance bottleneck in some
653 * common workload, threashold and synchonization as vmstat[] should be
654 * implemented.
655 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700656static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700657 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800658{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700659 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800660 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800661
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700662 get_online_cpus();
663 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700664 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700665#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700666 spin_lock(&memcg->pcp_counter_lock);
667 val += memcg->nocpu_base.count[idx];
668 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700669#endif
670 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800671 return val;
672}
673
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700674static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700675 bool charge)
676{
677 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700678 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700679}
680
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700681static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700682 enum mem_cgroup_events_index idx)
683{
684 unsigned long val = 0;
685 int cpu;
686
687 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700689#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700690 spin_lock(&memcg->pcp_counter_lock);
691 val += memcg->nocpu_base.events[idx];
692 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700693#endif
694 return val;
695}
696
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700697static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700698 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800699{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800700 preempt_disable();
701
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700702 /*
703 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
704 * counted as CACHE even if it's on ANON LRU.
705 */
706 if (anon)
707 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700708 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800709 else
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700710 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700711 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700712
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800713 /* pagein of a big page is an event. So, ignore page size */
714 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700715 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800716 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700717 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800718 nr_pages = -nr_pages; /* for event */
719 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800720
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700721 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800722
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800723 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800724}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800725
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700726unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700727mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700728 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700729{
730 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700731 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700732 unsigned long ret = 0;
733
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700734 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700735
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700736 for_each_lru(lru) {
737 if (BIT(lru) & lru_mask)
738 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700739 }
740 return ret;
741}
742
743static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700744mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700745 int nid, unsigned int lru_mask)
746{
Ying Han889976d2011-05-26 16:25:33 -0700747 u64 total = 0;
748 int zid;
749
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700750 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700751 total += mem_cgroup_zone_nr_lru_pages(memcg,
752 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700753
Ying Han889976d2011-05-26 16:25:33 -0700754 return total;
755}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700756
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700757static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700758 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800759{
Ying Han889976d2011-05-26 16:25:33 -0700760 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800761 u64 total = 0;
762
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700763 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800765 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800766}
767
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800768static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
769 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800770{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700771 unsigned long val, next;
772
Steven Rostedt47994012011-11-02 13:38:33 -0700773 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
774 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700775 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800776 if ((long)next - (long)val < 0) {
777 switch (target) {
778 case MEM_CGROUP_TARGET_THRESH:
779 next = val + THRESHOLDS_EVENTS_TARGET;
780 break;
781 case MEM_CGROUP_TARGET_SOFTLIMIT:
782 next = val + SOFTLIMIT_EVENTS_TARGET;
783 break;
784 case MEM_CGROUP_TARGET_NUMAINFO:
785 next = val + NUMAINFO_EVENTS_TARGET;
786 break;
787 default:
788 break;
789 }
790 __this_cpu_write(memcg->stat->targets[target], next);
791 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700792 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800793 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800794}
795
796/*
797 * Check events in order.
798 *
799 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700800static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800801{
Steven Rostedt47994012011-11-02 13:38:33 -0700802 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800803 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800804 if (unlikely(mem_cgroup_event_ratelimit(memcg,
805 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800806 bool do_softlimit;
807 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800808
809 do_softlimit = mem_cgroup_event_ratelimit(memcg,
810 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700811#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800812 do_numainfo = mem_cgroup_event_ratelimit(memcg,
813 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700814#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800815 preempt_enable();
816
817 mem_cgroup_threshold(memcg);
818 if (unlikely(do_softlimit))
819 mem_cgroup_update_tree(memcg, page);
820#if MAX_NUMNODES > 1
821 if (unlikely(do_numainfo))
822 atomic_inc(&memcg->numainfo_events);
823#endif
824 } else
825 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800826}
827
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000828struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800829{
830 return container_of(cgroup_subsys_state(cont,
831 mem_cgroup_subsys_id), struct mem_cgroup,
832 css);
833}
834
Balbir Singhcf475ad2008-04-29 01:00:16 -0700835struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800836{
Balbir Singh31a78f22008-09-28 23:09:31 +0100837 /*
838 * mm_update_next_owner() may clear mm->owner to NULL
839 * if it races with swapoff, page migration, etc.
840 * So this can be called with p == NULL.
841 */
842 if (unlikely(!p))
843 return NULL;
844
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800845 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
846 struct mem_cgroup, css);
847}
848
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700849struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800850{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700851 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700852
853 if (!mm)
854 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800855 /*
856 * Because we have no locks, mm->owner's may be being moved to other
857 * cgroup. We use css_tryget() here even if this looks
858 * pessimistic (rather than adding locks here).
859 */
860 rcu_read_lock();
861 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700862 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
863 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800864 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700865 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800866 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700867 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800868}
869
Johannes Weiner56600482012-01-12 17:17:59 -0800870/**
871 * mem_cgroup_iter - iterate over memory cgroup hierarchy
872 * @root: hierarchy root
873 * @prev: previously returned memcg, NULL on first invocation
874 * @reclaim: cookie for shared reclaim walks, NULL for full walks
875 *
876 * Returns references to children of the hierarchy below @root, or
877 * @root itself, or %NULL after a full round-trip.
878 *
879 * Caller must pass the return value in @prev on subsequent
880 * invocations for reference counting, or use mem_cgroup_iter_break()
881 * to cancel a hierarchy walk before the round-trip is complete.
882 *
883 * Reclaimers can specify a zone and a priority level in @reclaim to
884 * divide up the memcgs in the hierarchy among all concurrent
885 * reclaimers operating on the same zone and priority.
886 */
887struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
888 struct mem_cgroup *prev,
889 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700890{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800891 struct mem_cgroup *memcg = NULL;
892 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700893
Johannes Weiner56600482012-01-12 17:17:59 -0800894 if (mem_cgroup_disabled())
895 return NULL;
896
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700897 if (!root)
898 root = root_mem_cgroup;
899
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800900 if (prev && !reclaim)
901 id = css_id(&prev->css);
902
903 if (prev && prev != root)
904 css_put(&prev->css);
905
906 if (!root->use_hierarchy && root != root_mem_cgroup) {
907 if (prev)
908 return NULL;
909 return root;
910 }
911
912 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800913 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800914 struct cgroup_subsys_state *css;
915
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800916 if (reclaim) {
917 int nid = zone_to_nid(reclaim->zone);
918 int zid = zone_idx(reclaim->zone);
919 struct mem_cgroup_per_zone *mz;
920
921 mz = mem_cgroup_zoneinfo(root, nid, zid);
922 iter = &mz->reclaim_iter[reclaim->priority];
923 if (prev && reclaim->generation != iter->generation)
924 return NULL;
925 id = iter->position;
926 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800927
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700928 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800929 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
930 if (css) {
931 if (css == &root->css || css_tryget(css))
932 memcg = container_of(css,
933 struct mem_cgroup, css);
934 } else
935 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700936 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700937
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800938 if (reclaim) {
939 iter->position = id;
940 if (!css)
941 iter->generation++;
942 else if (!prev && memcg)
943 reclaim->generation = iter->generation;
944 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800945
946 if (prev && !css)
947 return NULL;
948 }
949 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700950}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800951
Johannes Weiner56600482012-01-12 17:17:59 -0800952/**
953 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
954 * @root: hierarchy root
955 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
956 */
957void mem_cgroup_iter_break(struct mem_cgroup *root,
958 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800959{
960 if (!root)
961 root = root_mem_cgroup;
962 if (prev && prev != root)
963 css_put(&prev->css);
964}
965
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700966/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800967 * Iteration constructs for visiting all cgroups (under a tree). If
968 * loops are exited prematurely (break), mem_cgroup_iter_break() must
969 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700970 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800971#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800972 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800973 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800974 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700975
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800976#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800977 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800978 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800979 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700980
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700981static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700982{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700983 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700984}
985
Ying Han456f9982011-05-26 16:25:38 -0700986void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
987{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700988 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700989
990 if (!mm)
991 return;
992
993 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700994 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
995 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700996 goto out;
997
998 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -0700999 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001000 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1001 break;
1002 case PGMAJFAULT:
1003 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001004 break;
1005 default:
1006 BUG();
1007 }
1008out:
1009 rcu_read_unlock();
1010}
1011EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1012
Johannes Weiner925b7672012-01-12 17:18:15 -08001013/**
1014 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1015 * @zone: zone of the wanted lruvec
1016 * @mem: memcg of the wanted lruvec
1017 *
1018 * Returns the lru list vector holding pages for the given @zone and
1019 * @mem. This can be the global zone lruvec, if the memory controller
1020 * is disabled.
1021 */
1022struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1023 struct mem_cgroup *memcg)
1024{
1025 struct mem_cgroup_per_zone *mz;
1026
1027 if (mem_cgroup_disabled())
1028 return &zone->lruvec;
1029
1030 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1031 return &mz->lruvec;
1032}
1033
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001034/*
1035 * Following LRU functions are allowed to be used without PCG_LOCK.
1036 * Operations are called by routine of global LRU independently from memcg.
1037 * What we have to take care of here is validness of pc->mem_cgroup.
1038 *
1039 * Changes to pc->mem_cgroup happens when
1040 * 1. charge
1041 * 2. moving account
1042 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1043 * It is added to LRU before charge.
1044 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1045 * When moving account, the page is not on LRU. It's isolated.
1046 */
1047
Johannes Weiner925b7672012-01-12 17:18:15 -08001048/**
1049 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1050 * @zone: zone of the page
1051 * @page: the page
1052 * @lru: current lru
1053 *
1054 * This function accounts for @page being added to @lru, and returns
1055 * the lruvec for the given @zone and the memcg @page is charged to.
1056 *
1057 * The callsite is then responsible for physically linking the page to
1058 * the returned lruvec->lists[@lru].
Minchan Kim3f58a822011-03-22 16:32:53 -07001059 */
Johannes Weiner925b7672012-01-12 17:18:15 -08001060struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page,
1061 enum lru_list lru)
Minchan Kim3f58a822011-03-22 16:32:53 -07001062{
1063 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001064 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001065 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001066
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001067 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001068 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001069
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001070 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001071 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001072
1073 /*
1074 * Surreptitiously switch any uncharged page to root:
1075 * an uncharged page off lru does nothing to secure
1076 * its former mem_cgroup from sudden removal.
1077 *
1078 * Our caller holds lru_lock, and PageCgroupUsed is updated
1079 * under page_cgroup lock: between them, they make all uses
1080 * of pc->mem_cgroup safe.
1081 */
1082 if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1083 pc->mem_cgroup = memcg = root_mem_cgroup;
1084
Johannes Weiner925b7672012-01-12 17:18:15 -08001085 mz = page_cgroup_zoneinfo(memcg, page);
1086 /* compound_order() is stabilized through lru_lock */
Hugh Dickins1eb49272012-03-21 16:34:19 -07001087 mz->lru_size[lru] += 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001088 return &mz->lruvec;
1089}
1090
1091/**
1092 * mem_cgroup_lru_del_list - account for removing an lru page
1093 * @page: the page
1094 * @lru: target lru
1095 *
1096 * This function accounts for @page being removed from @lru.
1097 *
1098 * The callsite is then responsible for physically unlinking
1099 * @page->lru.
1100 */
1101void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru)
1102{
1103 struct mem_cgroup_per_zone *mz;
1104 struct mem_cgroup *memcg;
1105 struct page_cgroup *pc;
1106
1107 if (mem_cgroup_disabled())
1108 return;
1109
1110 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001111 memcg = pc->mem_cgroup;
1112 VM_BUG_ON(!memcg);
Johannes Weiner925b7672012-01-12 17:18:15 -08001113 mz = page_cgroup_zoneinfo(memcg, page);
1114 /* huge page split is done under lru_lock. so, we have no races. */
Hugh Dickins1eb49272012-03-21 16:34:19 -07001115 VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page)));
1116 mz->lru_size[lru] -= 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001117}
1118
Johannes Weiner925b7672012-01-12 17:18:15 -08001119/**
1120 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1121 * @zone: zone of the page
1122 * @page: the page
1123 * @from: current lru
1124 * @to: target lru
1125 *
1126 * This function accounts for @page being moved between the lrus @from
1127 * and @to, and returns the lruvec for the given @zone and the memcg
1128 * @page is charged to.
1129 *
1130 * The callsite is then responsible for physically relinking
1131 * @page->lru to the returned lruvec->lists[@to].
1132 */
1133struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone,
1134 struct page *page,
1135 enum lru_list from,
1136 enum lru_list to)
1137{
1138 /* XXX: Optimize this, especially for @from == @to */
1139 mem_cgroup_lru_del_list(page, from);
1140 return mem_cgroup_lru_add_list(zone, page, to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001141}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001142
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001143/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001144 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001145 * hierarchy subtree
1146 */
Johannes Weiner2e3d11f2012-05-29 15:06:25 -07001147bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1148 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001149{
Johannes Weiner9003b352012-05-29 15:06:24 -07001150 if (root_memcg == memcg)
1151 return true;
1152 if (!root_memcg->use_hierarchy)
1153 return false;
Johannes Weiner2e3d11f2012-05-29 15:06:25 -07001154 return css_is_ancestor(&memcg->css, &root_memcg->css);
1155}
1156
1157static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1158 struct mem_cgroup *memcg)
1159{
1160 bool ret;
1161
Johannes Weiner9003b352012-05-29 15:06:24 -07001162 rcu_read_lock();
Johannes Weiner2e3d11f2012-05-29 15:06:25 -07001163 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner9003b352012-05-29 15:06:24 -07001164 rcu_read_unlock();
1165 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001166}
1167
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001168int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001169{
1170 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001171 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001172 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001173
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001174 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001175 if (p) {
1176 curr = try_get_mem_cgroup_from_mm(p->mm);
1177 task_unlock(p);
1178 } else {
1179 /*
1180 * All threads may have already detached their mm's, but the oom
1181 * killer still needs to detect if they have already been oom
1182 * killed to prevent needlessly killing additional tasks.
1183 */
1184 task_lock(task);
1185 curr = mem_cgroup_from_task(task);
1186 if (curr)
1187 css_get(&curr->css);
1188 task_unlock(task);
1189 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001190 if (!curr)
1191 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001192 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001193 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001194 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001195 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1196 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001197 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001198 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001199 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001200 return ret;
1201}
1202
Johannes Weiner9b272972011-11-02 13:38:23 -07001203int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001204{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001205 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001206 int nid = zone_to_nid(zone);
1207 int zid = zone_idx(zone);
1208 unsigned long inactive;
1209 unsigned long active;
1210 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001211
Johannes Weiner9b272972011-11-02 13:38:23 -07001212 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1213 BIT(LRU_INACTIVE_ANON));
1214 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1215 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001216
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001217 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1218 if (gb)
1219 inactive_ratio = int_sqrt(10 * gb);
1220 else
1221 inactive_ratio = 1;
1222
Johannes Weiner9b272972011-11-02 13:38:23 -07001223 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001224}
1225
Johannes Weiner9b272972011-11-02 13:38:23 -07001226int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001227{
1228 unsigned long active;
1229 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001230 int zid = zone_idx(zone);
1231 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001232
Johannes Weiner9b272972011-11-02 13:38:23 -07001233 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1234 BIT(LRU_INACTIVE_FILE));
1235 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1236 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001237
1238 return (active > inactive);
1239}
1240
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001241struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1242 struct zone *zone)
1243{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001244 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001245 int zid = zone_idx(zone);
1246 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1247
1248 return &mz->reclaim_stat;
1249}
1250
1251struct zone_reclaim_stat *
1252mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1253{
1254 struct page_cgroup *pc;
1255 struct mem_cgroup_per_zone *mz;
1256
1257 if (mem_cgroup_disabled())
1258 return NULL;
1259
1260 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001261 if (!PageCgroupUsed(pc))
1262 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001263 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1264 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001265 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001266 return &mz->reclaim_stat;
1267}
1268
Balbir Singh6d61ef42009-01-07 18:08:06 -08001269#define mem_cgroup_from_res_counter(counter, member) \
1270 container_of(counter, struct mem_cgroup, member)
1271
Johannes Weiner19942822011-02-01 15:52:43 -08001272/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001273 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1274 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001275 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001276 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001277 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001278 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001279static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001280{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001281 unsigned long long margin;
1282
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001283 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001284 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001285 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001286 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001287}
1288
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001289int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001290{
1291 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001292
1293 /* root ? */
1294 if (cgrp->parent == NULL)
1295 return vm_swappiness;
1296
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001297 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001298}
1299
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001300/*
1301 * memcg->moving_account is used for checking possibility that some thread is
1302 * calling move_account(). When a thread on CPU-A starts moving pages under
1303 * a memcg, other threads should check memcg->moving_account under
1304 * rcu_read_lock(), like this:
1305 *
1306 * CPU-A CPU-B
1307 * rcu_read_lock()
1308 * memcg->moving_account+1 if (memcg->mocing_account)
1309 * take heavy locks.
1310 * synchronize_rcu() update something.
1311 * rcu_read_unlock()
1312 * start move here.
1313 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001314
1315/* for quick checking without looking up memcg */
1316atomic_t memcg_moving __read_mostly;
1317
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001318static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001319{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001320 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001321 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001322 synchronize_rcu();
1323}
1324
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001325static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001326{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001327 /*
1328 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1329 * We check NULL in callee rather than caller.
1330 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001331 if (memcg) {
1332 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001333 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001334 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001335}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001336
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001337/*
1338 * 2 routines for checking "mem" is under move_account() or not.
1339 *
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001340 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1341 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001342 * pc->mem_cgroup may be overwritten.
1343 *
1344 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1345 * under hierarchy of moving cgroups. This is for
1346 * waiting at hith-memory prressure caused by "move".
1347 */
1348
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001349static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001350{
1351 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001352 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001353}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001355static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001356{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001357 struct mem_cgroup *from;
1358 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001359 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001360 /*
1361 * Unlike task_move routines, we access mc.to, mc.from not under
1362 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1363 */
1364 spin_lock(&mc.lock);
1365 from = mc.from;
1366 to = mc.to;
1367 if (!from)
1368 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001369
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001370 ret = mem_cgroup_same_or_subtree(memcg, from)
1371 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001372unlock:
1373 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001374 return ret;
1375}
1376
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001377static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001378{
1379 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001380 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001381 DEFINE_WAIT(wait);
1382 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1383 /* moving charge context might have finished. */
1384 if (mc.moving_task)
1385 schedule();
1386 finish_wait(&mc.waitq, &wait);
1387 return true;
1388 }
1389 }
1390 return false;
1391}
1392
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001393/*
1394 * Take this lock when
1395 * - a code tries to modify page's memcg while it's USED.
1396 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001397 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001398 */
1399static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1400 unsigned long *flags)
1401{
1402 spin_lock_irqsave(&memcg->move_lock, *flags);
1403}
1404
1405static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1406 unsigned long *flags)
1407{
1408 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1409}
1410
Balbir Singhe2224322009-04-02 16:57:39 -07001411/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001412 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001413 * @memcg: The memory cgroup that went over limit
1414 * @p: Task that is going to be killed
1415 *
1416 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1417 * enabled
1418 */
1419void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1420{
1421 struct cgroup *task_cgrp;
1422 struct cgroup *mem_cgrp;
1423 /*
1424 * Need a buffer in BSS, can't rely on allocations. The code relies
1425 * on the assumption that OOM is serialized for memory controller.
1426 * If this assumption is broken, revisit this code.
1427 */
1428 static char memcg_name[PATH_MAX];
1429 int ret;
1430
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001431 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001432 return;
1433
Balbir Singhe2224322009-04-02 16:57:39 -07001434 rcu_read_lock();
1435
1436 mem_cgrp = memcg->css.cgroup;
1437 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1438
1439 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1440 if (ret < 0) {
1441 /*
1442 * Unfortunately, we are unable to convert to a useful name
1443 * But we'll still print out the usage information
1444 */
1445 rcu_read_unlock();
1446 goto done;
1447 }
1448 rcu_read_unlock();
1449
1450 printk(KERN_INFO "Task in %s killed", memcg_name);
1451
1452 rcu_read_lock();
1453 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1454 if (ret < 0) {
1455 rcu_read_unlock();
1456 goto done;
1457 }
1458 rcu_read_unlock();
1459
1460 /*
1461 * Continues from above, so we don't need an KERN_ level
1462 */
1463 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1464done:
1465
1466 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1467 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1468 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1469 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1470 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1471 "failcnt %llu\n",
1472 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1473 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1474 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1475}
1476
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001477/*
1478 * This function returns the number of memcg under hierarchy tree. Returns
1479 * 1(self count) if no children.
1480 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001481static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001482{
1483 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001484 struct mem_cgroup *iter;
1485
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001486 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001487 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001488 return num;
1489}
1490
Balbir Singh6d61ef42009-01-07 18:08:06 -08001491/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001492 * Return the memory (and swap, if configured) limit for a memcg.
1493 */
1494u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1495{
1496 u64 limit;
1497 u64 memsw;
1498
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001499 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1500 limit += total_swap_pages << PAGE_SHIFT;
1501
David Rientjesa63d83f2010-08-09 17:19:46 -07001502 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1503 /*
1504 * If memsw is finite and limits the amount of swap space available
1505 * to this memcg, return that limit.
1506 */
1507 return min(limit, memsw);
1508}
1509
Johannes Weiner56600482012-01-12 17:17:59 -08001510static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1511 gfp_t gfp_mask,
1512 unsigned long flags)
1513{
1514 unsigned long total = 0;
1515 bool noswap = false;
1516 int loop;
1517
1518 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1519 noswap = true;
1520 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1521 noswap = true;
1522
1523 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1524 if (loop)
1525 drain_all_stock_async(memcg);
1526 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1527 /*
1528 * Allow limit shrinkers, which are triggered directly
1529 * by userspace, to catch signals and stop reclaim
1530 * after minimal progress, regardless of the margin.
1531 */
1532 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1533 break;
1534 if (mem_cgroup_margin(memcg))
1535 break;
1536 /*
1537 * If nothing was reclaimed after two attempts, there
1538 * may be no reclaimable pages in this hierarchy.
1539 */
1540 if (loop && !total)
1541 break;
1542 }
1543 return total;
1544}
1545
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001546/**
1547 * test_mem_cgroup_node_reclaimable
1548 * @mem: the target memcg
1549 * @nid: the node ID to be checked.
1550 * @noswap : specify true here if the user wants flle only information.
1551 *
1552 * This function returns whether the specified memcg contains any
1553 * reclaimable pages on a node. Returns true if there are any reclaimable
1554 * pages in the node.
1555 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001556static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001557 int nid, bool noswap)
1558{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001559 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001560 return true;
1561 if (noswap || !total_swap_pages)
1562 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001563 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001564 return true;
1565 return false;
1566
1567}
Ying Han889976d2011-05-26 16:25:33 -07001568#if MAX_NUMNODES > 1
1569
1570/*
1571 * Always updating the nodemask is not very good - even if we have an empty
1572 * list or the wrong list here, we can start from some node and traverse all
1573 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1574 *
1575 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001576static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001577{
1578 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001579 /*
1580 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1581 * pagein/pageout changes since the last update.
1582 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001584 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001585 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001586 return;
1587
Ying Han889976d2011-05-26 16:25:33 -07001588 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001589 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001590
1591 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001593 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1594 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001595 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001596
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001597 atomic_set(&memcg->numainfo_events, 0);
1598 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001599}
1600
1601/*
1602 * Selecting a node where we start reclaim from. Because what we need is just
1603 * reducing usage counter, start from anywhere is O,K. Considering
1604 * memory reclaim from current node, there are pros. and cons.
1605 *
1606 * Freeing memory from current node means freeing memory from a node which
1607 * we'll use or we've used. So, it may make LRU bad. And if several threads
1608 * hit limits, it will see a contention on a node. But freeing from remote
1609 * node means more costs for memory reclaim because of memory latency.
1610 *
1611 * Now, we use round-robin. Better algorithm is welcomed.
1612 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001613int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001614{
1615 int node;
1616
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001617 mem_cgroup_may_update_nodemask(memcg);
1618 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001619
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001620 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001621 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001622 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001623 /*
1624 * We call this when we hit limit, not when pages are added to LRU.
1625 * No LRU may hold pages because all pages are UNEVICTABLE or
1626 * memcg is too small and all pages are not on LRU. In that case,
1627 * we use curret node.
1628 */
1629 if (unlikely(node == MAX_NUMNODES))
1630 node = numa_node_id();
1631
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001632 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001633 return node;
1634}
1635
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001636/*
1637 * Check all nodes whether it contains reclaimable pages or not.
1638 * For quick scan, we make use of scan_nodes. This will allow us to skip
1639 * unused nodes. But scan_nodes is lazily updated and may not cotain
1640 * enough new information. We need to do double check.
1641 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001642bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001643{
1644 int nid;
1645
1646 /*
1647 * quick check...making use of scan_node.
1648 * We can skip unused nodes.
1649 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001650 if (!nodes_empty(memcg->scan_nodes)) {
1651 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001652 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001653 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001654
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001655 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001656 return true;
1657 }
1658 }
1659 /*
1660 * Check rest of nodes.
1661 */
1662 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001663 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001664 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001665 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001666 return true;
1667 }
1668 return false;
1669}
1670
Ying Han889976d2011-05-26 16:25:33 -07001671#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001672int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001673{
1674 return 0;
1675}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001676
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001677bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001678{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001679 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001680}
Ying Han889976d2011-05-26 16:25:33 -07001681#endif
1682
Johannes Weiner56600482012-01-12 17:17:59 -08001683static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1684 struct zone *zone,
1685 gfp_t gfp_mask,
1686 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001687{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001688 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001689 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001690 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001691 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001692 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001693 struct mem_cgroup_reclaim_cookie reclaim = {
1694 .zone = zone,
1695 .priority = 0,
1696 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001697
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001698 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001699
Balbir Singh4e416952009-09-23 15:56:39 -07001700 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001701 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001702 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001703 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001704 if (loop >= 2) {
1705 /*
1706 * If we have not been able to reclaim
1707 * anything, it might because there are
1708 * no reclaimable pages under this hierarchy
1709 */
Johannes Weiner56600482012-01-12 17:17:59 -08001710 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001711 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001712 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001713 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001714 * excess >> 2 is not to excessive so as to
1715 * reclaim too much, nor too less that we keep
1716 * coming back to reclaim from this cgroup
1717 */
1718 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001719 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001720 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001721 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001722 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001723 }
Johannes Weiner56600482012-01-12 17:17:59 -08001724 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001725 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001726 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1727 zone, &nr_scanned);
1728 *total_scanned += nr_scanned;
1729 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001730 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001731 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001732 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001733 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001734}
1735
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001736/*
1737 * Check OOM-Killer is already running under our hierarchy.
1738 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001739 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001740 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001741static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001742{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001743 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001744
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001745 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001746 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001747 /*
1748 * this subtree of our hierarchy is already locked
1749 * so we cannot give a lock.
1750 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001751 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001752 mem_cgroup_iter_break(memcg, iter);
1753 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001754 } else
1755 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001756 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001757
Michal Hocko79dfdac2011-07-26 16:08:23 -07001758 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001759 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001760
1761 /*
1762 * OK, we failed to lock the whole subtree so we have to clean up
1763 * what we set up to the failing subtree
1764 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001765 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001766 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001767 mem_cgroup_iter_break(memcg, iter);
1768 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001769 }
1770 iter->oom_lock = false;
1771 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001772 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001773}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001774
Michal Hocko79dfdac2011-07-26 16:08:23 -07001775/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001776 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001777 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001778static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001779{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001780 struct mem_cgroup *iter;
1781
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001782 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001783 iter->oom_lock = false;
1784 return 0;
1785}
1786
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001787static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001788{
1789 struct mem_cgroup *iter;
1790
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001791 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001792 atomic_inc(&iter->under_oom);
1793}
1794
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001795static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001796{
1797 struct mem_cgroup *iter;
1798
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001799 /*
1800 * When a new child is created while the hierarchy is under oom,
1801 * mem_cgroup_oom_lock() may not be called. We have to use
1802 * atomic_add_unless() here.
1803 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001804 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001805 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001806}
1807
Michal Hocko1af8efe2011-07-26 16:08:24 -07001808static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001809static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1810
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001811struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001812 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001813 wait_queue_t wait;
1814};
1815
1816static int memcg_oom_wake_function(wait_queue_t *wait,
1817 unsigned mode, int sync, void *arg)
1818{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001819 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1820 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001821 struct oom_wait_info *oom_wait_info;
1822
1823 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001824 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001825
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001826 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001827 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001828 * Then we can use css_is_ancestor without taking care of RCU.
1829 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001830 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1831 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001832 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001833 return autoremove_wake_function(wait, mode, sync, arg);
1834}
1835
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001836static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001837{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001838 /* for filtering, pass "memcg" as argument. */
1839 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001840}
1841
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001842static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001843{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001844 if (memcg && atomic_read(&memcg->under_oom))
1845 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001846}
1847
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001848/*
1849 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1850 */
David Rientjese845e192012-03-21 16:34:10 -07001851bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001852{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001853 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001854 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001855
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001856 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001857 owait.wait.flags = 0;
1858 owait.wait.func = memcg_oom_wake_function;
1859 owait.wait.private = current;
1860 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001861 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001862 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001863
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001864 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001865 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001866 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001867 /*
1868 * Even if signal_pending(), we can't quit charge() loop without
1869 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1870 * under OOM is always welcomed, use TASK_KILLABLE here.
1871 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001872 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001873 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001874 need_to_kill = false;
1875 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001876 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001877 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001878
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001879 if (need_to_kill) {
1880 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07001881 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001882 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001883 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001884 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001885 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001886 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001887 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001888 mem_cgroup_oom_unlock(memcg);
1889 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001890 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001891
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001893
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001894 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1895 return false;
1896 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001897 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001898 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001899}
1900
Balbir Singhd69b0422009-06-17 16:26:34 -07001901/*
1902 * Currently used to update mapped file statistics, but the routine can be
1903 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001904 *
1905 * Notes: Race condition
1906 *
1907 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1908 * it tends to be costly. But considering some conditions, we doesn't need
1909 * to do so _always_.
1910 *
1911 * Considering "charge", lock_page_cgroup() is not required because all
1912 * file-stat operations happen after a page is attached to radix-tree. There
1913 * are no race with "charge".
1914 *
1915 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1916 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1917 * if there are race with "uncharge". Statistics itself is properly handled
1918 * by flags.
1919 *
1920 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001921 * small, we check mm->moving_account and detect there are possibility of race
1922 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001923 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001924
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001925void __mem_cgroup_begin_update_page_stat(struct page *page,
1926 bool *locked, unsigned long *flags)
1927{
1928 struct mem_cgroup *memcg;
1929 struct page_cgroup *pc;
1930
1931 pc = lookup_page_cgroup(page);
1932again:
1933 memcg = pc->mem_cgroup;
1934 if (unlikely(!memcg || !PageCgroupUsed(pc)))
1935 return;
1936 /*
1937 * If this memory cgroup is not under account moving, we don't
1938 * need to take move_lock_page_cgroup(). Because we already hold
1939 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001940 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001941 */
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001942 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001943 return;
1944
1945 move_lock_mem_cgroup(memcg, flags);
1946 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
1947 move_unlock_mem_cgroup(memcg, flags);
1948 goto again;
1949 }
1950 *locked = true;
1951}
1952
1953void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
1954{
1955 struct page_cgroup *pc = lookup_page_cgroup(page);
1956
1957 /*
1958 * It's guaranteed that pc->mem_cgroup never changes while
1959 * lock is held because a routine modifies pc->mem_cgroup
1960 * should take move_lock_page_cgroup().
1961 */
1962 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1963}
1964
Greg Thelen2a7106f2011-01-13 15:47:37 -08001965void mem_cgroup_update_page_stat(struct page *page,
1966 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001967{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001968 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001969 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001970 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001971
Johannes Weinercfa44942012-01-12 17:18:38 -08001972 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001973 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001974
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001975 memcg = pc->mem_cgroup;
1976 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001977 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001978
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001979 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001980 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08001981 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001982 break;
1983 default:
1984 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001985 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001986
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001987 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07001988}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001989
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001990/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001991 * size of first charge trial. "32" comes from vmscan.c's magic value.
1992 * TODO: maybe necessary to use big numbers in big irons.
1993 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001994#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001995struct memcg_stock_pcp {
1996 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001997 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001998 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001999 unsigned long flags;
2000#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002001};
2002static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002003static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002004
2005/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002006 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002007 * from local stock and true is returned. If the stock is 0 or charges from a
2008 * cgroup which is not current target, returns false. This stock will be
2009 * refilled.
2010 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002011static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002012{
2013 struct memcg_stock_pcp *stock;
2014 bool ret = true;
2015
2016 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002017 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002018 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002019 else /* need to call res_counter_charge */
2020 ret = false;
2021 put_cpu_var(memcg_stock);
2022 return ret;
2023}
2024
2025/*
2026 * Returns stocks cached in percpu to res_counter and reset cached information.
2027 */
2028static void drain_stock(struct memcg_stock_pcp *stock)
2029{
2030 struct mem_cgroup *old = stock->cached;
2031
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002032 if (stock->nr_pages) {
2033 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2034
2035 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002036 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002037 res_counter_uncharge(&old->memsw, bytes);
2038 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002039 }
2040 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002041}
2042
2043/*
2044 * This must be called under preempt disabled or must be called by
2045 * a thread which is pinned to local cpu.
2046 */
2047static void drain_local_stock(struct work_struct *dummy)
2048{
2049 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2050 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002051 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002052}
2053
2054/*
2055 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002056 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002057 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002058static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002059{
2060 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2061
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002062 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002063 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002064 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002065 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002066 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002067 put_cpu_var(memcg_stock);
2068}
2069
2070/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002071 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002072 * of the hierarchy under it. sync flag says whether we should block
2073 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002074 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002075static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002076{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002077 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002078
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002079 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002080 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002081 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002082 for_each_online_cpu(cpu) {
2083 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002084 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002085
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002086 memcg = stock->cached;
2087 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002088 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002089 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002090 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002091 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2092 if (cpu == curcpu)
2093 drain_local_stock(&stock->work);
2094 else
2095 schedule_work_on(cpu, &stock->work);
2096 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002097 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002098 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002099
2100 if (!sync)
2101 goto out;
2102
2103 for_each_online_cpu(cpu) {
2104 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002105 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002106 flush_work(&stock->work);
2107 }
2108out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002109 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002110}
2111
2112/*
2113 * Tries to drain stocked charges in other cpus. This function is asynchronous
2114 * and just put a work per cpu for draining localy on each cpu. Caller can
2115 * expects some charges will be back to res_counter later but cannot wait for
2116 * it.
2117 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002118static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002119{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002120 /*
2121 * If someone calls draining, avoid adding more kworker runs.
2122 */
2123 if (!mutex_trylock(&percpu_charge_mutex))
2124 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002125 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002126 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002127}
2128
2129/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002130static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002131{
2132 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002133 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002134 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002135 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002136}
2137
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002138/*
2139 * This function drains percpu counter value from DEAD cpu and
2140 * move it to local cpu. Note that this function can be preempted.
2141 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002142static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002143{
2144 int i;
2145
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002146 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002147 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002148 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002149
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002150 per_cpu(memcg->stat->count[i], cpu) = 0;
2151 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002152 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002153 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002154 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002155
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002156 per_cpu(memcg->stat->events[i], cpu) = 0;
2157 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002158 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002159 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002160}
2161
2162static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002163 unsigned long action,
2164 void *hcpu)
2165{
2166 int cpu = (unsigned long)hcpu;
2167 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002168 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002169
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002170 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002171 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002172
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002173 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002174 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002175
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002176 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002177 mem_cgroup_drain_pcp_counter(iter, cpu);
2178
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002179 stock = &per_cpu(memcg_stock, cpu);
2180 drain_stock(stock);
2181 return NOTIFY_OK;
2182}
2183
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002184
2185/* See __mem_cgroup_try_charge() for details */
2186enum {
2187 CHARGE_OK, /* success */
2188 CHARGE_RETRY, /* need to retry but retry is not bad */
2189 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2190 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2191 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2192};
2193
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002194static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002195 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002196{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002197 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002198 struct mem_cgroup *mem_over_limit;
2199 struct res_counter *fail_res;
2200 unsigned long flags = 0;
2201 int ret;
2202
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002203 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002204
2205 if (likely(!ret)) {
2206 if (!do_swap_account)
2207 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002208 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002209 if (likely(!ret))
2210 return CHARGE_OK;
2211
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002212 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002213 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2214 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2215 } else
2216 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002217 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002218 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2219 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002220 *
2221 * Never reclaim on behalf of optional batching, retry with a
2222 * single page instead.
2223 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002224 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002225 return CHARGE_RETRY;
2226
2227 if (!(gfp_mask & __GFP_WAIT))
2228 return CHARGE_WOULDBLOCK;
2229
Johannes Weiner56600482012-01-12 17:17:59 -08002230 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002231 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002232 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002233 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002234 * Even though the limit is exceeded at this point, reclaim
2235 * may have been able to free some pages. Retry the charge
2236 * before killing the task.
2237 *
2238 * Only for regular pages, though: huge pages are rather
2239 * unlikely to succeed so close to the limit, and we fall back
2240 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002241 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002242 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002243 return CHARGE_RETRY;
2244
2245 /*
2246 * At task move, charge accounts can be doubly counted. So, it's
2247 * better to wait until the end of task_move if something is going on.
2248 */
2249 if (mem_cgroup_wait_acct_move(mem_over_limit))
2250 return CHARGE_RETRY;
2251
2252 /* If we don't need to call oom-killer at el, return immediately */
2253 if (!oom_check)
2254 return CHARGE_NOMEM;
2255 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002256 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002257 return CHARGE_OOM_DIE;
2258
2259 return CHARGE_RETRY;
2260}
2261
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002262/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002263 * __mem_cgroup_try_charge() does
2264 * 1. detect memcg to be charged against from passed *mm and *ptr,
2265 * 2. update res_counter
2266 * 3. call memory reclaim if necessary.
2267 *
2268 * In some special case, if the task is fatal, fatal_signal_pending() or
2269 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2270 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2271 * as possible without any hazards. 2: all pages should have a valid
2272 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2273 * pointer, that is treated as a charge to root_mem_cgroup.
2274 *
2275 * So __mem_cgroup_try_charge() will return
2276 * 0 ... on success, filling *ptr with a valid memcg pointer.
2277 * -ENOMEM ... charge failure because of resource limits.
2278 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2279 *
2280 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2281 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002282 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002283static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002284 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002285 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002286 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002287 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002288{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002289 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002290 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002291 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002292 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002293
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002294 /*
2295 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2296 * in system level. So, allow to go ahead dying process in addition to
2297 * MEMDIE process.
2298 */
2299 if (unlikely(test_thread_flag(TIF_MEMDIE)
2300 || fatal_signal_pending(current)))
2301 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002302
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002303 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002304 * We always charge the cgroup the mm_struct belongs to.
2305 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002306 * thread group leader migrates. It's possible that mm is not
2307 * set, if so charge the init_mm (happens for pagecache usage).
2308 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002309 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002310 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002311again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002312 if (*ptr) { /* css should be a valid one */
2313 memcg = *ptr;
2314 VM_BUG_ON(css_is_removed(&memcg->css));
2315 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002316 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002317 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002318 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002319 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002320 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002321 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002322
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002323 rcu_read_lock();
2324 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002325 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002326 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002327 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002328 * race with swapoff. Then, we have small risk of mis-accouning.
2329 * But such kind of mis-account by race always happens because
2330 * we don't have cgroup_mutex(). It's overkill and we allo that
2331 * small race, here.
2332 * (*) swapoff at el will charge against mm-struct not against
2333 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002334 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002335 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002336 if (!memcg)
2337 memcg = root_mem_cgroup;
2338 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002339 rcu_read_unlock();
2340 goto done;
2341 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002342 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002343 /*
2344 * It seems dagerous to access memcg without css_get().
2345 * But considering how consume_stok works, it's not
2346 * necessary. If consume_stock success, some charges
2347 * from this memcg are cached on this cpu. So, we
2348 * don't need to call css_get()/css_tryget() before
2349 * calling consume_stock().
2350 */
2351 rcu_read_unlock();
2352 goto done;
2353 }
2354 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002355 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002356 rcu_read_unlock();
2357 goto again;
2358 }
2359 rcu_read_unlock();
2360 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002361
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002362 do {
2363 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002364
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002365 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002366 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002367 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002368 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002369 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002370
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002371 oom_check = false;
2372 if (oom && !nr_oom_retries) {
2373 oom_check = true;
2374 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2375 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002376
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002377 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002378 switch (ret) {
2379 case CHARGE_OK:
2380 break;
2381 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002382 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002383 css_put(&memcg->css);
2384 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002385 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002386 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002387 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002388 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002389 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002390 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002391 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002392 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002393 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002394 /* If oom, we never return -ENOMEM */
2395 nr_oom_retries--;
2396 break;
2397 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002398 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002399 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002400 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002401 } while (ret != CHARGE_OK);
2402
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002403 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002404 refill_stock(memcg, batch - nr_pages);
2405 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002406done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002407 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002408 return 0;
2409nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002410 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002411 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002412bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002413 *ptr = root_mem_cgroup;
2414 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002415}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002416
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002417/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002418 * Somemtimes we have to undo a charge we got by try_charge().
2419 * This function is for that and do uncharge, put css's refcnt.
2420 * gotten by try_charge().
2421 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002422static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b82011-03-23 16:42:33 -07002423 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002424{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002425 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b82011-03-23 16:42:33 -07002426 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002427
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002428 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002429 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002430 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002431 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002432}
2433
2434/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002435 * A helper function to get mem_cgroup from ID. must be called under
2436 * rcu_read_lock(). The caller must check css_is_removed() or some if
2437 * it's concern. (dropping refcnt from swap can be called against removed
2438 * memcg.)
2439 */
2440static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2441{
2442 struct cgroup_subsys_state *css;
2443
2444 /* ID 0 is unused ID */
2445 if (!id)
2446 return NULL;
2447 css = css_lookup(&mem_cgroup_subsys, id);
2448 if (!css)
2449 return NULL;
2450 return container_of(css, struct mem_cgroup, css);
2451}
2452
Wu Fengguange42d9d52009-12-16 12:19:59 +01002453struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002454{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002455 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002456 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002457 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002458 swp_entry_t ent;
2459
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002460 VM_BUG_ON(!PageLocked(page));
2461
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002462 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002463 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002464 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002465 memcg = pc->mem_cgroup;
2466 if (memcg && !css_tryget(&memcg->css))
2467 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002468 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002469 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002470 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002471 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002472 memcg = mem_cgroup_lookup(id);
2473 if (memcg && !css_tryget(&memcg->css))
2474 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002475 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002476 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002477 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002478 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002479}
2480
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002481static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002482 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002483 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002484 enum charge_type ctype,
2485 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002486{
Johannes Weinerce587e62012-04-24 20:22:33 +02002487 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002488 struct zone *uninitialized_var(zone);
2489 bool was_on_lru = false;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002490 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002491
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002492 lock_page_cgroup(pc);
2493 if (unlikely(PageCgroupUsed(pc))) {
2494 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002495 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002496 return;
2497 }
2498 /*
2499 * we don't need page_cgroup_lock about tail pages, becase they are not
2500 * accessed by any other context at this point.
2501 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002502
2503 /*
2504 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2505 * may already be on some other mem_cgroup's LRU. Take care of it.
2506 */
2507 if (lrucare) {
2508 zone = page_zone(page);
2509 spin_lock_irq(&zone->lru_lock);
2510 if (PageLRU(page)) {
2511 ClearPageLRU(page);
2512 del_page_from_lru_list(zone, page, page_lru(page));
2513 was_on_lru = true;
2514 }
2515 }
2516
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002517 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002518 /*
2519 * We access a page_cgroup asynchronously without lock_page_cgroup().
2520 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2521 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2522 * before USED bit, we need memory barrier here.
2523 * See mem_cgroup_add_lru_list(), etc.
2524 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002525 smp_wmb();
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002526 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002527
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002528 if (lrucare) {
2529 if (was_on_lru) {
2530 VM_BUG_ON(PageLRU(page));
2531 SetPageLRU(page);
2532 add_page_to_lru_list(zone, page, page_lru(page));
2533 }
2534 spin_unlock_irq(&zone->lru_lock);
2535 }
2536
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002537 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2538 anon = true;
2539 else
2540 anon = false;
2541
2542 mem_cgroup_charge_statistics(memcg, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002543 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002544
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002545 /*
2546 * "charge_statistics" updated event counter. Then, check it.
2547 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2548 * if they exceeds softlimit.
2549 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002550 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002551}
2552
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002553#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2554
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002555#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION))
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002556/*
2557 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002558 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2559 * charge/uncharge will be never happen and move_account() is done under
2560 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002561 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002562void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002563{
2564 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002565 struct page_cgroup *pc;
2566 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002567
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002568 if (mem_cgroup_disabled())
2569 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002570 for (i = 1; i < HPAGE_PMD_NR; i++) {
2571 pc = head_pc + i;
2572 pc->mem_cgroup = head_pc->mem_cgroup;
2573 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002574 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2575 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002576}
Hugh Dickins12d27102012-01-12 17:19:52 -08002577#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002578
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002579/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002580 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002581 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002582 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002583 * @pc: page_cgroup of the page.
2584 * @from: mem_cgroup which the page is moved from.
2585 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002586 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002587 *
2588 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002589 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002590 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002591 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002592 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002593 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002594 * true, this function does "uncharge" from old cgroup, but it doesn't if
2595 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002596 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002597static int mem_cgroup_move_account(struct page *page,
2598 unsigned int nr_pages,
2599 struct page_cgroup *pc,
2600 struct mem_cgroup *from,
2601 struct mem_cgroup *to,
2602 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002603{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002604 unsigned long flags;
2605 int ret;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002606 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002607
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002608 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002609 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002610 /*
2611 * The page is isolated from LRU. So, collapse function
2612 * will not handle this page. But page splitting can happen.
2613 * Do this check under compound_page_lock(). The caller should
2614 * hold it.
2615 */
2616 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002617 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002618 goto out;
2619
2620 lock_page_cgroup(pc);
2621
2622 ret = -EINVAL;
2623 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2624 goto unlock;
2625
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002626 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002627
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002628 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002629 /* Update mapped_file data for mem_cgroup */
2630 preempt_disable();
2631 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2632 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2633 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002634 }
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002635 mem_cgroup_charge_statistics(from, anon, -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002636 if (uncharge)
2637 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b82011-03-23 16:42:33 -07002638 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002639
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002640 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002641 pc->mem_cgroup = to;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002642 mem_cgroup_charge_statistics(to, anon, nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002643 /*
2644 * We charges against "to" which may not have any tasks. Then, "to"
2645 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002646 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002647 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002648 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002649 */
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002650 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002651 ret = 0;
2652unlock:
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002653 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002654 /*
2655 * check events
2656 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002657 memcg_check_events(to, page);
2658 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002659out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002660 return ret;
2661}
2662
2663/*
2664 * move charges to its parent.
2665 */
2666
Johannes Weiner5564e882011-03-23 16:42:29 -07002667static int mem_cgroup_move_parent(struct page *page,
2668 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002669 struct mem_cgroup *child,
2670 gfp_t gfp_mask)
2671{
2672 struct cgroup *cg = child->css.cgroup;
2673 struct cgroup *pcg = cg->parent;
2674 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002675 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002676 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002677 int ret;
2678
2679 /* Is ROOT ? */
2680 if (!pcg)
2681 return -EINVAL;
2682
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002683 ret = -EBUSY;
2684 if (!get_page_unless_zero(page))
2685 goto out;
2686 if (isolate_lru_page(page))
2687 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002688
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002689 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002690
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002691 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002692 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002693 if (ret)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002694 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002695
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002696 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002697 flags = compound_lock_irqsave(page);
2698
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002699 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002700 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002701 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002702
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002703 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002704 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002705put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002706 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002707put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002708 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002709out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002710 return ret;
2711}
2712
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002713/*
2714 * Charge the memory controller for page usage.
2715 * Return
2716 * 0 if the charge was successful
2717 * < 0 if the cgroup is over its limit
2718 */
2719static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002720 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002721{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002723 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002724 bool oom = true;
2725 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002726
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002727 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002728 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002729 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002730 /*
2731 * Never OOM-kill a process for a huge page. The
2732 * fault handler will fall back to regular pages.
2733 */
2734 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002735 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002736
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002737 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002738 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002739 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02002740 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002741 return 0;
2742}
2743
2744int mem_cgroup_newpage_charge(struct page *page,
2745 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002746{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002747 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002748 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002749 VM_BUG_ON(page_mapped(page));
2750 VM_BUG_ON(page->mapping && !PageAnon(page));
2751 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002752 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002753 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002754}
2755
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002756static void
2757__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2758 enum charge_type ctype);
2759
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002760int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2761 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002762{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002763 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002764 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002765 int ret;
2766
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002767 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002768 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002769 if (PageCompound(page))
2770 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002771
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002772 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002773 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002774 if (!page_is_file_cache(page))
2775 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002776
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002777 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002778 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002779 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002780 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002781 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002782 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2783 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002784 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002785}
2786
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002787/*
2788 * While swap-in, try_charge -> commit or cancel, the page is locked.
2789 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002790 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002791 * "commit()" or removed by "cancel()"
2792 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002793int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2794 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002795 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002796{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002797 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002798 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002799
Johannes Weiner72835c82012-01-12 17:18:32 -08002800 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002801
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002802 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002803 return 0;
2804
2805 if (!do_swap_account)
2806 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002807 /*
2808 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002809 * the pte, and even removed page from swap cache: in those cases
2810 * do_swap_page()'s pte_same() test will fail; but there's also a
2811 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002812 */
2813 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002814 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002815 memcg = try_get_mem_cgroup_from_page(page);
2816 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002817 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002818 *memcgp = memcg;
2819 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002820 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002821 if (ret == -EINTR)
2822 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002823 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002824charge_cur_mm:
2825 if (unlikely(!mm))
2826 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002827 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2828 if (ret == -EINTR)
2829 ret = 0;
2830 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002831}
2832
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002833static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002834__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002835 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002836{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002837 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002838 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002839 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002840 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002841 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002842
Johannes Weinerce587e62012-04-24 20:22:33 +02002843 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002844 /*
2845 * Now swap is on-memory. This means this page may be
2846 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002847 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2848 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2849 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002850 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002851 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002852 swp_entry_t ent = {.val = page_private(page)};
Johannes Weiner72835c82012-01-12 17:18:32 -08002853 struct mem_cgroup *swap_memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002854 unsigned short id;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002855
2856 id = swap_cgroup_record(ent, 0);
2857 rcu_read_lock();
Johannes Weiner72835c82012-01-12 17:18:32 -08002858 swap_memcg = mem_cgroup_lookup(id);
2859 if (swap_memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002860 /*
2861 * This recorded memcg can be obsolete one. So, avoid
2862 * calling css_tryget
2863 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002864 if (!mem_cgroup_is_root(swap_memcg))
2865 res_counter_uncharge(&swap_memcg->memsw,
2866 PAGE_SIZE);
2867 mem_cgroup_swap_statistics(swap_memcg, false);
2868 mem_cgroup_put(swap_memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002869 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002870 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002871 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002872 /*
2873 * At swapin, we may charge account against cgroup which has no tasks.
2874 * So, rmdir()->pre_destroy() can be called while we do this charge.
2875 * In that case, we need to call pre_destroy() again. check it here.
2876 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002877 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002878}
2879
Johannes Weiner72835c82012-01-12 17:18:32 -08002880void mem_cgroup_commit_charge_swapin(struct page *page,
2881 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002882{
Johannes Weiner72835c82012-01-12 17:18:32 -08002883 __mem_cgroup_commit_charge_swapin(page, memcg,
2884 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002885}
2886
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002887void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002888{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002889 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002890 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002891 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002892 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002893 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002894}
2895
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002896static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002897 unsigned int nr_pages,
2898 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002899{
2900 struct memcg_batch_info *batch = NULL;
2901 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002902
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002903 /* If swapout, usage of swap doesn't decrease */
2904 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2905 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002906
2907 batch = &current->memcg_batch;
2908 /*
2909 * In usual, we do css_get() when we remember memcg pointer.
2910 * But in this case, we keep res->usage until end of a series of
2911 * uncharges. Then, it's ok to ignore memcg's refcnt.
2912 */
2913 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002914 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002915 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002916 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002917 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002918 * the same cgroup and we have chance to coalesce uncharges.
2919 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2920 * because we want to do uncharge as soon as possible.
2921 */
2922
2923 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2924 goto direct_uncharge;
2925
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002926 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002927 goto direct_uncharge;
2928
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002929 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002930 * In typical case, batch->memcg == mem. This means we can
2931 * merge a series of uncharges to an uncharge of res_counter.
2932 * If not, we uncharge res_counter ony by one.
2933 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002934 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002935 goto direct_uncharge;
2936 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002937 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002938 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002939 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002940 return;
2941direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002942 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002943 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002944 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2945 if (unlikely(batch->memcg != memcg))
2946 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002947}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002948
Balbir Singh8697d332008-02-07 00:13:59 -08002949/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002950 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002951 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002952static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002953__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002954{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002955 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002956 unsigned int nr_pages = 1;
2957 struct page_cgroup *pc;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002958 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002959
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002960 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002961 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002962
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002963 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002964 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002965
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002966 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002967 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002968 VM_BUG_ON(!PageTransHuge(page));
2969 }
Balbir Singh8697d332008-02-07 00:13:59 -08002970 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002971 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002972 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002973 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002974 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002975 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002976
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002977 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002978
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002979 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002980
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002981 if (!PageCgroupUsed(pc))
2982 goto unlock_out;
2983
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002984 anon = PageAnon(page);
2985
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002986 switch (ctype) {
2987 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002988 /*
2989 * Generally PageAnon tells if it's the anon statistics to be
2990 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2991 * used before page reached the stage of being marked PageAnon.
2992 */
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002993 anon = true;
2994 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002995 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07002996 /* See mem_cgroup_prepare_migration() */
2997 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002998 goto unlock_out;
2999 break;
3000 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3001 if (!PageAnon(page)) { /* Shared memory */
3002 if (page->mapping && !page_is_file_cache(page))
3003 goto unlock_out;
3004 } else if (page_mapped(page)) /* Anon */
3005 goto unlock_out;
3006 break;
3007 default:
3008 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003009 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003010
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003011 mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003012
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003013 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003014 /*
3015 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3016 * freed from LRU. This is safe because uncharged page is expected not
3017 * to be reused (freed soon). Exception is SwapCache, it's handled by
3018 * special functions.
3019 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003020
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003021 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003022 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003023 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003024 * will never be freed.
3025 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003026 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003027 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003028 mem_cgroup_swap_statistics(memcg, true);
3029 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003030 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003031 if (!mem_cgroup_is_root(memcg))
3032 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003033
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003034 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003035
3036unlock_out:
3037 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003038 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003039}
3040
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003041void mem_cgroup_uncharge_page(struct page *page)
3042{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003043 /* early check. */
3044 if (page_mapped(page))
3045 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08003046 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003047 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3048}
3049
3050void mem_cgroup_uncharge_cache_page(struct page *page)
3051{
3052 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003053 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003054 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3055}
3056
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003057/*
3058 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3059 * In that cases, pages are freed continuously and we can expect pages
3060 * are in the same memcg. All these calls itself limits the number of
3061 * pages freed at once, then uncharge_start/end() is called properly.
3062 * This may be called prural(2) times in a context,
3063 */
3064
3065void mem_cgroup_uncharge_start(void)
3066{
3067 current->memcg_batch.do_batch++;
3068 /* We can do nest. */
3069 if (current->memcg_batch.do_batch == 1) {
3070 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003071 current->memcg_batch.nr_pages = 0;
3072 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003073 }
3074}
3075
3076void mem_cgroup_uncharge_end(void)
3077{
3078 struct memcg_batch_info *batch = &current->memcg_batch;
3079
3080 if (!batch->do_batch)
3081 return;
3082
3083 batch->do_batch--;
3084 if (batch->do_batch) /* If stacked, do nothing. */
3085 return;
3086
3087 if (!batch->memcg)
3088 return;
3089 /*
3090 * This "batch->memcg" is valid without any css_get/put etc...
3091 * bacause we hide charges behind us.
3092 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003093 if (batch->nr_pages)
3094 res_counter_uncharge(&batch->memcg->res,
3095 batch->nr_pages * PAGE_SIZE);
3096 if (batch->memsw_nr_pages)
3097 res_counter_uncharge(&batch->memcg->memsw,
3098 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003099 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003100 /* forget this pointer (for sanity check) */
3101 batch->memcg = NULL;
3102}
3103
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003104#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003105/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003106 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003107 * memcg information is recorded to swap_cgroup of "ent"
3108 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003109void
3110mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003111{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003112 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003113 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003114
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003115 if (!swapout) /* this was a swap cache but the swap is unused ! */
3116 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3117
3118 memcg = __mem_cgroup_uncharge_common(page, ctype);
3119
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003120 /*
3121 * record memcg information, if swapout && memcg != NULL,
3122 * mem_cgroup_get() was called in uncharge().
3123 */
3124 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003125 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003126}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003127#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003128
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003129#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3130/*
3131 * called from swap_entry_free(). remove record in swap_cgroup and
3132 * uncharge "memsw" account.
3133 */
3134void mem_cgroup_uncharge_swap(swp_entry_t ent)
3135{
3136 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003137 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003138
3139 if (!do_swap_account)
3140 return;
3141
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003142 id = swap_cgroup_record(ent, 0);
3143 rcu_read_lock();
3144 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003145 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003146 /*
3147 * We uncharge this because swap is freed.
3148 * This memcg can be obsolete one. We avoid calling css_tryget
3149 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003150 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003151 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003152 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003153 mem_cgroup_put(memcg);
3154 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003155 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003156}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003157
3158/**
3159 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3160 * @entry: swap entry to be moved
3161 * @from: mem_cgroup which the entry is moved from
3162 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003163 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003164 *
3165 * It succeeds only when the swap_cgroup's record for this entry is the same
3166 * as the mem_cgroup's id of @from.
3167 *
3168 * Returns 0 on success, -EINVAL on failure.
3169 *
3170 * The caller must have charged to @to, IOW, called res_counter_charge() about
3171 * both res and memsw, and called css_get().
3172 */
3173static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003174 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003175{
3176 unsigned short old_id, new_id;
3177
3178 old_id = css_id(&from->css);
3179 new_id = css_id(&to->css);
3180
3181 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003182 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003183 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003184 /*
3185 * This function is only called from task migration context now.
3186 * It postpones res_counter and refcount handling till the end
3187 * of task migration(mem_cgroup_clear_mc()) for performance
3188 * improvement. But we cannot postpone mem_cgroup_get(to)
3189 * because if the process that has been moved to @to does
3190 * swap-in, the refcount of @to might be decreased to 0.
3191 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003192 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003193 if (need_fixup) {
3194 if (!mem_cgroup_is_root(from))
3195 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3196 mem_cgroup_put(from);
3197 /*
3198 * we charged both to->res and to->memsw, so we should
3199 * uncharge to->res.
3200 */
3201 if (!mem_cgroup_is_root(to))
3202 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003203 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003204 return 0;
3205 }
3206 return -EINVAL;
3207}
3208#else
3209static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003210 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003211{
3212 return -EINVAL;
3213}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003214#endif
3215
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003216/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003217 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3218 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003219 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003220int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003221 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003222{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003223 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003224 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003225 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003226 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003227
Johannes Weiner72835c82012-01-12 17:18:32 -08003228 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003229
Andrea Arcangeliec168512011-01-13 15:46:56 -08003230 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003231 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003232 return 0;
3233
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003234 pc = lookup_page_cgroup(page);
3235 lock_page_cgroup(pc);
3236 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003237 memcg = pc->mem_cgroup;
3238 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003239 /*
3240 * At migrating an anonymous page, its mapcount goes down
3241 * to 0 and uncharge() will be called. But, even if it's fully
3242 * unmapped, migration may fail and this page has to be
3243 * charged again. We set MIGRATION flag here and delay uncharge
3244 * until end_migration() is called
3245 *
3246 * Corner Case Thinking
3247 * A)
3248 * When the old page was mapped as Anon and it's unmap-and-freed
3249 * while migration was ongoing.
3250 * If unmap finds the old page, uncharge() of it will be delayed
3251 * until end_migration(). If unmap finds a new page, it's
3252 * uncharged when it make mapcount to be 1->0. If unmap code
3253 * finds swap_migration_entry, the new page will not be mapped
3254 * and end_migration() will find it(mapcount==0).
3255 *
3256 * B)
3257 * When the old page was mapped but migraion fails, the kernel
3258 * remaps it. A charge for it is kept by MIGRATION flag even
3259 * if mapcount goes down to 0. We can do remap successfully
3260 * without charging it again.
3261 *
3262 * C)
3263 * The "old" page is under lock_page() until the end of
3264 * migration, so, the old page itself will not be swapped-out.
3265 * If the new page is swapped out before end_migraton, our
3266 * hook to usual swap-out path will catch the event.
3267 */
3268 if (PageAnon(page))
3269 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003270 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003271 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003272 /*
3273 * If the page is not charged at this point,
3274 * we return here.
3275 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003276 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003277 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003278
Johannes Weiner72835c82012-01-12 17:18:32 -08003279 *memcgp = memcg;
3280 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003281 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003282 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003283 if (PageAnon(page)) {
3284 lock_page_cgroup(pc);
3285 ClearPageCgroupMigration(pc);
3286 unlock_page_cgroup(pc);
3287 /*
3288 * The old page may be fully unmapped while we kept it.
3289 */
3290 mem_cgroup_uncharge_page(page);
3291 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003292 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003293 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003294 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003295 /*
3296 * We charge new page before it's used/mapped. So, even if unlock_page()
3297 * is called before end_migration, we can catch all events on this new
3298 * page. In the case new page is migrated but not remapped, new page's
3299 * mapcount will be finally 0 and we call uncharge in end_migration().
3300 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003301 if (PageAnon(page))
3302 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3303 else if (page_is_file_cache(page))
3304 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3305 else
3306 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Johannes Weinerce587e62012-04-24 20:22:33 +02003307 __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003308 return ret;
3309}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003310
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003311/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003312void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003313 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003314{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003315 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003316 struct page_cgroup *pc;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003317 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003318
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003319 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003320 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003321 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003322 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003323 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003324 used = oldpage;
3325 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003326 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003327 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003328 unused = oldpage;
3329 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003330 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003331 * We disallowed uncharge of pages under migration because mapcount
3332 * of the page goes down to zero, temporarly.
3333 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003334 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003335 pc = lookup_page_cgroup(oldpage);
3336 lock_page_cgroup(pc);
3337 ClearPageCgroupMigration(pc);
3338 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003339 anon = PageAnon(used);
3340 __mem_cgroup_uncharge_common(unused,
3341 anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED
3342 : MEM_CGROUP_CHARGE_TYPE_CACHE);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003343
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003344 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003345 * If a page is a file cache, radix-tree replacement is very atomic
3346 * and we can skip this check. When it was an Anon page, its mapcount
3347 * goes down to 0. But because we added MIGRATION flage, it's not
3348 * uncharged yet. There are several case but page->mapcount check
3349 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3350 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003351 */
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003352 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003353 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003354 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003355 * At migration, we may charge account against cgroup which has no
3356 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003357 * So, rmdir()->pre_destroy() can be called while we do this charge.
3358 * In that case, we need to call pre_destroy() again. check it here.
3359 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003360 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003361}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003362
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003363/*
3364 * At replace page cache, newpage is not under any memcg but it's on
3365 * LRU. So, this function doesn't touch res_counter but handles LRU
3366 * in correct way. Both pages are locked so we cannot race with uncharge.
3367 */
3368void mem_cgroup_replace_page_cache(struct page *oldpage,
3369 struct page *newpage)
3370{
3371 struct mem_cgroup *memcg;
3372 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003373 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003374
3375 if (mem_cgroup_disabled())
3376 return;
3377
3378 pc = lookup_page_cgroup(oldpage);
3379 /* fix accounting on old pages */
3380 lock_page_cgroup(pc);
3381 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003382 mem_cgroup_charge_statistics(memcg, false, -1);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003383 ClearPageCgroupUsed(pc);
3384 unlock_page_cgroup(pc);
3385
3386 if (PageSwapBacked(oldpage))
3387 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3388
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003389 /*
3390 * Even if newpage->mapping was NULL before starting replacement,
3391 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3392 * LRU while we overwrite pc->mem_cgroup.
3393 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003394 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003395}
3396
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003397#ifdef CONFIG_DEBUG_VM
3398static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3399{
3400 struct page_cgroup *pc;
3401
3402 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003403 /*
3404 * Can be NULL while feeding pages into the page allocator for
3405 * the first time, i.e. during boot or memory hotplug;
3406 * or when mem_cgroup_disabled().
3407 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003408 if (likely(pc) && PageCgroupUsed(pc))
3409 return pc;
3410 return NULL;
3411}
3412
3413bool mem_cgroup_bad_page_check(struct page *page)
3414{
3415 if (mem_cgroup_disabled())
3416 return false;
3417
3418 return lookup_page_cgroup_used(page) != NULL;
3419}
3420
3421void mem_cgroup_print_bad_page(struct page *page)
3422{
3423 struct page_cgroup *pc;
3424
3425 pc = lookup_page_cgroup_used(page);
3426 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003427 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003428 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003429 }
3430}
3431#endif
3432
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003433static DEFINE_MUTEX(set_limit_mutex);
3434
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003435static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003436 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003437{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003438 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003439 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003440 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003441 int children = mem_cgroup_count_children(memcg);
3442 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003443 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003444
3445 /*
3446 * For keeping hierarchical_reclaim simple, how long we should retry
3447 * is depends on callers. We set our retry-count to be function
3448 * of # of children which we should visit in this loop.
3449 */
3450 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3451
3452 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003453
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003454 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003455 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003456 if (signal_pending(current)) {
3457 ret = -EINTR;
3458 break;
3459 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003460 /*
3461 * Rather than hide all in some function, I do this in
3462 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003463 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003464 */
3465 mutex_lock(&set_limit_mutex);
3466 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3467 if (memswlimit < val) {
3468 ret = -EINVAL;
3469 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003470 break;
3471 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003472
3473 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3474 if (memlimit < val)
3475 enlarge = 1;
3476
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003477 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003478 if (!ret) {
3479 if (memswlimit == val)
3480 memcg->memsw_is_minimum = true;
3481 else
3482 memcg->memsw_is_minimum = false;
3483 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003484 mutex_unlock(&set_limit_mutex);
3485
3486 if (!ret)
3487 break;
3488
Johannes Weiner56600482012-01-12 17:17:59 -08003489 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3490 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003491 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3492 /* Usage is reduced ? */
3493 if (curusage >= oldusage)
3494 retry_count--;
3495 else
3496 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003497 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003498 if (!ret && enlarge)
3499 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003500
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003501 return ret;
3502}
3503
Li Zefan338c8432009-06-17 16:27:15 -07003504static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3505 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003506{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003507 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003508 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003509 int children = mem_cgroup_count_children(memcg);
3510 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003511 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003512
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003513 /* see mem_cgroup_resize_res_limit */
3514 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3515 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003516 while (retry_count) {
3517 if (signal_pending(current)) {
3518 ret = -EINTR;
3519 break;
3520 }
3521 /*
3522 * Rather than hide all in some function, I do this in
3523 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003524 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003525 */
3526 mutex_lock(&set_limit_mutex);
3527 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3528 if (memlimit > val) {
3529 ret = -EINVAL;
3530 mutex_unlock(&set_limit_mutex);
3531 break;
3532 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003533 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3534 if (memswlimit < val)
3535 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003536 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003537 if (!ret) {
3538 if (memlimit == val)
3539 memcg->memsw_is_minimum = true;
3540 else
3541 memcg->memsw_is_minimum = false;
3542 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003543 mutex_unlock(&set_limit_mutex);
3544
3545 if (!ret)
3546 break;
3547
Johannes Weiner56600482012-01-12 17:17:59 -08003548 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3549 MEM_CGROUP_RECLAIM_NOSWAP |
3550 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003551 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003552 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003553 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003554 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003555 else
3556 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003557 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003558 if (!ret && enlarge)
3559 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003560 return ret;
3561}
3562
Balbir Singh4e416952009-09-23 15:56:39 -07003563unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003564 gfp_t gfp_mask,
3565 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003566{
3567 unsigned long nr_reclaimed = 0;
3568 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3569 unsigned long reclaimed;
3570 int loop = 0;
3571 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003572 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003573 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003574
3575 if (order > 0)
3576 return 0;
3577
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003578 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003579 /*
3580 * This loop can run a while, specially if mem_cgroup's continuously
3581 * keep exceeding their soft limit and putting the system under
3582 * pressure
3583 */
3584 do {
3585 if (next_mz)
3586 mz = next_mz;
3587 else
3588 mz = mem_cgroup_largest_soft_limit_node(mctz);
3589 if (!mz)
3590 break;
3591
Ying Han0ae5e892011-05-26 16:25:25 -07003592 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003593 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08003594 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003595 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003596 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003597 spin_lock(&mctz->lock);
3598
3599 /*
3600 * If we failed to reclaim anything from this memory cgroup
3601 * it is time to move on to the next cgroup
3602 */
3603 next_mz = NULL;
3604 if (!reclaimed) {
3605 do {
3606 /*
3607 * Loop until we find yet another one.
3608 *
3609 * By the time we get the soft_limit lock
3610 * again, someone might have aded the
3611 * group back on the RB tree. Iterate to
3612 * make sure we get a different mem.
3613 * mem_cgroup_largest_soft_limit_node returns
3614 * NULL if no other cgroup is present on
3615 * the tree
3616 */
3617 next_mz =
3618 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003619 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003620 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003621 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003622 break;
3623 } while (1);
3624 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003625 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3626 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003627 /*
3628 * One school of thought says that we should not add
3629 * back the node to the tree if reclaim returns 0.
3630 * But our reclaim could return 0, simply because due
3631 * to priority we are exposing a smaller subset of
3632 * memory to reclaim from. Consider this as a longer
3633 * term TODO.
3634 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003635 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003636 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003637 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003638 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003639 loop++;
3640 /*
3641 * Could not reclaim anything and there are no more
3642 * mem cgroups to try or we seem to be looping without
3643 * reclaiming anything.
3644 */
3645 if (!nr_reclaimed &&
3646 (next_mz == NULL ||
3647 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3648 break;
3649 } while (!nr_reclaimed);
3650 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003651 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003652 return nr_reclaimed;
3653}
3654
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003655/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003656 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003657 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3658 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003659static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003660 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003661{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003662 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003663 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003664 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003665 struct page *busy;
3666 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003667 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003668
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003669 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003670 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003671 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003672
Hugh Dickins1eb49272012-03-21 16:34:19 -07003673 loop = mz->lru_size[lru];
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003674 /* give some margin against EBUSY etc...*/
3675 loop += 256;
3676 busy = NULL;
3677 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003678 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003679 struct page *page;
3680
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003681 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003682 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003683 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003684 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003685 break;
3686 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003687 page = list_entry(list->prev, struct page, lru);
3688 if (busy == page) {
3689 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003690 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003691 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003692 continue;
3693 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003694 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003695
Johannes Weiner925b7672012-01-12 17:18:15 -08003696 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003697
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003698 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003699 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003700 break;
3701
3702 if (ret == -EBUSY || ret == -EINVAL) {
3703 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003704 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003705 cond_resched();
3706 } else
3707 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003708 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003709
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003710 if (!ret && !list_empty(list))
3711 return -EBUSY;
3712 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003713}
3714
3715/*
3716 * make mem_cgroup's charge to be 0 if there is no task.
3717 * This enables deleting this mem_cgroup.
3718 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003719static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003720{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003721 int ret;
3722 int node, zid, shrink;
3723 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003724 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003725
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003726 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003727
3728 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003729 /* should free all ? */
3730 if (free_all)
3731 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003732move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003733 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003734 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003735 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003736 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003737 ret = -EINTR;
3738 if (signal_pending(current))
3739 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003740 /* This is for making all *used* pages to be on LRU. */
3741 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003742 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003743 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003744 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003745 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003746 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003747 enum lru_list lru;
3748 for_each_lru(lru) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003749 ret = mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003750 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003751 if (ret)
3752 break;
3753 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003754 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003755 if (ret)
3756 break;
3757 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003758 mem_cgroup_end_move(memcg);
3759 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003760 /* it seems parent cgroup doesn't have enough mem */
3761 if (ret == -ENOMEM)
3762 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003763 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003764 /* "ret" should also be checked to ensure all lists are empty. */
Glauber Costa569530f2012-04-12 12:49:13 -07003765 } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003766out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003767 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003768 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003769
3770try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003771 /* returns EBUSY if there is a task or if we come here twice. */
3772 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003773 ret = -EBUSY;
3774 goto out;
3775 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003776 /* we call try-to-free pages for make this cgroup empty */
3777 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003778 /* try to free all pages in this cgroup */
3779 shrink = 1;
Glauber Costa569530f2012-04-12 12:49:13 -07003780 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003781 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003782
3783 if (signal_pending(current)) {
3784 ret = -EINTR;
3785 goto out;
3786 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003787 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003788 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003789 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003790 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003791 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003792 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003793 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003794
3795 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003796 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003797 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003798 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003799}
3800
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003801int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3802{
3803 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3804}
3805
3806
Balbir Singh18f59ea2009-01-07 18:08:07 -08003807static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3808{
3809 return mem_cgroup_from_cont(cont)->use_hierarchy;
3810}
3811
3812static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3813 u64 val)
3814{
3815 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003816 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003817 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003818 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003819
3820 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003821 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003822
3823 cgroup_lock();
3824 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003825 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003826 * in the child subtrees. If it is unset, then the change can
3827 * occur, provided the current cgroup has no children.
3828 *
3829 * For the root cgroup, parent_mem is NULL, we allow value to be
3830 * set if there are no children.
3831 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003832 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003833 (val == 1 || val == 0)) {
3834 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003835 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003836 else
3837 retval = -EBUSY;
3838 } else
3839 retval = -EINVAL;
3840 cgroup_unlock();
3841
3842 return retval;
3843}
3844
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003845
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003846static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003847 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003848{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003849 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003850 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003851
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003852 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003853 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003854 val += mem_cgroup_read_stat(iter, idx);
3855
3856 if (val < 0) /* race ? */
3857 val = 0;
3858 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003859}
3860
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003861static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003862{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003863 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003864
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003865 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003866 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003867 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003868 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003869 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003870 }
3871
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003872 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3873 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003874
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003875 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003876 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003877
3878 return val << PAGE_SHIFT;
3879}
3880
Paul Menage2c3daa72008-04-29 00:59:58 -07003881static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003882{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003883 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003884 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003885 int type, name;
3886
3887 type = MEMFILE_TYPE(cft->private);
3888 name = MEMFILE_ATTR(cft->private);
3889 switch (type) {
3890 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003891 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003892 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003893 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003894 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003895 break;
3896 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003897 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003898 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003899 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003900 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003901 break;
3902 default:
3903 BUG();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003904 }
3905 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003906}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003907/*
3908 * The user of this function is...
3909 * RES_LIMIT.
3910 */
Paul Menage856c13a2008-07-25 01:47:04 -07003911static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3912 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003913{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003914 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003915 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003916 unsigned long long val;
3917 int ret;
3918
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003919 type = MEMFILE_TYPE(cft->private);
3920 name = MEMFILE_ATTR(cft->private);
3921 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003922 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003923 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3924 ret = -EINVAL;
3925 break;
3926 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003927 /* This function does all necessary parse...reuse it */
3928 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003929 if (ret)
3930 break;
3931 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003932 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003933 else
3934 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003935 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003936 case RES_SOFT_LIMIT:
3937 ret = res_counter_memparse_write_strategy(buffer, &val);
3938 if (ret)
3939 break;
3940 /*
3941 * For memsw, soft limits are hard to implement in terms
3942 * of semantics, for now, we support soft limits for
3943 * control without swap
3944 */
3945 if (type == _MEM)
3946 ret = res_counter_set_soft_limit(&memcg->res, val);
3947 else
3948 ret = -EINVAL;
3949 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003950 default:
3951 ret = -EINVAL; /* should be BUG() ? */
3952 break;
3953 }
3954 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003955}
3956
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003957static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3958 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3959{
3960 struct cgroup *cgroup;
3961 unsigned long long min_limit, min_memsw_limit, tmp;
3962
3963 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3964 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3965 cgroup = memcg->css.cgroup;
3966 if (!memcg->use_hierarchy)
3967 goto out;
3968
3969 while (cgroup->parent) {
3970 cgroup = cgroup->parent;
3971 memcg = mem_cgroup_from_cont(cgroup);
3972 if (!memcg->use_hierarchy)
3973 break;
3974 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3975 min_limit = min(min_limit, tmp);
3976 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3977 min_memsw_limit = min(min_memsw_limit, tmp);
3978 }
3979out:
3980 *mem_limit = min_limit;
3981 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003982}
3983
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003984static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003985{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003986 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003987 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003988
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003989 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003990 type = MEMFILE_TYPE(event);
3991 name = MEMFILE_ATTR(event);
3992 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003993 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003994 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003995 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003996 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003997 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003998 break;
3999 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004000 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004001 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004002 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004003 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004004 break;
4005 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004006
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07004007 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004008}
4009
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004010static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4011 struct cftype *cft)
4012{
4013 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4014}
4015
Daisuke Nishimura02491442010-03-10 15:22:17 -08004016#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004017static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4018 struct cftype *cft, u64 val)
4019{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004020 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004021
4022 if (val >= (1 << NR_MOVE_TYPE))
4023 return -EINVAL;
4024 /*
4025 * We check this value several times in both in can_attach() and
4026 * attach(), so we need cgroup lock to prevent this value from being
4027 * inconsistent.
4028 */
4029 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004030 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004031 cgroup_unlock();
4032
4033 return 0;
4034}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004035#else
4036static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4037 struct cftype *cft, u64 val)
4038{
4039 return -ENOSYS;
4040}
4041#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004042
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004043
4044/* For read statistics */
4045enum {
4046 MCS_CACHE,
4047 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004048 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004049 MCS_PGPGIN,
4050 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004051 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07004052 MCS_PGFAULT,
4053 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004054 MCS_INACTIVE_ANON,
4055 MCS_ACTIVE_ANON,
4056 MCS_INACTIVE_FILE,
4057 MCS_ACTIVE_FILE,
4058 MCS_UNEVICTABLE,
4059 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004060};
4061
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004062struct mcs_total_stat {
4063 s64 stat[NR_MCS_STAT];
4064};
4065
4066struct {
4067 char *local_name;
4068 char *total_name;
4069} memcg_stat_strings[NR_MCS_STAT] = {
4070 {"cache", "total_cache"},
4071 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004072 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004073 {"pgpgin", "total_pgpgin"},
4074 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004075 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004076 {"pgfault", "total_pgfault"},
4077 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004078 {"inactive_anon", "total_inactive_anon"},
4079 {"active_anon", "total_active_anon"},
4080 {"inactive_file", "total_inactive_file"},
4081 {"active_file", "total_active_file"},
4082 {"unevictable", "total_unevictable"}
4083};
4084
4085
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004086static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004087mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004088{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004089 s64 val;
4090
4091 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004092 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004093 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004094 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004095 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004096 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004097 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004098 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004099 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004100 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004101 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004102 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004103 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004104 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4105 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004106 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004107 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004108 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004109 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004110
4111 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004112 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004113 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004114 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004115 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004116 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004117 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004118 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004119 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004120 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004121 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004122}
4123
4124static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004125mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004126{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004127 struct mem_cgroup *iter;
4128
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004129 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004130 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004131}
4132
Ying Han406eb0c2011-05-26 16:25:37 -07004133#ifdef CONFIG_NUMA
4134static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4135{
4136 int nid;
4137 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4138 unsigned long node_nr;
4139 struct cgroup *cont = m->private;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004140 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07004141
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004142 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004143 seq_printf(m, "total=%lu", total_nr);
4144 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004145 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004146 seq_printf(m, " N%d=%lu", nid, node_nr);
4147 }
4148 seq_putc(m, '\n');
4149
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004150 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004151 seq_printf(m, "file=%lu", file_nr);
4152 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004153 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004154 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004155 seq_printf(m, " N%d=%lu", nid, node_nr);
4156 }
4157 seq_putc(m, '\n');
4158
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004159 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004160 seq_printf(m, "anon=%lu", anon_nr);
4161 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004162 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004163 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004164 seq_printf(m, " N%d=%lu", nid, node_nr);
4165 }
4166 seq_putc(m, '\n');
4167
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004168 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004169 seq_printf(m, "unevictable=%lu", unevictable_nr);
4170 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004171 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004172 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004173 seq_printf(m, " N%d=%lu", nid, node_nr);
4174 }
4175 seq_putc(m, '\n');
4176 return 0;
4177}
4178#endif /* CONFIG_NUMA */
4179
Paul Menagec64745c2008-04-29 01:00:02 -07004180static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4181 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004182{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004183 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004184 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004185 int i;
4186
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004187 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004188 mem_cgroup_get_local_stat(memcg, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004189
Ying Han406eb0c2011-05-26 16:25:37 -07004190
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004191 for (i = 0; i < NR_MCS_STAT; i++) {
4192 if (i == MCS_SWAP && !do_swap_account)
4193 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004194 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004195 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004196
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004197 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004198 {
4199 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004200 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004201 cb->fill(cb, "hierarchical_memory_limit", limit);
4202 if (do_swap_account)
4203 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4204 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004205
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004206 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004207 mem_cgroup_get_total_stat(memcg, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004208 for (i = 0; i < NR_MCS_STAT; i++) {
4209 if (i == MCS_SWAP && !do_swap_account)
4210 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004211 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004212 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004213
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004214#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004215 {
4216 int nid, zid;
4217 struct mem_cgroup_per_zone *mz;
4218 unsigned long recent_rotated[2] = {0, 0};
4219 unsigned long recent_scanned[2] = {0, 0};
4220
4221 for_each_online_node(nid)
4222 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004223 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004224
4225 recent_rotated[0] +=
4226 mz->reclaim_stat.recent_rotated[0];
4227 recent_rotated[1] +=
4228 mz->reclaim_stat.recent_rotated[1];
4229 recent_scanned[0] +=
4230 mz->reclaim_stat.recent_scanned[0];
4231 recent_scanned[1] +=
4232 mz->reclaim_stat.recent_scanned[1];
4233 }
4234 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4235 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4236 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4237 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4238 }
4239#endif
4240
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004241 return 0;
4242}
4243
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004244static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4245{
4246 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4247
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004248 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004249}
4250
4251static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4252 u64 val)
4253{
4254 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4255 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004256
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004257 if (val > 100)
4258 return -EINVAL;
4259
4260 if (cgrp->parent == NULL)
4261 return -EINVAL;
4262
4263 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004264
4265 cgroup_lock();
4266
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004267 /* If under hierarchy, only empty-root can set this value */
4268 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004269 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4270 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004271 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004272 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004273
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004274 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004275
Li Zefan068b38c2009-01-15 13:51:26 -08004276 cgroup_unlock();
4277
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004278 return 0;
4279}
4280
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004281static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4282{
4283 struct mem_cgroup_threshold_ary *t;
4284 u64 usage;
4285 int i;
4286
4287 rcu_read_lock();
4288 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004289 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004290 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004291 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004292
4293 if (!t)
4294 goto unlock;
4295
4296 usage = mem_cgroup_usage(memcg, swap);
4297
4298 /*
4299 * current_threshold points to threshold just below usage.
4300 * If it's not true, a threshold was crossed after last
4301 * call of __mem_cgroup_threshold().
4302 */
Phil Carmody5407a562010-05-26 14:42:42 -07004303 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004304
4305 /*
4306 * Iterate backward over array of thresholds starting from
4307 * current_threshold and check if a threshold is crossed.
4308 * If none of thresholds below usage is crossed, we read
4309 * only one element of the array here.
4310 */
4311 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4312 eventfd_signal(t->entries[i].eventfd, 1);
4313
4314 /* i = current_threshold + 1 */
4315 i++;
4316
4317 /*
4318 * Iterate forward over array of thresholds starting from
4319 * current_threshold+1 and check if a threshold is crossed.
4320 * If none of thresholds above usage is crossed, we read
4321 * only one element of the array here.
4322 */
4323 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4324 eventfd_signal(t->entries[i].eventfd, 1);
4325
4326 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004327 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004328unlock:
4329 rcu_read_unlock();
4330}
4331
4332static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4333{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004334 while (memcg) {
4335 __mem_cgroup_threshold(memcg, false);
4336 if (do_swap_account)
4337 __mem_cgroup_threshold(memcg, true);
4338
4339 memcg = parent_mem_cgroup(memcg);
4340 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004341}
4342
4343static int compare_thresholds(const void *a, const void *b)
4344{
4345 const struct mem_cgroup_threshold *_a = a;
4346 const struct mem_cgroup_threshold *_b = b;
4347
4348 return _a->threshold - _b->threshold;
4349}
4350
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004351static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004352{
4353 struct mem_cgroup_eventfd_list *ev;
4354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004355 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004356 eventfd_signal(ev->eventfd, 1);
4357 return 0;
4358}
4359
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004360static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004361{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004362 struct mem_cgroup *iter;
4363
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004364 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004365 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004366}
4367
4368static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4369 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004370{
4371 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004372 struct mem_cgroup_thresholds *thresholds;
4373 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004374 int type = MEMFILE_TYPE(cft->private);
4375 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004376 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004377
4378 ret = res_counter_memparse_write_strategy(args, &threshold);
4379 if (ret)
4380 return ret;
4381
4382 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004383
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004384 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004385 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004386 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004387 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004388 else
4389 BUG();
4390
4391 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4392
4393 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004394 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004395 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4396
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398
4399 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004400 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004401 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004402 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004403 ret = -ENOMEM;
4404 goto unlock;
4405 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004406 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004407
4408 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004409 if (thresholds->primary) {
4410 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004411 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004412 }
4413
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004414 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004415 new->entries[size - 1].eventfd = eventfd;
4416 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004417
4418 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004419 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004420 compare_thresholds, NULL);
4421
4422 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004423 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004424 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004425 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004426 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004427 * new->current_threshold will not be used until
4428 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004429 * it here.
4430 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004431 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004432 }
4433 }
4434
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004435 /* Free old spare buffer and save old primary buffer as spare */
4436 kfree(thresholds->spare);
4437 thresholds->spare = thresholds->primary;
4438
4439 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004440
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004441 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004442 synchronize_rcu();
4443
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004444unlock:
4445 mutex_unlock(&memcg->thresholds_lock);
4446
4447 return ret;
4448}
4449
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004450static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004451 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004452{
4453 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004454 struct mem_cgroup_thresholds *thresholds;
4455 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004456 int type = MEMFILE_TYPE(cft->private);
4457 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004458 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004459
4460 mutex_lock(&memcg->thresholds_lock);
4461 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004462 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004463 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004464 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004465 else
4466 BUG();
4467
Anton Vorontsov371528c2012-02-24 05:14:46 +04004468 if (!thresholds->primary)
4469 goto unlock;
4470
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004471 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4472
4473 /* Check if a threshold crossed before removing */
4474 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4475
4476 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004477 size = 0;
4478 for (i = 0; i < thresholds->primary->size; i++) {
4479 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004480 size++;
4481 }
4482
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004483 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004484
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004485 /* Set thresholds array to NULL if we don't have thresholds */
4486 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004487 kfree(new);
4488 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004489 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004490 }
4491
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004492 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004493
4494 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004495 new->current_threshold = -1;
4496 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4497 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004498 continue;
4499
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004500 new->entries[j] = thresholds->primary->entries[i];
4501 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004502 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004503 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004504 * until rcu_assign_pointer(), so it's safe to increment
4505 * it here.
4506 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004507 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004508 }
4509 j++;
4510 }
4511
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004512swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004513 /* Swap primary and spare array */
4514 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004515 /* If all events are unregistered, free the spare array */
4516 if (!new) {
4517 kfree(thresholds->spare);
4518 thresholds->spare = NULL;
4519 }
4520
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004521 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004522
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004523 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004524 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004525unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004526 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004527}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004528
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004529static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4530 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4531{
4532 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4533 struct mem_cgroup_eventfd_list *event;
4534 int type = MEMFILE_TYPE(cft->private);
4535
4536 BUG_ON(type != _OOM_TYPE);
4537 event = kmalloc(sizeof(*event), GFP_KERNEL);
4538 if (!event)
4539 return -ENOMEM;
4540
Michal Hocko1af8efe2011-07-26 16:08:24 -07004541 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004542
4543 event->eventfd = eventfd;
4544 list_add(&event->list, &memcg->oom_notify);
4545
4546 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004547 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004548 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004549 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004550
4551 return 0;
4552}
4553
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004554static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004555 struct cftype *cft, struct eventfd_ctx *eventfd)
4556{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004557 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004558 struct mem_cgroup_eventfd_list *ev, *tmp;
4559 int type = MEMFILE_TYPE(cft->private);
4560
4561 BUG_ON(type != _OOM_TYPE);
4562
Michal Hocko1af8efe2011-07-26 16:08:24 -07004563 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004564
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004565 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004566 if (ev->eventfd == eventfd) {
4567 list_del(&ev->list);
4568 kfree(ev);
4569 }
4570 }
4571
Michal Hocko1af8efe2011-07-26 16:08:24 -07004572 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004573}
4574
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004575static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4576 struct cftype *cft, struct cgroup_map_cb *cb)
4577{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004578 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004579
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004580 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004582 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004583 cb->fill(cb, "under_oom", 1);
4584 else
4585 cb->fill(cb, "under_oom", 0);
4586 return 0;
4587}
4588
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004589static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4590 struct cftype *cft, u64 val)
4591{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004592 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004593 struct mem_cgroup *parent;
4594
4595 /* cannot set to root cgroup and only 0 and 1 are allowed */
4596 if (!cgrp->parent || !((val == 0) || (val == 1)))
4597 return -EINVAL;
4598
4599 parent = mem_cgroup_from_cont(cgrp->parent);
4600
4601 cgroup_lock();
4602 /* oom-kill-disable is a flag for subhierarchy. */
4603 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004604 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004605 cgroup_unlock();
4606 return -EINVAL;
4607 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004608 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004609 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004610 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004611 cgroup_unlock();
4612 return 0;
4613}
4614
Ying Han406eb0c2011-05-26 16:25:37 -07004615#ifdef CONFIG_NUMA
4616static const struct file_operations mem_control_numa_stat_file_operations = {
4617 .read = seq_read,
4618 .llseek = seq_lseek,
4619 .release = single_release,
4620};
4621
4622static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4623{
4624 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4625
4626 file->f_op = &mem_control_numa_stat_file_operations;
4627 return single_open(file, mem_control_numa_stat_show, cont);
4628}
4629#endif /* CONFIG_NUMA */
4630
Glauber Costae5671df2011-12-11 21:47:01 +00004631#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae5671df2011-12-11 21:47:01 +00004632static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4633{
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004634 /*
4635 * Part of this would be better living in a separate allocation
4636 * function, leaving us with just the cgroup tree population work.
4637 * We, however, depend on state such as network's proto_list that
4638 * is only initialized after cgroup creation. I found the less
4639 * cumbersome way to deal with it to defer it all to populate time
4640 */
Glauber Costa65c64ce2011-12-22 01:02:27 +00004641 return mem_cgroup_sockets_init(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004642};
4643
Li Zefan761b3ef2012-01-31 13:47:36 +08004644static void kmem_cgroup_destroy(struct cgroup *cont)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004645{
Li Zefan761b3ef2012-01-31 13:47:36 +08004646 mem_cgroup_sockets_destroy(cont);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004647}
Glauber Costae5671df2011-12-11 21:47:01 +00004648#else
4649static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4650{
4651 return 0;
4652}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004653
Li Zefan761b3ef2012-01-31 13:47:36 +08004654static void kmem_cgroup_destroy(struct cgroup *cont)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004655{
4656}
Glauber Costae5671df2011-12-11 21:47:01 +00004657#endif
4658
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004659static struct cftype mem_cgroup_files[] = {
4660 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004661 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004662 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004663 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004664 .register_event = mem_cgroup_usage_register_event,
4665 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004666 },
4667 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004668 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004669 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004670 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004671 .read_u64 = mem_cgroup_read,
4672 },
4673 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004674 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004675 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004676 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004677 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004678 },
4679 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004680 .name = "soft_limit_in_bytes",
4681 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4682 .write_string = mem_cgroup_write,
4683 .read_u64 = mem_cgroup_read,
4684 },
4685 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004686 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004687 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004688 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004689 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004690 },
Balbir Singh8697d332008-02-07 00:13:59 -08004691 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004692 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004693 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004694 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004695 {
4696 .name = "force_empty",
4697 .trigger = mem_cgroup_force_empty_write,
4698 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004699 {
4700 .name = "use_hierarchy",
4701 .write_u64 = mem_cgroup_hierarchy_write,
4702 .read_u64 = mem_cgroup_hierarchy_read,
4703 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004704 {
4705 .name = "swappiness",
4706 .read_u64 = mem_cgroup_swappiness_read,
4707 .write_u64 = mem_cgroup_swappiness_write,
4708 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004709 {
4710 .name = "move_charge_at_immigrate",
4711 .read_u64 = mem_cgroup_move_charge_read,
4712 .write_u64 = mem_cgroup_move_charge_write,
4713 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004714 {
4715 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004716 .read_map = mem_cgroup_oom_control_read,
4717 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004718 .register_event = mem_cgroup_oom_register_event,
4719 .unregister_event = mem_cgroup_oom_unregister_event,
4720 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4721 },
Ying Han406eb0c2011-05-26 16:25:37 -07004722#ifdef CONFIG_NUMA
4723 {
4724 .name = "numa_stat",
4725 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004726 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004727 },
4728#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004729};
4730
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004731#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4732static struct cftype memsw_cgroup_files[] = {
4733 {
4734 .name = "memsw.usage_in_bytes",
4735 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4736 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004737 .register_event = mem_cgroup_usage_register_event,
4738 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004739 },
4740 {
4741 .name = "memsw.max_usage_in_bytes",
4742 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4743 .trigger = mem_cgroup_reset,
4744 .read_u64 = mem_cgroup_read,
4745 },
4746 {
4747 .name = "memsw.limit_in_bytes",
4748 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4749 .write_string = mem_cgroup_write,
4750 .read_u64 = mem_cgroup_read,
4751 },
4752 {
4753 .name = "memsw.failcnt",
4754 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4755 .trigger = mem_cgroup_reset,
4756 .read_u64 = mem_cgroup_read,
4757 },
4758};
4759
4760static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4761{
4762 if (!do_swap_account)
4763 return 0;
4764 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4765 ARRAY_SIZE(memsw_cgroup_files));
4766};
4767#else
4768static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4769{
4770 return 0;
4771}
4772#endif
4773
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004774static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004775{
4776 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004777 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004778 enum lru_list lru;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004779 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004780 /*
4781 * This routine is called against possible nodes.
4782 * But it's BUG to call kmalloc() against offline node.
4783 *
4784 * TODO: this routine can waste much memory for nodes which will
4785 * never be onlined. It's better to use memory hotplug callback
4786 * function.
4787 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004788 if (!node_state(node, N_NORMAL_MEMORY))
4789 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004790 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004791 if (!pn)
4792 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004793
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004794 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4795 mz = &pn->zoneinfo[zone];
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004796 for_each_lru(lru)
4797 INIT_LIST_HEAD(&mz->lruvec.lists[lru]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004798 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004799 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004800 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004801 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004802 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004803 return 0;
4804}
4805
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004806static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004807{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004808 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004809}
4810
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004811static struct mem_cgroup *mem_cgroup_alloc(void)
4812{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004813 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004814 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004815
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004816 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004817 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004818 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004819 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004820 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004821
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004822 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004823 return NULL;
4824
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004825 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4826 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004827 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004828 spin_lock_init(&memcg->pcp_counter_lock);
4829 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004830
4831out_free:
4832 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004833 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004834 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004835 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004836 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004837}
4838
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004839/*
Hugh Dickins59927fb2012-03-15 15:17:07 -07004840 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4841 * but in process context. The work_freeing structure is overlaid
4842 * on the rcu_freeing structure, which itself is overlaid on memsw.
4843 */
4844static void vfree_work(struct work_struct *work)
4845{
4846 struct mem_cgroup *memcg;
4847
4848 memcg = container_of(work, struct mem_cgroup, work_freeing);
4849 vfree(memcg);
4850}
4851static void vfree_rcu(struct rcu_head *rcu_head)
4852{
4853 struct mem_cgroup *memcg;
4854
4855 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4856 INIT_WORK(&memcg->work_freeing, vfree_work);
4857 schedule_work(&memcg->work_freeing);
4858}
4859
4860/*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004861 * At destroying mem_cgroup, references from swap_cgroup can remain.
4862 * (scanning all at force_empty is too costly...)
4863 *
4864 * Instead of clearing all references at force_empty, we remember
4865 * the number of reference from swap_cgroup and free mem_cgroup when
4866 * it goes down to 0.
4867 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004868 * Removal of cgroup itself succeeds regardless of refs from swap.
4869 */
4870
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004871static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004872{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004873 int node;
4874
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004875 mem_cgroup_remove_from_trees(memcg);
4876 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004877
Bob Liu3ed28fa2012-01-12 17:19:04 -08004878 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004879 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004880
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004881 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004882 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004883 kfree_rcu(memcg, rcu_freeing);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004884 else
Hugh Dickins59927fb2012-03-15 15:17:07 -07004885 call_rcu(&memcg->rcu_freeing, vfree_rcu);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004886}
4887
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004888static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004890 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004891}
4892
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004893static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004894{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004895 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4896 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4897 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004898 if (parent)
4899 mem_cgroup_put(parent);
4900 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004901}
4902
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004903static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004904{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004905 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004906}
4907
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004908/*
4909 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4910 */
Glauber Costae1aab162011-12-11 21:47:03 +00004911struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004912{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004913 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004914 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004915 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004916}
Glauber Costae1aab162011-12-11 21:47:03 +00004917EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004918
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004919#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4920static void __init enable_swap_cgroup(void)
4921{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08004922 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004923 do_swap_account = 1;
4924}
4925#else
4926static void __init enable_swap_cgroup(void)
4927{
4928}
4929#endif
4930
Balbir Singhf64c3f52009-09-23 15:56:37 -07004931static int mem_cgroup_soft_limit_tree_init(void)
4932{
4933 struct mem_cgroup_tree_per_node *rtpn;
4934 struct mem_cgroup_tree_per_zone *rtpz;
4935 int tmp, node, zone;
4936
Bob Liu3ed28fa2012-01-12 17:19:04 -08004937 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004938 tmp = node;
4939 if (!node_state(node, N_NORMAL_MEMORY))
4940 tmp = -1;
4941 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4942 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004943 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004944
4945 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4946
4947 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4948 rtpz = &rtpn->rb_tree_per_zone[zone];
4949 rtpz->rb_root = RB_ROOT;
4950 spin_lock_init(&rtpz->lock);
4951 }
4952 }
4953 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004954
4955err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004956 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004957 if (!soft_limit_tree.rb_tree_per_node[node])
4958 break;
4959 kfree(soft_limit_tree.rb_tree_per_node[node]);
4960 soft_limit_tree.rb_tree_per_node[node] = NULL;
4961 }
4962 return 1;
4963
Balbir Singhf64c3f52009-09-23 15:56:37 -07004964}
4965
Li Zefan0eb253e2009-01-15 13:51:25 -08004966static struct cgroup_subsys_state * __ref
Li Zefan761b3ef2012-01-31 13:47:36 +08004967mem_cgroup_create(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004968{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004969 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004970 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004971 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004972
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004973 memcg = mem_cgroup_alloc();
4974 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004975 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004976
Bob Liu3ed28fa2012-01-12 17:19:04 -08004977 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004978 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004979 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004980
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004981 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004982 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004983 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004984 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004985 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004986 if (mem_cgroup_soft_limit_tree_init())
4987 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004988 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004989 for_each_possible_cpu(cpu) {
4990 struct memcg_stock_pcp *stock =
4991 &per_cpu(memcg_stock, cpu);
4992 INIT_WORK(&stock->work, drain_local_stock);
4993 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004994 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004995 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004996 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004997 memcg->use_hierarchy = parent->use_hierarchy;
4998 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004999 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005000
Balbir Singh18f59ea2009-01-07 18:08:07 -08005001 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005002 res_counter_init(&memcg->res, &parent->res);
5003 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005004 /*
5005 * We increment refcnt of the parent to ensure that we can
5006 * safely access it on res_counter_charge/uncharge.
5007 * This refcnt will be decremented when freeing this
5008 * mem_cgroup(see mem_cgroup_put).
5009 */
5010 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005011 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005012 res_counter_init(&memcg->res, NULL);
5013 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005014 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005015 memcg->last_scanned_node = MAX_NUMNODES;
5016 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08005017
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005018 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005019 memcg->swappiness = mem_cgroup_swappiness(parent);
5020 atomic_set(&memcg->refcnt, 1);
5021 memcg->move_charge_at_immigrate = 0;
5022 mutex_init(&memcg->thresholds_lock);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07005023 spin_lock_init(&memcg->move_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005024 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005025free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005026 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005027 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005028}
5029
Li Zefan761b3ef2012-01-31 13:47:36 +08005030static int mem_cgroup_pre_destroy(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005031{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005032 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005033
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005034 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005035}
5036
Li Zefan761b3ef2012-01-31 13:47:36 +08005037static void mem_cgroup_destroy(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005038{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005039 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005040
Li Zefan761b3ef2012-01-31 13:47:36 +08005041 kmem_cgroup_destroy(cont);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005042
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005043 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005044}
5045
5046static int mem_cgroup_populate(struct cgroup_subsys *ss,
5047 struct cgroup *cont)
5048{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005049 int ret;
5050
5051 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
5052 ARRAY_SIZE(mem_cgroup_files));
5053
5054 if (!ret)
5055 ret = register_memsw_files(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00005056
5057 if (!ret)
5058 ret = register_kmem_files(cont, ss);
5059
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005060 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005061}
5062
Daisuke Nishimura02491442010-03-10 15:22:17 -08005063#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005064/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005065#define PRECHARGE_COUNT_AT_ONCE 256
5066static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005067{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005068 int ret = 0;
5069 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005070 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005071
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005073 mc.precharge += count;
5074 /* we don't need css_get for root */
5075 return ret;
5076 }
5077 /* try to charge at once */
5078 if (count > 1) {
5079 struct res_counter *dummy;
5080 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005081 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005082 * by cgroup_lock_live_cgroup() that it is not removed and we
5083 * are still under the same cgroup_mutex. So we can postpone
5084 * css_get().
5085 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005086 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005087 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005088 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005089 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005090 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005091 goto one_by_one;
5092 }
5093 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005094 return ret;
5095 }
5096one_by_one:
5097 /* fall back to one by one charge */
5098 while (count--) {
5099 if (signal_pending(current)) {
5100 ret = -EINTR;
5101 break;
5102 }
5103 if (!batch_count--) {
5104 batch_count = PRECHARGE_COUNT_AT_ONCE;
5105 cond_resched();
5106 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005107 ret = __mem_cgroup_try_charge(NULL,
5108 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005109 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005110 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005111 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005112 mc.precharge++;
5113 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005114 return ret;
5115}
5116
5117/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005118 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005119 * @vma: the vma the pte to be checked belongs
5120 * @addr: the address corresponding to the pte to be checked
5121 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005122 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005123 *
5124 * Returns
5125 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5126 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5127 * move charge. if @target is not NULL, the page is stored in target->page
5128 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005129 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5130 * target for charge migration. if @target is not NULL, the entry is stored
5131 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005132 *
5133 * Called with pte lock held.
5134 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005135union mc_target {
5136 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005137 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005138};
5139
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005140enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005141 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005142 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005143 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005144};
5145
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005146static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5147 unsigned long addr, pte_t ptent)
5148{
5149 struct page *page = vm_normal_page(vma, addr, ptent);
5150
5151 if (!page || !page_mapped(page))
5152 return NULL;
5153 if (PageAnon(page)) {
5154 /* we don't move shared anon */
Hugh Dickinsbe22aec2012-03-09 13:37:32 -08005155 if (!move_anon() || page_mapcount(page) > 2)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005156 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005157 } else if (!move_file())
5158 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005159 return NULL;
5160 if (!get_page_unless_zero(page))
5161 return NULL;
5162
5163 return page;
5164}
5165
5166static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5167 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5168{
5169 int usage_count;
5170 struct page *page = NULL;
5171 swp_entry_t ent = pte_to_swp_entry(ptent);
5172
5173 if (!move_anon() || non_swap_entry(ent))
5174 return NULL;
5175 usage_count = mem_cgroup_count_swap_user(ent, &page);
5176 if (usage_count > 1) { /* we don't move shared anon */
5177 if (page)
5178 put_page(page);
5179 return NULL;
5180 }
5181 if (do_swap_account)
5182 entry->val = ent.val;
5183
5184 return page;
5185}
5186
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005187static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5188 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5189{
5190 struct page *page = NULL;
5191 struct inode *inode;
5192 struct address_space *mapping;
5193 pgoff_t pgoff;
5194
5195 if (!vma->vm_file) /* anonymous vma */
5196 return NULL;
5197 if (!move_file())
5198 return NULL;
5199
5200 inode = vma->vm_file->f_path.dentry->d_inode;
5201 mapping = vma->vm_file->f_mapping;
5202 if (pte_none(ptent))
5203 pgoff = linear_page_index(vma, addr);
5204 else /* pte_file(ptent) is true */
5205 pgoff = pte_to_pgoff(ptent);
5206
5207 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005208 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005209
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005210#ifdef CONFIG_SWAP
5211 /* shmem/tmpfs may report page out on swap: account for that too. */
5212 if (radix_tree_exceptional_entry(page)) {
5213 swp_entry_t swap = radix_to_swp_entry(page);
5214 if (do_swap_account)
5215 *entry = swap;
5216 page = find_get_page(&swapper_space, swap.val);
5217 }
5218#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005219 return page;
5220}
5221
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005222static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005223 unsigned long addr, pte_t ptent, union mc_target *target)
5224{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005225 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005226 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005227 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005228 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005229
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005230 if (pte_present(ptent))
5231 page = mc_handle_present_pte(vma, addr, ptent);
5232 else if (is_swap_pte(ptent))
5233 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005234 else if (pte_none(ptent) || pte_file(ptent))
5235 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005236
5237 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005238 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005239 if (page) {
5240 pc = lookup_page_cgroup(page);
5241 /*
5242 * Do only loose check w/o page_cgroup lock.
5243 * mem_cgroup_move_account() checks the pc is valid or not under
5244 * the lock.
5245 */
5246 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5247 ret = MC_TARGET_PAGE;
5248 if (target)
5249 target->page = page;
5250 }
5251 if (!ret || !target)
5252 put_page(page);
5253 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005254 /* There is a swap entry and a page doesn't exist or isn't charged */
5255 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005256 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005257 ret = MC_TARGET_SWAP;
5258 if (target)
5259 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005260 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005261 return ret;
5262}
5263
Naoya Horiguchi12724852012-03-21 16:34:28 -07005264#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5265/*
5266 * We don't consider swapping or file mapped pages because THP does not
5267 * support them for now.
5268 * Caller should make sure that pmd_trans_huge(pmd) is true.
5269 */
5270static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5271 unsigned long addr, pmd_t pmd, union mc_target *target)
5272{
5273 struct page *page = NULL;
5274 struct page_cgroup *pc;
5275 enum mc_target_type ret = MC_TARGET_NONE;
5276
5277 page = pmd_page(pmd);
5278 VM_BUG_ON(!page || !PageHead(page));
5279 if (!move_anon())
5280 return ret;
5281 pc = lookup_page_cgroup(page);
5282 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5283 ret = MC_TARGET_PAGE;
5284 if (target) {
5285 get_page(page);
5286 target->page = page;
5287 }
5288 }
5289 return ret;
5290}
5291#else
5292static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5293 unsigned long addr, pmd_t pmd, union mc_target *target)
5294{
5295 return MC_TARGET_NONE;
5296}
5297#endif
5298
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005299static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5300 unsigned long addr, unsigned long end,
5301 struct mm_walk *walk)
5302{
5303 struct vm_area_struct *vma = walk->private;
5304 pte_t *pte;
5305 spinlock_t *ptl;
5306
Naoya Horiguchi12724852012-03-21 16:34:28 -07005307 if (pmd_trans_huge_lock(pmd, vma) == 1) {
5308 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5309 mc.precharge += HPAGE_PMD_NR;
5310 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005311 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005312 }
Dave Hansen03319322011-03-22 16:32:56 -07005313
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005314 if (pmd_trans_unstable(pmd))
5315 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005316 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5317 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005318 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005319 mc.precharge++; /* increment precharge temporarily */
5320 pte_unmap_unlock(pte - 1, ptl);
5321 cond_resched();
5322
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005323 return 0;
5324}
5325
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005326static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5327{
5328 unsigned long precharge;
5329 struct vm_area_struct *vma;
5330
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005331 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005332 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5333 struct mm_walk mem_cgroup_count_precharge_walk = {
5334 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5335 .mm = mm,
5336 .private = vma,
5337 };
5338 if (is_vm_hugetlb_page(vma))
5339 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005340 walk_page_range(vma->vm_start, vma->vm_end,
5341 &mem_cgroup_count_precharge_walk);
5342 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005343 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005344
5345 precharge = mc.precharge;
5346 mc.precharge = 0;
5347
5348 return precharge;
5349}
5350
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005351static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5352{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005353 unsigned long precharge = mem_cgroup_count_precharge(mm);
5354
5355 VM_BUG_ON(mc.moving_task);
5356 mc.moving_task = current;
5357 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005358}
5359
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005360/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5361static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005362{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005363 struct mem_cgroup *from = mc.from;
5364 struct mem_cgroup *to = mc.to;
5365
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005366 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005367 if (mc.precharge) {
5368 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5369 mc.precharge = 0;
5370 }
5371 /*
5372 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5373 * we must uncharge here.
5374 */
5375 if (mc.moved_charge) {
5376 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5377 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005378 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005379 /* we must fixup refcnts and charges */
5380 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005381 /* uncharge swap account from the old cgroup */
5382 if (!mem_cgroup_is_root(mc.from))
5383 res_counter_uncharge(&mc.from->memsw,
5384 PAGE_SIZE * mc.moved_swap);
5385 __mem_cgroup_put(mc.from, mc.moved_swap);
5386
5387 if (!mem_cgroup_is_root(mc.to)) {
5388 /*
5389 * we charged both to->res and to->memsw, so we should
5390 * uncharge to->res.
5391 */
5392 res_counter_uncharge(&mc.to->res,
5393 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005394 }
5395 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005396 mc.moved_swap = 0;
5397 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005398 memcg_oom_recover(from);
5399 memcg_oom_recover(to);
5400 wake_up_all(&mc.waitq);
5401}
5402
5403static void mem_cgroup_clear_mc(void)
5404{
5405 struct mem_cgroup *from = mc.from;
5406
5407 /*
5408 * we must clear moving_task before waking up waiters at the end of
5409 * task migration.
5410 */
5411 mc.moving_task = NULL;
5412 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005413 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005414 mc.from = NULL;
5415 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005416 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005417 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005418}
5419
Li Zefan761b3ef2012-01-31 13:47:36 +08005420static int mem_cgroup_can_attach(struct cgroup *cgroup,
5421 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005422{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005423 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005424 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005425 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005426
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005427 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005428 struct mm_struct *mm;
5429 struct mem_cgroup *from = mem_cgroup_from_task(p);
5430
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005431 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005432
5433 mm = get_task_mm(p);
5434 if (!mm)
5435 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005436 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005437 if (mm->owner == p) {
5438 VM_BUG_ON(mc.from);
5439 VM_BUG_ON(mc.to);
5440 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005441 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005442 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005443 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005444 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005445 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005446 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005447 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005448 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005449
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005450 ret = mem_cgroup_precharge_mc(mm);
5451 if (ret)
5452 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005453 }
5454 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005455 }
5456 return ret;
5457}
5458
Li Zefan761b3ef2012-01-31 13:47:36 +08005459static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5460 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005461{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005462 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005463}
5464
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005465static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5466 unsigned long addr, unsigned long end,
5467 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005468{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005469 int ret = 0;
5470 struct vm_area_struct *vma = walk->private;
5471 pte_t *pte;
5472 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005473 enum mc_target_type target_type;
5474 union mc_target target;
5475 struct page *page;
5476 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005477
Naoya Horiguchi12724852012-03-21 16:34:28 -07005478 /*
5479 * We don't take compound_lock() here but no race with splitting thp
5480 * happens because:
5481 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5482 * under splitting, which means there's no concurrent thp split,
5483 * - if another thread runs into split_huge_page() just after we
5484 * entered this if-block, the thread must wait for page table lock
5485 * to be unlocked in __split_huge_page_splitting(), where the main
5486 * part of thp split is not executed yet.
5487 */
5488 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005489 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005490 spin_unlock(&vma->vm_mm->page_table_lock);
5491 return 0;
5492 }
5493 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5494 if (target_type == MC_TARGET_PAGE) {
5495 page = target.page;
5496 if (!isolate_lru_page(page)) {
5497 pc = lookup_page_cgroup(page);
5498 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5499 pc, mc.from, mc.to,
5500 false)) {
5501 mc.precharge -= HPAGE_PMD_NR;
5502 mc.moved_charge += HPAGE_PMD_NR;
5503 }
5504 putback_lru_page(page);
5505 }
5506 put_page(page);
5507 }
5508 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005509 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005510 }
5511
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005512 if (pmd_trans_unstable(pmd))
5513 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005514retry:
5515 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5516 for (; addr != end; addr += PAGE_SIZE) {
5517 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005518 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005519
5520 if (!mc.precharge)
5521 break;
5522
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005523 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005524 case MC_TARGET_PAGE:
5525 page = target.page;
5526 if (isolate_lru_page(page))
5527 goto put;
5528 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005529 if (!mem_cgroup_move_account(page, 1, pc,
5530 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005531 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005532 /* we uncharge from mc.from later. */
5533 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005534 }
5535 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005536put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005537 put_page(page);
5538 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005539 case MC_TARGET_SWAP:
5540 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005541 if (!mem_cgroup_move_swap_account(ent,
5542 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005543 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005544 /* we fixup refcnts and charges later. */
5545 mc.moved_swap++;
5546 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005547 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005548 default:
5549 break;
5550 }
5551 }
5552 pte_unmap_unlock(pte - 1, ptl);
5553 cond_resched();
5554
5555 if (addr != end) {
5556 /*
5557 * We have consumed all precharges we got in can_attach().
5558 * We try charge one by one, but don't do any additional
5559 * charges to mc.to if we have failed in charge once in attach()
5560 * phase.
5561 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005562 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005563 if (!ret)
5564 goto retry;
5565 }
5566
5567 return ret;
5568}
5569
5570static void mem_cgroup_move_charge(struct mm_struct *mm)
5571{
5572 struct vm_area_struct *vma;
5573
5574 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005575retry:
5576 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5577 /*
5578 * Someone who are holding the mmap_sem might be waiting in
5579 * waitq. So we cancel all extra charges, wake up all waiters,
5580 * and retry. Because we cancel precharges, we might not be able
5581 * to move enough charges, but moving charge is a best-effort
5582 * feature anyway, so it wouldn't be a big problem.
5583 */
5584 __mem_cgroup_clear_mc();
5585 cond_resched();
5586 goto retry;
5587 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005588 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5589 int ret;
5590 struct mm_walk mem_cgroup_move_charge_walk = {
5591 .pmd_entry = mem_cgroup_move_charge_pte_range,
5592 .mm = mm,
5593 .private = vma,
5594 };
5595 if (is_vm_hugetlb_page(vma))
5596 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005597 ret = walk_page_range(vma->vm_start, vma->vm_end,
5598 &mem_cgroup_move_charge_walk);
5599 if (ret)
5600 /*
5601 * means we have consumed all precharges and failed in
5602 * doing additional charge. Just abandon here.
5603 */
5604 break;
5605 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005606 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005607}
5608
Li Zefan761b3ef2012-01-31 13:47:36 +08005609static void mem_cgroup_move_task(struct cgroup *cont,
5610 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005611{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005612 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005613 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005614
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005615 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005616 if (mc.to)
5617 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005618 mmput(mm);
5619 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005620 if (mc.to)
5621 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005622}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005623#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08005624static int mem_cgroup_can_attach(struct cgroup *cgroup,
5625 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005626{
5627 return 0;
5628}
Li Zefan761b3ef2012-01-31 13:47:36 +08005629static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5630 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005631{
5632}
Li Zefan761b3ef2012-01-31 13:47:36 +08005633static void mem_cgroup_move_task(struct cgroup *cont,
5634 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005635{
5636}
5637#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005638
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005639struct cgroup_subsys mem_cgroup_subsys = {
5640 .name = "memory",
5641 .subsys_id = mem_cgroup_subsys_id,
5642 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005643 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005644 .destroy = mem_cgroup_destroy,
5645 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005646 .can_attach = mem_cgroup_can_attach,
5647 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005648 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005649 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005650 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005651};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005652
5653#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005654static int __init enable_swap_account(char *s)
5655{
5656 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005657 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005658 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005659 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005660 really_do_swap_account = 0;
5661 return 1;
5662}
Michal Hockoa2c89902011-05-24 17:12:50 -07005663__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005664
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005665#endif