blob: 4f71219cc53e955bc78612de37b94abce81f6aed [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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki453a9bf32011-07-08 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 Hiroyuki453a9bf32011-07-08 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki453a9bf32011-07-08 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki8a9478ca2009-06-17 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyuki8c7c6e342009-01-07 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 Hiroyukib24028572012-03-21 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 Hiroyukib24028572012-03-21 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 Hiroyukib24028572012-03-21 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 Dickinsf156ab932012-03-21 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 Dickinsf156ab932012-03-21 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 Hiroyuki453a9bf32011-07-08 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 Hiroyuki453a9bf32011-07-08 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 Takahashif8d665422009-01-07 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
1119void mem_cgroup_lru_del(struct page *page)
1120{
1121 mem_cgroup_lru_del_list(page, page_lru(page));
1122}
1123
1124/**
1125 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1126 * @zone: zone of the page
1127 * @page: the page
1128 * @from: current lru
1129 * @to: target lru
1130 *
1131 * This function accounts for @page being moved between the lrus @from
1132 * and @to, and returns the lruvec for the given @zone and the memcg
1133 * @page is charged to.
1134 *
1135 * The callsite is then responsible for physically relinking
1136 * @page->lru to the returned lruvec->lists[@to].
1137 */
1138struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone,
1139 struct page *page,
1140 enum lru_list from,
1141 enum lru_list to)
1142{
1143 /* XXX: Optimize this, especially for @from == @to */
1144 mem_cgroup_lru_del_list(page, from);
1145 return mem_cgroup_lru_add_list(zone, page, to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001146}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001147
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001148/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001149 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001150 * hierarchy subtree
1151 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001152bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1153 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001154{
Johannes Weiner91c637342012-05-29 15:06:24 -07001155 if (root_memcg == memcg)
1156 return true;
1157 if (!root_memcg->use_hierarchy)
1158 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001159 return css_is_ancestor(&memcg->css, &root_memcg->css);
1160}
1161
1162static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1163 struct mem_cgroup *memcg)
1164{
1165 bool ret;
1166
Johannes Weiner91c637342012-05-29 15:06:24 -07001167 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001168 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001169 rcu_read_unlock();
1170 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001171}
1172
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001173int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001174{
1175 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001176 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001177 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001178
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001179 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001180 if (p) {
1181 curr = try_get_mem_cgroup_from_mm(p->mm);
1182 task_unlock(p);
1183 } else {
1184 /*
1185 * All threads may have already detached their mm's, but the oom
1186 * killer still needs to detect if they have already been oom
1187 * killed to prevent needlessly killing additional tasks.
1188 */
1189 task_lock(task);
1190 curr = mem_cgroup_from_task(task);
1191 if (curr)
1192 css_get(&curr->css);
1193 task_unlock(task);
1194 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001195 if (!curr)
1196 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001197 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001198 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001199 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001200 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1201 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001202 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001203 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001204 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001205 return ret;
1206}
1207
Johannes Weiner9b272972011-11-02 13:38:23 -07001208int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001209{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001210 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001211 int nid = zone_to_nid(zone);
1212 int zid = zone_idx(zone);
1213 unsigned long inactive;
1214 unsigned long active;
1215 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001216
Johannes Weiner9b272972011-11-02 13:38:23 -07001217 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1218 BIT(LRU_INACTIVE_ANON));
1219 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1220 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001221
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001222 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1223 if (gb)
1224 inactive_ratio = int_sqrt(10 * gb);
1225 else
1226 inactive_ratio = 1;
1227
Johannes Weiner9b272972011-11-02 13:38:23 -07001228 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001229}
1230
Johannes Weiner9b272972011-11-02 13:38:23 -07001231int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001232{
1233 unsigned long active;
1234 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001235 int zid = zone_idx(zone);
1236 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001237
Johannes Weiner9b272972011-11-02 13:38:23 -07001238 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1239 BIT(LRU_INACTIVE_FILE));
1240 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1241 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001242
1243 return (active > inactive);
1244}
1245
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001246struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1247 struct zone *zone)
1248{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001249 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001250 int zid = zone_idx(zone);
1251 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1252
1253 return &mz->reclaim_stat;
1254}
1255
1256struct zone_reclaim_stat *
1257mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1258{
1259 struct page_cgroup *pc;
1260 struct mem_cgroup_per_zone *mz;
1261
1262 if (mem_cgroup_disabled())
1263 return NULL;
1264
1265 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001266 if (!PageCgroupUsed(pc))
1267 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001268 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1269 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001270 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001271 return &mz->reclaim_stat;
1272}
1273
Balbir Singh6d61ef42009-01-07 18:08:06 -08001274#define mem_cgroup_from_res_counter(counter, member) \
1275 container_of(counter, struct mem_cgroup, member)
1276
Johannes Weiner19942822011-02-01 15:52:43 -08001277/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001278 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1279 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001280 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001281 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001282 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001283 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001284static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001285{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001286 unsigned long long margin;
1287
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001289 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001290 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001291 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001292}
1293
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001294int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001295{
1296 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001297
1298 /* root ? */
1299 if (cgrp->parent == NULL)
1300 return vm_swappiness;
1301
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001302 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001303}
1304
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001305/*
1306 * memcg->moving_account is used for checking possibility that some thread is
1307 * calling move_account(). When a thread on CPU-A starts moving pages under
1308 * a memcg, other threads should check memcg->moving_account under
1309 * rcu_read_lock(), like this:
1310 *
1311 * CPU-A CPU-B
1312 * rcu_read_lock()
1313 * memcg->moving_account+1 if (memcg->mocing_account)
1314 * take heavy locks.
1315 * synchronize_rcu() update something.
1316 * rcu_read_unlock()
1317 * start move here.
1318 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001319
1320/* for quick checking without looking up memcg */
1321atomic_t memcg_moving __read_mostly;
1322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001323static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001324{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001325 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001326 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001327 synchronize_rcu();
1328}
1329
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001330static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001331{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001332 /*
1333 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1334 * We check NULL in callee rather than caller.
1335 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001336 if (memcg) {
1337 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001338 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001339 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001340}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001341
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001342/*
1343 * 2 routines for checking "mem" is under move_account() or not.
1344 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001345 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1346 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001347 * pc->mem_cgroup may be overwritten.
1348 *
1349 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1350 * under hierarchy of moving cgroups. This is for
1351 * waiting at hith-memory prressure caused by "move".
1352 */
1353
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001354static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001355{
1356 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001357 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001358}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001359
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001360static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001361{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001362 struct mem_cgroup *from;
1363 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001364 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001365 /*
1366 * Unlike task_move routines, we access mc.to, mc.from not under
1367 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1368 */
1369 spin_lock(&mc.lock);
1370 from = mc.from;
1371 to = mc.to;
1372 if (!from)
1373 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001374
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001375 ret = mem_cgroup_same_or_subtree(memcg, from)
1376 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001377unlock:
1378 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001379 return ret;
1380}
1381
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001382static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001383{
1384 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001385 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001386 DEFINE_WAIT(wait);
1387 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1388 /* moving charge context might have finished. */
1389 if (mc.moving_task)
1390 schedule();
1391 finish_wait(&mc.waitq, &wait);
1392 return true;
1393 }
1394 }
1395 return false;
1396}
1397
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001398/*
1399 * Take this lock when
1400 * - a code tries to modify page's memcg while it's USED.
1401 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001402 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001403 */
1404static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1405 unsigned long *flags)
1406{
1407 spin_lock_irqsave(&memcg->move_lock, *flags);
1408}
1409
1410static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1411 unsigned long *flags)
1412{
1413 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1414}
1415
Balbir Singhe2224322009-04-02 16:57:39 -07001416/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001417 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001418 * @memcg: The memory cgroup that went over limit
1419 * @p: Task that is going to be killed
1420 *
1421 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1422 * enabled
1423 */
1424void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1425{
1426 struct cgroup *task_cgrp;
1427 struct cgroup *mem_cgrp;
1428 /*
1429 * Need a buffer in BSS, can't rely on allocations. The code relies
1430 * on the assumption that OOM is serialized for memory controller.
1431 * If this assumption is broken, revisit this code.
1432 */
1433 static char memcg_name[PATH_MAX];
1434 int ret;
1435
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001436 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001437 return;
1438
Balbir Singhe2224322009-04-02 16:57:39 -07001439 rcu_read_lock();
1440
1441 mem_cgrp = memcg->css.cgroup;
1442 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1443
1444 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1445 if (ret < 0) {
1446 /*
1447 * Unfortunately, we are unable to convert to a useful name
1448 * But we'll still print out the usage information
1449 */
1450 rcu_read_unlock();
1451 goto done;
1452 }
1453 rcu_read_unlock();
1454
1455 printk(KERN_INFO "Task in %s killed", memcg_name);
1456
1457 rcu_read_lock();
1458 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1459 if (ret < 0) {
1460 rcu_read_unlock();
1461 goto done;
1462 }
1463 rcu_read_unlock();
1464
1465 /*
1466 * Continues from above, so we don't need an KERN_ level
1467 */
1468 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1469done:
1470
1471 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1472 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1473 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1474 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1475 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1476 "failcnt %llu\n",
1477 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1478 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1479 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1480}
1481
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001482/*
1483 * This function returns the number of memcg under hierarchy tree. Returns
1484 * 1(self count) if no children.
1485 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001486static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001487{
1488 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001489 struct mem_cgroup *iter;
1490
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001491 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001492 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001493 return num;
1494}
1495
Balbir Singh6d61ef42009-01-07 18:08:06 -08001496/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001497 * Return the memory (and swap, if configured) limit for a memcg.
1498 */
1499u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1500{
1501 u64 limit;
1502 u64 memsw;
1503
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001504 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1505 limit += total_swap_pages << PAGE_SHIFT;
1506
David Rientjesa63d83f2010-08-09 17:19:46 -07001507 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1508 /*
1509 * If memsw is finite and limits the amount of swap space available
1510 * to this memcg, return that limit.
1511 */
1512 return min(limit, memsw);
1513}
1514
Johannes Weiner56600482012-01-12 17:17:59 -08001515static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1516 gfp_t gfp_mask,
1517 unsigned long flags)
1518{
1519 unsigned long total = 0;
1520 bool noswap = false;
1521 int loop;
1522
1523 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1524 noswap = true;
1525 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1526 noswap = true;
1527
1528 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1529 if (loop)
1530 drain_all_stock_async(memcg);
1531 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1532 /*
1533 * Allow limit shrinkers, which are triggered directly
1534 * by userspace, to catch signals and stop reclaim
1535 * after minimal progress, regardless of the margin.
1536 */
1537 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1538 break;
1539 if (mem_cgroup_margin(memcg))
1540 break;
1541 /*
1542 * If nothing was reclaimed after two attempts, there
1543 * may be no reclaimable pages in this hierarchy.
1544 */
1545 if (loop && !total)
1546 break;
1547 }
1548 return total;
1549}
1550
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001551/**
1552 * test_mem_cgroup_node_reclaimable
1553 * @mem: the target memcg
1554 * @nid: the node ID to be checked.
1555 * @noswap : specify true here if the user wants flle only information.
1556 *
1557 * This function returns whether the specified memcg contains any
1558 * reclaimable pages on a node. Returns true if there are any reclaimable
1559 * pages in the node.
1560 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001561static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001562 int nid, bool noswap)
1563{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001564 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001565 return true;
1566 if (noswap || !total_swap_pages)
1567 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001568 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001569 return true;
1570 return false;
1571
1572}
Ying Han889976d2011-05-26 16:25:33 -07001573#if MAX_NUMNODES > 1
1574
1575/*
1576 * Always updating the nodemask is not very good - even if we have an empty
1577 * list or the wrong list here, we can start from some node and traverse all
1578 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1579 *
1580 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001581static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001582{
1583 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001584 /*
1585 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1586 * pagein/pageout changes since the last update.
1587 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001588 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001589 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001590 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001591 return;
1592
Ying Han889976d2011-05-26 16:25:33 -07001593 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001594 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001595
1596 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001598 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1599 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001600 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001601
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001602 atomic_set(&memcg->numainfo_events, 0);
1603 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001604}
1605
1606/*
1607 * Selecting a node where we start reclaim from. Because what we need is just
1608 * reducing usage counter, start from anywhere is O,K. Considering
1609 * memory reclaim from current node, there are pros. and cons.
1610 *
1611 * Freeing memory from current node means freeing memory from a node which
1612 * we'll use or we've used. So, it may make LRU bad. And if several threads
1613 * hit limits, it will see a contention on a node. But freeing from remote
1614 * node means more costs for memory reclaim because of memory latency.
1615 *
1616 * Now, we use round-robin. Better algorithm is welcomed.
1617 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001618int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001619{
1620 int node;
1621
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001622 mem_cgroup_may_update_nodemask(memcg);
1623 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001624
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001625 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001626 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001628 /*
1629 * We call this when we hit limit, not when pages are added to LRU.
1630 * No LRU may hold pages because all pages are UNEVICTABLE or
1631 * memcg is too small and all pages are not on LRU. In that case,
1632 * we use curret node.
1633 */
1634 if (unlikely(node == MAX_NUMNODES))
1635 node = numa_node_id();
1636
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001637 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001638 return node;
1639}
1640
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001641/*
1642 * Check all nodes whether it contains reclaimable pages or not.
1643 * For quick scan, we make use of scan_nodes. This will allow us to skip
1644 * unused nodes. But scan_nodes is lazily updated and may not cotain
1645 * enough new information. We need to do double check.
1646 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001647bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001648{
1649 int nid;
1650
1651 /*
1652 * quick check...making use of scan_node.
1653 * We can skip unused nodes.
1654 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001655 if (!nodes_empty(memcg->scan_nodes)) {
1656 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001657 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001658 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001659
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001660 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001661 return true;
1662 }
1663 }
1664 /*
1665 * Check rest of nodes.
1666 */
1667 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001668 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001669 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001670 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001671 return true;
1672 }
1673 return false;
1674}
1675
Ying Han889976d2011-05-26 16:25:33 -07001676#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001677int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001678{
1679 return 0;
1680}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001681
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001682bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001683{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001684 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001685}
Ying Han889976d2011-05-26 16:25:33 -07001686#endif
1687
Johannes Weiner56600482012-01-12 17:17:59 -08001688static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1689 struct zone *zone,
1690 gfp_t gfp_mask,
1691 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001692{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001693 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001694 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001695 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001696 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001697 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001698 struct mem_cgroup_reclaim_cookie reclaim = {
1699 .zone = zone,
1700 .priority = 0,
1701 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001702
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001703 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001704
Balbir Singh4e416952009-09-23 15:56:39 -07001705 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001706 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001707 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001708 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001709 if (loop >= 2) {
1710 /*
1711 * If we have not been able to reclaim
1712 * anything, it might because there are
1713 * no reclaimable pages under this hierarchy
1714 */
Johannes Weiner56600482012-01-12 17:17:59 -08001715 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001716 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001717 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001718 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001719 * excess >> 2 is not to excessive so as to
1720 * reclaim too much, nor too less that we keep
1721 * coming back to reclaim from this cgroup
1722 */
1723 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001724 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001725 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001726 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001727 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001728 }
Johannes Weiner56600482012-01-12 17:17:59 -08001729 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001730 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001731 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1732 zone, &nr_scanned);
1733 *total_scanned += nr_scanned;
1734 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001735 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001736 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001737 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001738 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001739}
1740
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001741/*
1742 * Check OOM-Killer is already running under our hierarchy.
1743 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001744 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001745 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001746static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001747{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001748 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001749
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001750 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001751 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001752 /*
1753 * this subtree of our hierarchy is already locked
1754 * so we cannot give a lock.
1755 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001756 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001757 mem_cgroup_iter_break(memcg, iter);
1758 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001759 } else
1760 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001761 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001762
Michal Hocko79dfdac2011-07-26 16:08:23 -07001763 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001764 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001765
1766 /*
1767 * OK, we failed to lock the whole subtree so we have to clean up
1768 * what we set up to the failing subtree
1769 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001770 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001771 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001772 mem_cgroup_iter_break(memcg, iter);
1773 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001774 }
1775 iter->oom_lock = false;
1776 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001777 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001778}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001779
Michal Hocko79dfdac2011-07-26 16:08:23 -07001780/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001781 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001782 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001783static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001784{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001785 struct mem_cgroup *iter;
1786
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001787 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001788 iter->oom_lock = false;
1789 return 0;
1790}
1791
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001792static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001793{
1794 struct mem_cgroup *iter;
1795
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001796 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001797 atomic_inc(&iter->under_oom);
1798}
1799
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001800static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001801{
1802 struct mem_cgroup *iter;
1803
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001804 /*
1805 * When a new child is created while the hierarchy is under oom,
1806 * mem_cgroup_oom_lock() may not be called. We have to use
1807 * atomic_add_unless() here.
1808 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001809 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001810 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001811}
1812
Michal Hocko1af8efe2011-07-26 16:08:24 -07001813static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001814static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1815
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001816struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001817 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001818 wait_queue_t wait;
1819};
1820
1821static int memcg_oom_wake_function(wait_queue_t *wait,
1822 unsigned mode, int sync, void *arg)
1823{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001824 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1825 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001826 struct oom_wait_info *oom_wait_info;
1827
1828 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001829 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001830
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001831 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001832 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001833 * Then we can use css_is_ancestor without taking care of RCU.
1834 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001835 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1836 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001837 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001838 return autoremove_wake_function(wait, mode, sync, arg);
1839}
1840
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001841static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001842{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001843 /* for filtering, pass "memcg" as argument. */
1844 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001845}
1846
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001847static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001848{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001849 if (memcg && atomic_read(&memcg->under_oom))
1850 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001851}
1852
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001853/*
1854 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1855 */
David Rientjese845e192012-03-21 16:34:10 -07001856bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001857{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001858 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001859 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001860
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001861 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001862 owait.wait.flags = 0;
1863 owait.wait.func = memcg_oom_wake_function;
1864 owait.wait.private = current;
1865 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001866 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001867 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001868
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001869 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001870 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001871 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001872 /*
1873 * Even if signal_pending(), we can't quit charge() loop without
1874 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1875 * under OOM is always welcomed, use TASK_KILLABLE here.
1876 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001877 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001878 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001879 need_to_kill = false;
1880 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001881 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001882 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001883
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001884 if (need_to_kill) {
1885 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07001886 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001887 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001888 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001889 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001890 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001891 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001892 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001893 mem_cgroup_oom_unlock(memcg);
1894 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001895 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001896
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001897 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001898
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001899 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1900 return false;
1901 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001902 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001903 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001904}
1905
Balbir Singhd69b0422009-06-17 16:26:34 -07001906/*
1907 * Currently used to update mapped file statistics, but the routine can be
1908 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001909 *
1910 * Notes: Race condition
1911 *
1912 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1913 * it tends to be costly. But considering some conditions, we doesn't need
1914 * to do so _always_.
1915 *
1916 * Considering "charge", lock_page_cgroup() is not required because all
1917 * file-stat operations happen after a page is attached to radix-tree. There
1918 * are no race with "charge".
1919 *
1920 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1921 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1922 * if there are race with "uncharge". Statistics itself is properly handled
1923 * by flags.
1924 *
1925 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001926 * small, we check mm->moving_account and detect there are possibility of race
1927 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001928 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001929
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001930void __mem_cgroup_begin_update_page_stat(struct page *page,
1931 bool *locked, unsigned long *flags)
1932{
1933 struct mem_cgroup *memcg;
1934 struct page_cgroup *pc;
1935
1936 pc = lookup_page_cgroup(page);
1937again:
1938 memcg = pc->mem_cgroup;
1939 if (unlikely(!memcg || !PageCgroupUsed(pc)))
1940 return;
1941 /*
1942 * If this memory cgroup is not under account moving, we don't
1943 * need to take move_lock_page_cgroup(). Because we already hold
1944 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001945 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001946 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001947 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001948 return;
1949
1950 move_lock_mem_cgroup(memcg, flags);
1951 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
1952 move_unlock_mem_cgroup(memcg, flags);
1953 goto again;
1954 }
1955 *locked = true;
1956}
1957
1958void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
1959{
1960 struct page_cgroup *pc = lookup_page_cgroup(page);
1961
1962 /*
1963 * It's guaranteed that pc->mem_cgroup never changes while
1964 * lock is held because a routine modifies pc->mem_cgroup
1965 * should take move_lock_page_cgroup().
1966 */
1967 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1968}
1969
Greg Thelen2a7106f2011-01-13 15:47:37 -08001970void mem_cgroup_update_page_stat(struct page *page,
1971 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001972{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001973 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001974 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001975 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001976
Johannes Weinercfa44942012-01-12 17:18:38 -08001977 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001978 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001979
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001980 memcg = pc->mem_cgroup;
1981 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001982 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001983
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001984 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001985 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08001986 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001987 break;
1988 default:
1989 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001990 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001991
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001992 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07001993}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001994
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001995/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001996 * size of first charge trial. "32" comes from vmscan.c's magic value.
1997 * TODO: maybe necessary to use big numbers in big irons.
1998 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001999#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002000struct memcg_stock_pcp {
2001 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002002 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002003 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002004 unsigned long flags;
2005#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002006};
2007static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002008static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002009
2010/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002011 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002012 * from local stock and true is returned. If the stock is 0 or charges from a
2013 * cgroup which is not current target, returns false. This stock will be
2014 * refilled.
2015 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002016static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002017{
2018 struct memcg_stock_pcp *stock;
2019 bool ret = true;
2020
2021 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002022 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002023 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002024 else /* need to call res_counter_charge */
2025 ret = false;
2026 put_cpu_var(memcg_stock);
2027 return ret;
2028}
2029
2030/*
2031 * Returns stocks cached in percpu to res_counter and reset cached information.
2032 */
2033static void drain_stock(struct memcg_stock_pcp *stock)
2034{
2035 struct mem_cgroup *old = stock->cached;
2036
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002037 if (stock->nr_pages) {
2038 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2039
2040 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002041 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002042 res_counter_uncharge(&old->memsw, bytes);
2043 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002044 }
2045 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002046}
2047
2048/*
2049 * This must be called under preempt disabled or must be called by
2050 * a thread which is pinned to local cpu.
2051 */
2052static void drain_local_stock(struct work_struct *dummy)
2053{
2054 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2055 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002056 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002057}
2058
2059/*
2060 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002061 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002062 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002063static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002064{
2065 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2066
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002067 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002068 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002069 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002070 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002071 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002072 put_cpu_var(memcg_stock);
2073}
2074
2075/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002076 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002077 * of the hierarchy under it. sync flag says whether we should block
2078 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002079 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002080static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002081{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002082 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002083
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002084 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002085 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002086 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002087 for_each_online_cpu(cpu) {
2088 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002089 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002090
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002091 memcg = stock->cached;
2092 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002093 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002094 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002095 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002096 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2097 if (cpu == curcpu)
2098 drain_local_stock(&stock->work);
2099 else
2100 schedule_work_on(cpu, &stock->work);
2101 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002102 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002103 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002104
2105 if (!sync)
2106 goto out;
2107
2108 for_each_online_cpu(cpu) {
2109 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002110 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002111 flush_work(&stock->work);
2112 }
2113out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002114 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002115}
2116
2117/*
2118 * Tries to drain stocked charges in other cpus. This function is asynchronous
2119 * and just put a work per cpu for draining localy on each cpu. Caller can
2120 * expects some charges will be back to res_counter later but cannot wait for
2121 * it.
2122 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002123static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002124{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002125 /*
2126 * If someone calls draining, avoid adding more kworker runs.
2127 */
2128 if (!mutex_trylock(&percpu_charge_mutex))
2129 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002130 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002131 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002132}
2133
2134/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002136{
2137 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002138 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002139 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002140 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002141}
2142
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002143/*
2144 * This function drains percpu counter value from DEAD cpu and
2145 * move it to local cpu. Note that this function can be preempted.
2146 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002147static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002148{
2149 int i;
2150
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002151 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002152 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002153 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002154
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002155 per_cpu(memcg->stat->count[i], cpu) = 0;
2156 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002157 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002158 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002159 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002160
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002161 per_cpu(memcg->stat->events[i], cpu) = 0;
2162 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002163 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002164 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002165}
2166
2167static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002168 unsigned long action,
2169 void *hcpu)
2170{
2171 int cpu = (unsigned long)hcpu;
2172 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002173 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002174
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002175 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002176 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002177
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002178 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002179 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002180
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002181 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002182 mem_cgroup_drain_pcp_counter(iter, cpu);
2183
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002184 stock = &per_cpu(memcg_stock, cpu);
2185 drain_stock(stock);
2186 return NOTIFY_OK;
2187}
2188
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002189
2190/* See __mem_cgroup_try_charge() for details */
2191enum {
2192 CHARGE_OK, /* success */
2193 CHARGE_RETRY, /* need to retry but retry is not bad */
2194 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2195 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2196 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2197};
2198
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002199static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002200 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002201{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002202 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002203 struct mem_cgroup *mem_over_limit;
2204 struct res_counter *fail_res;
2205 unsigned long flags = 0;
2206 int ret;
2207
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002208 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002209
2210 if (likely(!ret)) {
2211 if (!do_swap_account)
2212 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002213 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002214 if (likely(!ret))
2215 return CHARGE_OK;
2216
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002217 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002218 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2219 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2220 } else
2221 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002222 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002223 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2224 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002225 *
2226 * Never reclaim on behalf of optional batching, retry with a
2227 * single page instead.
2228 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002229 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002230 return CHARGE_RETRY;
2231
2232 if (!(gfp_mask & __GFP_WAIT))
2233 return CHARGE_WOULDBLOCK;
2234
Johannes Weiner56600482012-01-12 17:17:59 -08002235 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002236 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002237 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002238 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002239 * Even though the limit is exceeded at this point, reclaim
2240 * may have been able to free some pages. Retry the charge
2241 * before killing the task.
2242 *
2243 * Only for regular pages, though: huge pages are rather
2244 * unlikely to succeed so close to the limit, and we fall back
2245 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002246 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002247 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002248 return CHARGE_RETRY;
2249
2250 /*
2251 * At task move, charge accounts can be doubly counted. So, it's
2252 * better to wait until the end of task_move if something is going on.
2253 */
2254 if (mem_cgroup_wait_acct_move(mem_over_limit))
2255 return CHARGE_RETRY;
2256
2257 /* If we don't need to call oom-killer at el, return immediately */
2258 if (!oom_check)
2259 return CHARGE_NOMEM;
2260 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002261 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002262 return CHARGE_OOM_DIE;
2263
2264 return CHARGE_RETRY;
2265}
2266
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002267/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002268 * __mem_cgroup_try_charge() does
2269 * 1. detect memcg to be charged against from passed *mm and *ptr,
2270 * 2. update res_counter
2271 * 3. call memory reclaim if necessary.
2272 *
2273 * In some special case, if the task is fatal, fatal_signal_pending() or
2274 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2275 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2276 * as possible without any hazards. 2: all pages should have a valid
2277 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2278 * pointer, that is treated as a charge to root_mem_cgroup.
2279 *
2280 * So __mem_cgroup_try_charge() will return
2281 * 0 ... on success, filling *ptr with a valid memcg pointer.
2282 * -ENOMEM ... charge failure because of resource limits.
2283 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2284 *
2285 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2286 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002287 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002288static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002289 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002290 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002291 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002292 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002293{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002294 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002295 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002296 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002297 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002298
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002299 /*
2300 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2301 * in system level. So, allow to go ahead dying process in addition to
2302 * MEMDIE process.
2303 */
2304 if (unlikely(test_thread_flag(TIF_MEMDIE)
2305 || fatal_signal_pending(current)))
2306 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002307
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002308 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002309 * We always charge the cgroup the mm_struct belongs to.
2310 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002311 * thread group leader migrates. It's possible that mm is not
2312 * set, if so charge the init_mm (happens for pagecache usage).
2313 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002314 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002315 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002316again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002317 if (*ptr) { /* css should be a valid one */
2318 memcg = *ptr;
2319 VM_BUG_ON(css_is_removed(&memcg->css));
2320 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002321 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002322 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002323 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002324 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002325 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002326 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002327
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002328 rcu_read_lock();
2329 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002330 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002331 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002332 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002333 * race with swapoff. Then, we have small risk of mis-accouning.
2334 * But such kind of mis-account by race always happens because
2335 * we don't have cgroup_mutex(). It's overkill and we allo that
2336 * small race, here.
2337 * (*) swapoff at el will charge against mm-struct not against
2338 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002339 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002340 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002341 if (!memcg)
2342 memcg = root_mem_cgroup;
2343 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002344 rcu_read_unlock();
2345 goto done;
2346 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002347 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002348 /*
2349 * It seems dagerous to access memcg without css_get().
2350 * But considering how consume_stok works, it's not
2351 * necessary. If consume_stock success, some charges
2352 * from this memcg are cached on this cpu. So, we
2353 * don't need to call css_get()/css_tryget() before
2354 * calling consume_stock().
2355 */
2356 rcu_read_unlock();
2357 goto done;
2358 }
2359 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002360 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002361 rcu_read_unlock();
2362 goto again;
2363 }
2364 rcu_read_unlock();
2365 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002366
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002367 do {
2368 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002369
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002370 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002371 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002372 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002373 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002374 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002375
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002376 oom_check = false;
2377 if (oom && !nr_oom_retries) {
2378 oom_check = true;
2379 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2380 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002381
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002382 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002383 switch (ret) {
2384 case CHARGE_OK:
2385 break;
2386 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002387 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002388 css_put(&memcg->css);
2389 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002390 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002391 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002392 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002393 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002394 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002395 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002396 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002397 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002398 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002399 /* If oom, we never return -ENOMEM */
2400 nr_oom_retries--;
2401 break;
2402 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002403 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002404 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002405 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002406 } while (ret != CHARGE_OK);
2407
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002408 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002409 refill_stock(memcg, batch - nr_pages);
2410 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002411done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002412 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002413 return 0;
2414nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002415 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002416 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002417bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002418 *ptr = root_mem_cgroup;
2419 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002420}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002421
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002422/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002423 * Somemtimes we have to undo a charge we got by try_charge().
2424 * This function is for that and do uncharge, put css's refcnt.
2425 * gotten by try_charge().
2426 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002427static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002428 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002429{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002430 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002431 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002432
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002433 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002434 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002435 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002436 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002437}
2438
2439/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002440 * A helper function to get mem_cgroup from ID. must be called under
2441 * rcu_read_lock(). The caller must check css_is_removed() or some if
2442 * it's concern. (dropping refcnt from swap can be called against removed
2443 * memcg.)
2444 */
2445static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2446{
2447 struct cgroup_subsys_state *css;
2448
2449 /* ID 0 is unused ID */
2450 if (!id)
2451 return NULL;
2452 css = css_lookup(&mem_cgroup_subsys, id);
2453 if (!css)
2454 return NULL;
2455 return container_of(css, struct mem_cgroup, css);
2456}
2457
Wu Fengguange42d9d52009-12-16 12:19:59 +01002458struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002459{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002460 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002461 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002462 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002463 swp_entry_t ent;
2464
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002465 VM_BUG_ON(!PageLocked(page));
2466
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002467 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002468 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002469 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002470 memcg = pc->mem_cgroup;
2471 if (memcg && !css_tryget(&memcg->css))
2472 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002473 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002474 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002475 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002476 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002477 memcg = mem_cgroup_lookup(id);
2478 if (memcg && !css_tryget(&memcg->css))
2479 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002480 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002481 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002482 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002483 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002484}
2485
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002486static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002487 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002488 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002489 enum charge_type ctype,
2490 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002491{
Johannes Weinerce587e62012-04-24 20:22:33 +02002492 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002493 struct zone *uninitialized_var(zone);
2494 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002495 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002496
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002497 lock_page_cgroup(pc);
2498 if (unlikely(PageCgroupUsed(pc))) {
2499 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002500 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002501 return;
2502 }
2503 /*
2504 * we don't need page_cgroup_lock about tail pages, becase they are not
2505 * accessed by any other context at this point.
2506 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002507
2508 /*
2509 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2510 * may already be on some other mem_cgroup's LRU. Take care of it.
2511 */
2512 if (lrucare) {
2513 zone = page_zone(page);
2514 spin_lock_irq(&zone->lru_lock);
2515 if (PageLRU(page)) {
2516 ClearPageLRU(page);
2517 del_page_from_lru_list(zone, page, page_lru(page));
2518 was_on_lru = true;
2519 }
2520 }
2521
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002522 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002523 /*
2524 * We access a page_cgroup asynchronously without lock_page_cgroup().
2525 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2526 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2527 * before USED bit, we need memory barrier here.
2528 * See mem_cgroup_add_lru_list(), etc.
2529 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002530 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002531 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002532
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002533 if (lrucare) {
2534 if (was_on_lru) {
2535 VM_BUG_ON(PageLRU(page));
2536 SetPageLRU(page);
2537 add_page_to_lru_list(zone, page, page_lru(page));
2538 }
2539 spin_unlock_irq(&zone->lru_lock);
2540 }
2541
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002542 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2543 anon = true;
2544 else
2545 anon = false;
2546
2547 mem_cgroup_charge_statistics(memcg, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002548 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002549
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002550 /*
2551 * "charge_statistics" updated event counter. Then, check it.
2552 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2553 * if they exceeds softlimit.
2554 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002555 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002556}
2557
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002558#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2559
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002560#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION))
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002561/*
2562 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002563 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2564 * charge/uncharge will be never happen and move_account() is done under
2565 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002566 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002567void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002568{
2569 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002570 struct page_cgroup *pc;
2571 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002572
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002573 if (mem_cgroup_disabled())
2574 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002575 for (i = 1; i < HPAGE_PMD_NR; i++) {
2576 pc = head_pc + i;
2577 pc->mem_cgroup = head_pc->mem_cgroup;
2578 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002579 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2580 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002581}
Hugh Dickins12d27102012-01-12 17:19:52 -08002582#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002583
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002584/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002585 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002586 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002587 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002588 * @pc: page_cgroup of the page.
2589 * @from: mem_cgroup which the page is moved from.
2590 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002591 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002592 *
2593 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002594 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002595 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002596 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002597 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002598 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002599 * true, this function does "uncharge" from old cgroup, but it doesn't if
2600 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002601 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002602static int mem_cgroup_move_account(struct page *page,
2603 unsigned int nr_pages,
2604 struct page_cgroup *pc,
2605 struct mem_cgroup *from,
2606 struct mem_cgroup *to,
2607 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002608{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002609 unsigned long flags;
2610 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002611 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002612
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002613 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002614 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002615 /*
2616 * The page is isolated from LRU. So, collapse function
2617 * will not handle this page. But page splitting can happen.
2618 * Do this check under compound_page_lock(). The caller should
2619 * hold it.
2620 */
2621 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002622 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002623 goto out;
2624
2625 lock_page_cgroup(pc);
2626
2627 ret = -EINVAL;
2628 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2629 goto unlock;
2630
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002631 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002632
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002633 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002634 /* Update mapped_file data for mem_cgroup */
2635 preempt_disable();
2636 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2637 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2638 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002639 }
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002640 mem_cgroup_charge_statistics(from, anon, -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002641 if (uncharge)
2642 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002643 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002644
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002645 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002646 pc->mem_cgroup = to;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002647 mem_cgroup_charge_statistics(to, anon, nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002648 /*
2649 * We charges against "to" which may not have any tasks. Then, "to"
2650 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002651 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002652 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002653 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002654 */
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002655 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002656 ret = 0;
2657unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002658 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002659 /*
2660 * check events
2661 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002662 memcg_check_events(to, page);
2663 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002664out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002665 return ret;
2666}
2667
2668/*
2669 * move charges to its parent.
2670 */
2671
Johannes Weiner5564e882011-03-23 16:42:29 -07002672static int mem_cgroup_move_parent(struct page *page,
2673 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002674 struct mem_cgroup *child,
2675 gfp_t gfp_mask)
2676{
2677 struct cgroup *cg = child->css.cgroup;
2678 struct cgroup *pcg = cg->parent;
2679 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002680 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002681 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002682 int ret;
2683
2684 /* Is ROOT ? */
2685 if (!pcg)
2686 return -EINVAL;
2687
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002688 ret = -EBUSY;
2689 if (!get_page_unless_zero(page))
2690 goto out;
2691 if (isolate_lru_page(page))
2692 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002693
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002694 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002695
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002696 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002697 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002698 if (ret)
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002699 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002700
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002701 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002702 flags = compound_lock_irqsave(page);
2703
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002704 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002705 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002706 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002707
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002708 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002709 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002710put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002711 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002712put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002713 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002714out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002715 return ret;
2716}
2717
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002718/*
2719 * Charge the memory controller for page usage.
2720 * Return
2721 * 0 if the charge was successful
2722 * < 0 if the cgroup is over its limit
2723 */
2724static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002725 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002726{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002727 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002728 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002729 bool oom = true;
2730 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002731
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002732 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002733 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002734 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002735 /*
2736 * Never OOM-kill a process for a huge page. The
2737 * fault handler will fall back to regular pages.
2738 */
2739 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002740 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002741
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002742 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002743 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002744 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02002745 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002746 return 0;
2747}
2748
2749int mem_cgroup_newpage_charge(struct page *page,
2750 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002751{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002752 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002753 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002754 VM_BUG_ON(page_mapped(page));
2755 VM_BUG_ON(page->mapping && !PageAnon(page));
2756 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002757 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002758 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002759}
2760
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002761static void
2762__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2763 enum charge_type ctype);
2764
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002765int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2766 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002767{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002768 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002769 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002770 int ret;
2771
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002772 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002773 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002774 if (PageCompound(page))
2775 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002776
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002777 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002778 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002779 if (!page_is_file_cache(page))
2780 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002781
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002782 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002783 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002784 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002785 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002786 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002787 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2788 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002789 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002790}
2791
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002792/*
2793 * While swap-in, try_charge -> commit or cancel, the page is locked.
2794 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002795 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002796 * "commit()" or removed by "cancel()"
2797 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002798int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2799 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002800 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002801{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002802 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002803 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002804
Johannes Weiner72835c82012-01-12 17:18:32 -08002805 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002806
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002807 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002808 return 0;
2809
2810 if (!do_swap_account)
2811 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002812 /*
2813 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002814 * the pte, and even removed page from swap cache: in those cases
2815 * do_swap_page()'s pte_same() test will fail; but there's also a
2816 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002817 */
2818 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002819 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002820 memcg = try_get_mem_cgroup_from_page(page);
2821 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002822 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002823 *memcgp = memcg;
2824 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002825 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002826 if (ret == -EINTR)
2827 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002828 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002829charge_cur_mm:
2830 if (unlikely(!mm))
2831 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002832 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2833 if (ret == -EINTR)
2834 ret = 0;
2835 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002836}
2837
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002838static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002839__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002840 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002841{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002842 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002843 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002844 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002845 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002846 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002847
Johannes Weinerce587e62012-04-24 20:22:33 +02002848 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002849 /*
2850 * Now swap is on-memory. This means this page may be
2851 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002852 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2853 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2854 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002855 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002856 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002857 swp_entry_t ent = {.val = page_private(page)};
Johannes Weiner72835c82012-01-12 17:18:32 -08002858 struct mem_cgroup *swap_memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002859 unsigned short id;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002860
2861 id = swap_cgroup_record(ent, 0);
2862 rcu_read_lock();
Johannes Weiner72835c82012-01-12 17:18:32 -08002863 swap_memcg = mem_cgroup_lookup(id);
2864 if (swap_memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002865 /*
2866 * This recorded memcg can be obsolete one. So, avoid
2867 * calling css_tryget
2868 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002869 if (!mem_cgroup_is_root(swap_memcg))
2870 res_counter_uncharge(&swap_memcg->memsw,
2871 PAGE_SIZE);
2872 mem_cgroup_swap_statistics(swap_memcg, false);
2873 mem_cgroup_put(swap_memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002874 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002875 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002876 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002877 /*
2878 * At swapin, we may charge account against cgroup which has no tasks.
2879 * So, rmdir()->pre_destroy() can be called while we do this charge.
2880 * In that case, we need to call pre_destroy() again. check it here.
2881 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002882 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002883}
2884
Johannes Weiner72835c82012-01-12 17:18:32 -08002885void mem_cgroup_commit_charge_swapin(struct page *page,
2886 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002887{
Johannes Weiner72835c82012-01-12 17:18:32 -08002888 __mem_cgroup_commit_charge_swapin(page, memcg,
2889 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002890}
2891
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002892void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002893{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002894 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002895 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002896 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002897 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002898 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002899}
2900
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002901static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002902 unsigned int nr_pages,
2903 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002904{
2905 struct memcg_batch_info *batch = NULL;
2906 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002907
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002908 /* If swapout, usage of swap doesn't decrease */
2909 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2910 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002911
2912 batch = &current->memcg_batch;
2913 /*
2914 * In usual, we do css_get() when we remember memcg pointer.
2915 * But in this case, we keep res->usage until end of a series of
2916 * uncharges. Then, it's ok to ignore memcg's refcnt.
2917 */
2918 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002919 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002920 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002921 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002922 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002923 * the same cgroup and we have chance to coalesce uncharges.
2924 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2925 * because we want to do uncharge as soon as possible.
2926 */
2927
2928 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2929 goto direct_uncharge;
2930
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002931 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002932 goto direct_uncharge;
2933
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002934 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002935 * In typical case, batch->memcg == mem. This means we can
2936 * merge a series of uncharges to an uncharge of res_counter.
2937 * If not, we uncharge res_counter ony by one.
2938 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002939 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002940 goto direct_uncharge;
2941 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002942 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002943 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002944 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002945 return;
2946direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002947 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002948 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002949 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2950 if (unlikely(batch->memcg != memcg))
2951 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002952}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002953
Balbir Singh8697d332008-02-07 00:13:59 -08002954/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002955 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002956 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002957static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002958__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002959{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002960 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002961 unsigned int nr_pages = 1;
2962 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002963 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002964
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002965 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002966 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002967
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002968 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002969 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002970
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002971 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002972 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002973 VM_BUG_ON(!PageTransHuge(page));
2974 }
Balbir Singh8697d332008-02-07 00:13:59 -08002975 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002976 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002977 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002978 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002979 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002980 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002981
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002982 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002983
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002984 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002985
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002986 if (!PageCgroupUsed(pc))
2987 goto unlock_out;
2988
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002989 anon = PageAnon(page);
2990
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002991 switch (ctype) {
2992 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002993 /*
2994 * Generally PageAnon tells if it's the anon statistics to be
2995 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2996 * used before page reached the stage of being marked PageAnon.
2997 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002998 anon = true;
2999 /* fallthrough */
KAMEZAWA Hiroyuki8a9478ca2009-06-17 16:27:17 -07003000 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003001 /* See mem_cgroup_prepare_migration() */
3002 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003003 goto unlock_out;
3004 break;
3005 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3006 if (!PageAnon(page)) { /* Shared memory */
3007 if (page->mapping && !page_is_file_cache(page))
3008 goto unlock_out;
3009 } else if (page_mapped(page)) /* Anon */
3010 goto unlock_out;
3011 break;
3012 default:
3013 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003014 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003015
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003016 mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003017
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003018 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003019 /*
3020 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3021 * freed from LRU. This is safe because uncharged page is expected not
3022 * to be reused (freed soon). Exception is SwapCache, it's handled by
3023 * special functions.
3024 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003025
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003026 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003027 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003028 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003029 * will never be freed.
3030 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003031 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003032 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003033 mem_cgroup_swap_statistics(memcg, true);
3034 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003035 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003036 if (!mem_cgroup_is_root(memcg))
3037 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003038
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003039 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003040
3041unlock_out:
3042 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003043 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003044}
3045
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003046void mem_cgroup_uncharge_page(struct page *page)
3047{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003048 /* early check. */
3049 if (page_mapped(page))
3050 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08003051 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003052 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3053}
3054
3055void mem_cgroup_uncharge_cache_page(struct page *page)
3056{
3057 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003058 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003059 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3060}
3061
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003062/*
3063 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3064 * In that cases, pages are freed continuously and we can expect pages
3065 * are in the same memcg. All these calls itself limits the number of
3066 * pages freed at once, then uncharge_start/end() is called properly.
3067 * This may be called prural(2) times in a context,
3068 */
3069
3070void mem_cgroup_uncharge_start(void)
3071{
3072 current->memcg_batch.do_batch++;
3073 /* We can do nest. */
3074 if (current->memcg_batch.do_batch == 1) {
3075 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003076 current->memcg_batch.nr_pages = 0;
3077 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003078 }
3079}
3080
3081void mem_cgroup_uncharge_end(void)
3082{
3083 struct memcg_batch_info *batch = &current->memcg_batch;
3084
3085 if (!batch->do_batch)
3086 return;
3087
3088 batch->do_batch--;
3089 if (batch->do_batch) /* If stacked, do nothing. */
3090 return;
3091
3092 if (!batch->memcg)
3093 return;
3094 /*
3095 * This "batch->memcg" is valid without any css_get/put etc...
3096 * bacause we hide charges behind us.
3097 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003098 if (batch->nr_pages)
3099 res_counter_uncharge(&batch->memcg->res,
3100 batch->nr_pages * PAGE_SIZE);
3101 if (batch->memsw_nr_pages)
3102 res_counter_uncharge(&batch->memcg->memsw,
3103 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003104 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003105 /* forget this pointer (for sanity check) */
3106 batch->memcg = NULL;
3107}
3108
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003109#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003110/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003111 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003112 * memcg information is recorded to swap_cgroup of "ent"
3113 */
KAMEZAWA Hiroyuki8a9478ca2009-06-17 16:27:17 -07003114void
3115mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003116{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003117 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478ca2009-06-17 16:27:17 -07003118 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003119
KAMEZAWA Hiroyuki8a9478ca2009-06-17 16:27:17 -07003120 if (!swapout) /* this was a swap cache but the swap is unused ! */
3121 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3122
3123 memcg = __mem_cgroup_uncharge_common(page, ctype);
3124
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003125 /*
3126 * record memcg information, if swapout && memcg != NULL,
3127 * mem_cgroup_get() was called in uncharge().
3128 */
3129 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003130 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003131}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003132#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003133
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003134#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3135/*
3136 * called from swap_entry_free(). remove record in swap_cgroup and
3137 * uncharge "memsw" account.
3138 */
3139void mem_cgroup_uncharge_swap(swp_entry_t ent)
3140{
3141 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003142 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003143
3144 if (!do_swap_account)
3145 return;
3146
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003147 id = swap_cgroup_record(ent, 0);
3148 rcu_read_lock();
3149 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003150 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003151 /*
3152 * We uncharge this because swap is freed.
3153 * This memcg can be obsolete one. We avoid calling css_tryget
3154 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003155 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003156 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003157 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003158 mem_cgroup_put(memcg);
3159 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003160 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003161}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003162
3163/**
3164 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3165 * @entry: swap entry to be moved
3166 * @from: mem_cgroup which the entry is moved from
3167 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003168 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003169 *
3170 * It succeeds only when the swap_cgroup's record for this entry is the same
3171 * as the mem_cgroup's id of @from.
3172 *
3173 * Returns 0 on success, -EINVAL on failure.
3174 *
3175 * The caller must have charged to @to, IOW, called res_counter_charge() about
3176 * both res and memsw, and called css_get().
3177 */
3178static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003179 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003180{
3181 unsigned short old_id, new_id;
3182
3183 old_id = css_id(&from->css);
3184 new_id = css_id(&to->css);
3185
3186 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003187 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003188 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003189 /*
3190 * This function is only called from task migration context now.
3191 * It postpones res_counter and refcount handling till the end
3192 * of task migration(mem_cgroup_clear_mc()) for performance
3193 * improvement. But we cannot postpone mem_cgroup_get(to)
3194 * because if the process that has been moved to @to does
3195 * swap-in, the refcount of @to might be decreased to 0.
3196 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003197 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003198 if (need_fixup) {
3199 if (!mem_cgroup_is_root(from))
3200 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3201 mem_cgroup_put(from);
3202 /*
3203 * we charged both to->res and to->memsw, so we should
3204 * uncharge to->res.
3205 */
3206 if (!mem_cgroup_is_root(to))
3207 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003208 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003209 return 0;
3210 }
3211 return -EINVAL;
3212}
3213#else
3214static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003215 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003216{
3217 return -EINVAL;
3218}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003219#endif
3220
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003221/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003222 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3223 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003224 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003225int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003226 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003227{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003228 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003229 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003230 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003231 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003232
Johannes Weiner72835c82012-01-12 17:18:32 -08003233 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003234
Andrea Arcangeliec168512011-01-13 15:46:56 -08003235 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003236 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003237 return 0;
3238
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003239 pc = lookup_page_cgroup(page);
3240 lock_page_cgroup(pc);
3241 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003242 memcg = pc->mem_cgroup;
3243 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003244 /*
3245 * At migrating an anonymous page, its mapcount goes down
3246 * to 0 and uncharge() will be called. But, even if it's fully
3247 * unmapped, migration may fail and this page has to be
3248 * charged again. We set MIGRATION flag here and delay uncharge
3249 * until end_migration() is called
3250 *
3251 * Corner Case Thinking
3252 * A)
3253 * When the old page was mapped as Anon and it's unmap-and-freed
3254 * while migration was ongoing.
3255 * If unmap finds the old page, uncharge() of it will be delayed
3256 * until end_migration(). If unmap finds a new page, it's
3257 * uncharged when it make mapcount to be 1->0. If unmap code
3258 * finds swap_migration_entry, the new page will not be mapped
3259 * and end_migration() will find it(mapcount==0).
3260 *
3261 * B)
3262 * When the old page was mapped but migraion fails, the kernel
3263 * remaps it. A charge for it is kept by MIGRATION flag even
3264 * if mapcount goes down to 0. We can do remap successfully
3265 * without charging it again.
3266 *
3267 * C)
3268 * The "old" page is under lock_page() until the end of
3269 * migration, so, the old page itself will not be swapped-out.
3270 * If the new page is swapped out before end_migraton, our
3271 * hook to usual swap-out path will catch the event.
3272 */
3273 if (PageAnon(page))
3274 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003275 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003276 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003277 /*
3278 * If the page is not charged at this point,
3279 * we return here.
3280 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003281 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003282 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003283
Johannes Weiner72835c82012-01-12 17:18:32 -08003284 *memcgp = memcg;
3285 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003286 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003287 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003288 if (PageAnon(page)) {
3289 lock_page_cgroup(pc);
3290 ClearPageCgroupMigration(pc);
3291 unlock_page_cgroup(pc);
3292 /*
3293 * The old page may be fully unmapped while we kept it.
3294 */
3295 mem_cgroup_uncharge_page(page);
3296 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003297 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003298 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003299 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003300 /*
3301 * We charge new page before it's used/mapped. So, even if unlock_page()
3302 * is called before end_migration, we can catch all events on this new
3303 * page. In the case new page is migrated but not remapped, new page's
3304 * mapcount will be finally 0 and we call uncharge in end_migration().
3305 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003306 if (PageAnon(page))
3307 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3308 else if (page_is_file_cache(page))
3309 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3310 else
3311 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Johannes Weinerce587e62012-04-24 20:22:33 +02003312 __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003313 return ret;
3314}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003315
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003316/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003317void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003318 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003319{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003320 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003321 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003322 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003323
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003324 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003325 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003326 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003327 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003328 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003329 used = oldpage;
3330 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003331 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003332 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003333 unused = oldpage;
3334 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003335 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003336 * We disallowed uncharge of pages under migration because mapcount
3337 * of the page goes down to zero, temporarly.
3338 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003339 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003340 pc = lookup_page_cgroup(oldpage);
3341 lock_page_cgroup(pc);
3342 ClearPageCgroupMigration(pc);
3343 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003344 anon = PageAnon(used);
3345 __mem_cgroup_uncharge_common(unused,
3346 anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED
3347 : MEM_CGROUP_CHARGE_TYPE_CACHE);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003348
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003349 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003350 * If a page is a file cache, radix-tree replacement is very atomic
3351 * and we can skip this check. When it was an Anon page, its mapcount
3352 * goes down to 0. But because we added MIGRATION flage, it's not
3353 * uncharged yet. There are several case but page->mapcount check
3354 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3355 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003356 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003357 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003358 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003359 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003360 * At migration, we may charge account against cgroup which has no
3361 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003362 * So, rmdir()->pre_destroy() can be called while we do this charge.
3363 * In that case, we need to call pre_destroy() again. check it here.
3364 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003365 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003366}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003367
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003368/*
3369 * At replace page cache, newpage is not under any memcg but it's on
3370 * LRU. So, this function doesn't touch res_counter but handles LRU
3371 * in correct way. Both pages are locked so we cannot race with uncharge.
3372 */
3373void mem_cgroup_replace_page_cache(struct page *oldpage,
3374 struct page *newpage)
3375{
3376 struct mem_cgroup *memcg;
3377 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003378 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003379
3380 if (mem_cgroup_disabled())
3381 return;
3382
3383 pc = lookup_page_cgroup(oldpage);
3384 /* fix accounting on old pages */
3385 lock_page_cgroup(pc);
3386 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003387 mem_cgroup_charge_statistics(memcg, false, -1);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003388 ClearPageCgroupUsed(pc);
3389 unlock_page_cgroup(pc);
3390
3391 if (PageSwapBacked(oldpage))
3392 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3393
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003394 /*
3395 * Even if newpage->mapping was NULL before starting replacement,
3396 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3397 * LRU while we overwrite pc->mem_cgroup.
3398 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003399 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003400}
3401
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003402#ifdef CONFIG_DEBUG_VM
3403static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3404{
3405 struct page_cgroup *pc;
3406
3407 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003408 /*
3409 * Can be NULL while feeding pages into the page allocator for
3410 * the first time, i.e. during boot or memory hotplug;
3411 * or when mem_cgroup_disabled().
3412 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003413 if (likely(pc) && PageCgroupUsed(pc))
3414 return pc;
3415 return NULL;
3416}
3417
3418bool mem_cgroup_bad_page_check(struct page *page)
3419{
3420 if (mem_cgroup_disabled())
3421 return false;
3422
3423 return lookup_page_cgroup_used(page) != NULL;
3424}
3425
3426void mem_cgroup_print_bad_page(struct page *page)
3427{
3428 struct page_cgroup *pc;
3429
3430 pc = lookup_page_cgroup_used(page);
3431 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003432 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003433 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003434 }
3435}
3436#endif
3437
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003438static DEFINE_MUTEX(set_limit_mutex);
3439
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003440static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003441 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003442{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003443 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003444 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003445 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003446 int children = mem_cgroup_count_children(memcg);
3447 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003448 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003449
3450 /*
3451 * For keeping hierarchical_reclaim simple, how long we should retry
3452 * is depends on callers. We set our retry-count to be function
3453 * of # of children which we should visit in this loop.
3454 */
3455 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3456
3457 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003458
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003459 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003460 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003461 if (signal_pending(current)) {
3462 ret = -EINTR;
3463 break;
3464 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003465 /*
3466 * Rather than hide all in some function, I do this in
3467 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003468 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003469 */
3470 mutex_lock(&set_limit_mutex);
3471 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3472 if (memswlimit < val) {
3473 ret = -EINVAL;
3474 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003475 break;
3476 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003477
3478 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3479 if (memlimit < val)
3480 enlarge = 1;
3481
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003482 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003483 if (!ret) {
3484 if (memswlimit == val)
3485 memcg->memsw_is_minimum = true;
3486 else
3487 memcg->memsw_is_minimum = false;
3488 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003489 mutex_unlock(&set_limit_mutex);
3490
3491 if (!ret)
3492 break;
3493
Johannes Weiner56600482012-01-12 17:17:59 -08003494 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3495 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003496 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3497 /* Usage is reduced ? */
3498 if (curusage >= oldusage)
3499 retry_count--;
3500 else
3501 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003502 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003503 if (!ret && enlarge)
3504 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003505
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003506 return ret;
3507}
3508
Li Zefan338c8432009-06-17 16:27:15 -07003509static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3510 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003511{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003512 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003513 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003514 int children = mem_cgroup_count_children(memcg);
3515 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003516 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003517
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003518 /* see mem_cgroup_resize_res_limit */
3519 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3520 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003521 while (retry_count) {
3522 if (signal_pending(current)) {
3523 ret = -EINTR;
3524 break;
3525 }
3526 /*
3527 * Rather than hide all in some function, I do this in
3528 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003529 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003530 */
3531 mutex_lock(&set_limit_mutex);
3532 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3533 if (memlimit > val) {
3534 ret = -EINVAL;
3535 mutex_unlock(&set_limit_mutex);
3536 break;
3537 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003538 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3539 if (memswlimit < val)
3540 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003541 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003542 if (!ret) {
3543 if (memlimit == val)
3544 memcg->memsw_is_minimum = true;
3545 else
3546 memcg->memsw_is_minimum = false;
3547 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003548 mutex_unlock(&set_limit_mutex);
3549
3550 if (!ret)
3551 break;
3552
Johannes Weiner56600482012-01-12 17:17:59 -08003553 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3554 MEM_CGROUP_RECLAIM_NOSWAP |
3555 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003556 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003557 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003558 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003559 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003560 else
3561 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003562 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003563 if (!ret && enlarge)
3564 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003565 return ret;
3566}
3567
Balbir Singh4e416952009-09-23 15:56:39 -07003568unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003569 gfp_t gfp_mask,
3570 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003571{
3572 unsigned long nr_reclaimed = 0;
3573 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3574 unsigned long reclaimed;
3575 int loop = 0;
3576 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003577 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003578 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003579
3580 if (order > 0)
3581 return 0;
3582
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003583 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003584 /*
3585 * This loop can run a while, specially if mem_cgroup's continuously
3586 * keep exceeding their soft limit and putting the system under
3587 * pressure
3588 */
3589 do {
3590 if (next_mz)
3591 mz = next_mz;
3592 else
3593 mz = mem_cgroup_largest_soft_limit_node(mctz);
3594 if (!mz)
3595 break;
3596
Ying Han0ae5e892011-05-26 16:25:25 -07003597 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003598 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08003599 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003600 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003601 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003602 spin_lock(&mctz->lock);
3603
3604 /*
3605 * If we failed to reclaim anything from this memory cgroup
3606 * it is time to move on to the next cgroup
3607 */
3608 next_mz = NULL;
3609 if (!reclaimed) {
3610 do {
3611 /*
3612 * Loop until we find yet another one.
3613 *
3614 * By the time we get the soft_limit lock
3615 * again, someone might have aded the
3616 * group back on the RB tree. Iterate to
3617 * make sure we get a different mem.
3618 * mem_cgroup_largest_soft_limit_node returns
3619 * NULL if no other cgroup is present on
3620 * the tree
3621 */
3622 next_mz =
3623 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003624 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003625 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003626 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003627 break;
3628 } while (1);
3629 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003630 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3631 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003632 /*
3633 * One school of thought says that we should not add
3634 * back the node to the tree if reclaim returns 0.
3635 * But our reclaim could return 0, simply because due
3636 * to priority we are exposing a smaller subset of
3637 * memory to reclaim from. Consider this as a longer
3638 * term TODO.
3639 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003640 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003641 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003642 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003643 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003644 loop++;
3645 /*
3646 * Could not reclaim anything and there are no more
3647 * mem cgroups to try or we seem to be looping without
3648 * reclaiming anything.
3649 */
3650 if (!nr_reclaimed &&
3651 (next_mz == NULL ||
3652 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3653 break;
3654 } while (!nr_reclaimed);
3655 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003656 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003657 return nr_reclaimed;
3658}
3659
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003660/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003661 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003662 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3663 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003664static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003665 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003666{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003667 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003668 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003669 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003670 struct page *busy;
3671 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003672 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003673
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003674 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003675 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003676 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003677
Hugh Dickins1eb49272012-03-21 16:34:19 -07003678 loop = mz->lru_size[lru];
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003679 /* give some margin against EBUSY etc...*/
3680 loop += 256;
3681 busy = NULL;
3682 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003683 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003684 struct page *page;
3685
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003686 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003687 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003688 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003689 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003690 break;
3691 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003692 page = list_entry(list->prev, struct page, lru);
3693 if (busy == page) {
3694 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003695 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003696 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003697 continue;
3698 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003699 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003700
Johannes Weiner925b7672012-01-12 17:18:15 -08003701 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003702
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003703 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003704 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003705 break;
3706
3707 if (ret == -EBUSY || ret == -EINVAL) {
3708 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003709 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003710 cond_resched();
3711 } else
3712 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003713 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003714
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003715 if (!ret && !list_empty(list))
3716 return -EBUSY;
3717 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003718}
3719
3720/*
3721 * make mem_cgroup's charge to be 0 if there is no task.
3722 * This enables deleting this mem_cgroup.
3723 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003724static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003725{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003726 int ret;
3727 int node, zid, shrink;
3728 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003729 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003730
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003731 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003732
3733 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003734 /* should free all ? */
3735 if (free_all)
3736 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003737move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003738 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003739 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003740 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003741 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003742 ret = -EINTR;
3743 if (signal_pending(current))
3744 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003745 /* This is for making all *used* pages to be on LRU. */
3746 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003747 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003748 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003749 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003750 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003751 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab932012-03-21 16:34:19 -07003752 enum lru_list lru;
3753 for_each_lru(lru) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003754 ret = mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab932012-03-21 16:34:19 -07003755 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003756 if (ret)
3757 break;
3758 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003759 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003760 if (ret)
3761 break;
3762 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003763 mem_cgroup_end_move(memcg);
3764 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003765 /* it seems parent cgroup doesn't have enough mem */
3766 if (ret == -ENOMEM)
3767 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003768 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003769 /* "ret" should also be checked to ensure all lists are empty. */
Glauber Costa569530f2012-04-12 12:49:13 -07003770 } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003771out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003772 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003773 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003774
3775try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003776 /* returns EBUSY if there is a task or if we come here twice. */
3777 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003778 ret = -EBUSY;
3779 goto out;
3780 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003781 /* we call try-to-free pages for make this cgroup empty */
3782 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003783 /* try to free all pages in this cgroup */
3784 shrink = 1;
Glauber Costa569530f2012-04-12 12:49:13 -07003785 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003786 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003787
3788 if (signal_pending(current)) {
3789 ret = -EINTR;
3790 goto out;
3791 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003792 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003793 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003794 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003795 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003796 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003797 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003798 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003799
3800 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003801 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003802 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003803 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003804}
3805
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003806int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3807{
3808 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3809}
3810
3811
Balbir Singh18f59ea2009-01-07 18:08:07 -08003812static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3813{
3814 return mem_cgroup_from_cont(cont)->use_hierarchy;
3815}
3816
3817static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3818 u64 val)
3819{
3820 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003821 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003822 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003823 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003824
3825 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003826 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003827
3828 cgroup_lock();
3829 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003830 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003831 * in the child subtrees. If it is unset, then the change can
3832 * occur, provided the current cgroup has no children.
3833 *
3834 * For the root cgroup, parent_mem is NULL, we allow value to be
3835 * set if there are no children.
3836 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003837 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003838 (val == 1 || val == 0)) {
3839 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003840 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003841 else
3842 retval = -EBUSY;
3843 } else
3844 retval = -EINVAL;
3845 cgroup_unlock();
3846
3847 return retval;
3848}
3849
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003850
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003851static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003852 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003853{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003854 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003855 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003856
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003857 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003858 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003859 val += mem_cgroup_read_stat(iter, idx);
3860
3861 if (val < 0) /* race ? */
3862 val = 0;
3863 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003864}
3865
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003866static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003867{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003868 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003869
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003870 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003871 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003872 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003873 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003874 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003875 }
3876
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003877 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3878 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003879
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003880 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003881 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003882
3883 return val << PAGE_SHIFT;
3884}
3885
Tejun Heoaf36f902012-04-01 12:09:55 -07003886static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
3887 struct file *file, char __user *buf,
3888 size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003890 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Tejun Heoaf36f902012-04-01 12:09:55 -07003891 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003892 u64 val;
Tejun Heoaf36f902012-04-01 12:09:55 -07003893 int type, name, len;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003894
3895 type = MEMFILE_TYPE(cft->private);
3896 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003897
3898 if (!do_swap_account && type == _MEMSWAP)
3899 return -EOPNOTSUPP;
3900
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003901 switch (type) {
3902 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003903 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003904 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003905 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003906 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003907 break;
3908 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003909 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003910 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003911 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003912 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003913 break;
3914 default:
3915 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003916 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003917
3918 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3919 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003920}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003921/*
3922 * The user of this function is...
3923 * RES_LIMIT.
3924 */
Paul Menage856c13a2008-07-25 01:47:04 -07003925static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3926 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003927{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003928 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003929 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003930 unsigned long long val;
3931 int ret;
3932
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003933 type = MEMFILE_TYPE(cft->private);
3934 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003935
3936 if (!do_swap_account && type == _MEMSWAP)
3937 return -EOPNOTSUPP;
3938
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003939 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003940 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003941 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3942 ret = -EINVAL;
3943 break;
3944 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003945 /* This function does all necessary parse...reuse it */
3946 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003947 if (ret)
3948 break;
3949 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003950 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003951 else
3952 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003953 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003954 case RES_SOFT_LIMIT:
3955 ret = res_counter_memparse_write_strategy(buffer, &val);
3956 if (ret)
3957 break;
3958 /*
3959 * For memsw, soft limits are hard to implement in terms
3960 * of semantics, for now, we support soft limits for
3961 * control without swap
3962 */
3963 if (type == _MEM)
3964 ret = res_counter_set_soft_limit(&memcg->res, val);
3965 else
3966 ret = -EINVAL;
3967 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003968 default:
3969 ret = -EINVAL; /* should be BUG() ? */
3970 break;
3971 }
3972 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003973}
3974
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003975static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3976 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3977{
3978 struct cgroup *cgroup;
3979 unsigned long long min_limit, min_memsw_limit, tmp;
3980
3981 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3982 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3983 cgroup = memcg->css.cgroup;
3984 if (!memcg->use_hierarchy)
3985 goto out;
3986
3987 while (cgroup->parent) {
3988 cgroup = cgroup->parent;
3989 memcg = mem_cgroup_from_cont(cgroup);
3990 if (!memcg->use_hierarchy)
3991 break;
3992 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3993 min_limit = min(min_limit, tmp);
3994 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3995 min_memsw_limit = min(min_memsw_limit, tmp);
3996 }
3997out:
3998 *mem_limit = min_limit;
3999 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004000}
4001
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004002static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004003{
Tejun Heoaf36f902012-04-01 12:09:55 -07004004 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004005 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004006
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004007 type = MEMFILE_TYPE(event);
4008 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07004009
4010 if (!do_swap_account && type == _MEMSWAP)
4011 return -EOPNOTSUPP;
4012
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004013 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004014 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004015 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004016 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004017 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004018 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004019 break;
4020 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004021 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004022 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004023 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004024 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004025 break;
4026 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004027
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07004028 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004029}
4030
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004031static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4032 struct cftype *cft)
4033{
4034 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4035}
4036
Daisuke Nishimura02491442010-03-10 15:22:17 -08004037#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004038static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4039 struct cftype *cft, u64 val)
4040{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004041 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004042
4043 if (val >= (1 << NR_MOVE_TYPE))
4044 return -EINVAL;
4045 /*
4046 * We check this value several times in both in can_attach() and
4047 * attach(), so we need cgroup lock to prevent this value from being
4048 * inconsistent.
4049 */
4050 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004051 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004052 cgroup_unlock();
4053
4054 return 0;
4055}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004056#else
4057static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4058 struct cftype *cft, u64 val)
4059{
4060 return -ENOSYS;
4061}
4062#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004063
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004064
4065/* For read statistics */
4066enum {
4067 MCS_CACHE,
4068 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004069 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004070 MCS_PGPGIN,
4071 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004072 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07004073 MCS_PGFAULT,
4074 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004075 MCS_INACTIVE_ANON,
4076 MCS_ACTIVE_ANON,
4077 MCS_INACTIVE_FILE,
4078 MCS_ACTIVE_FILE,
4079 MCS_UNEVICTABLE,
4080 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004081};
4082
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004083struct mcs_total_stat {
4084 s64 stat[NR_MCS_STAT];
4085};
4086
4087struct {
4088 char *local_name;
4089 char *total_name;
4090} memcg_stat_strings[NR_MCS_STAT] = {
4091 {"cache", "total_cache"},
4092 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004093 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004094 {"pgpgin", "total_pgpgin"},
4095 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004096 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004097 {"pgfault", "total_pgfault"},
4098 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004099 {"inactive_anon", "total_inactive_anon"},
4100 {"active_anon", "total_active_anon"},
4101 {"inactive_file", "total_inactive_file"},
4102 {"active_file", "total_active_file"},
4103 {"unevictable", "total_unevictable"}
4104};
4105
4106
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004107static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004108mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004109{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004110 s64 val;
4111
4112 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004113 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004114 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004115 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004116 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004117 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004118 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004119 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004120 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004121 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004122 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004123 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004124 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004125 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4126 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004127 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004128 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004129 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004130 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004131
4132 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004133 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004134 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004135 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004136 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004137 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004138 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004139 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004140 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004141 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004142 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004143}
4144
4145static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004146mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004147{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004148 struct mem_cgroup *iter;
4149
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004150 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004151 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004152}
4153
Ying Han406eb0c2011-05-26 16:25:37 -07004154#ifdef CONFIG_NUMA
4155static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4156{
4157 int nid;
4158 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4159 unsigned long node_nr;
4160 struct cgroup *cont = m->private;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004161 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07004162
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004163 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004164 seq_printf(m, "total=%lu", total_nr);
4165 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004166 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004167 seq_printf(m, " N%d=%lu", nid, node_nr);
4168 }
4169 seq_putc(m, '\n');
4170
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004171 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004172 seq_printf(m, "file=%lu", file_nr);
4173 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004174 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004175 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004176 seq_printf(m, " N%d=%lu", nid, node_nr);
4177 }
4178 seq_putc(m, '\n');
4179
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004180 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004181 seq_printf(m, "anon=%lu", anon_nr);
4182 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004183 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004184 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004185 seq_printf(m, " N%d=%lu", nid, node_nr);
4186 }
4187 seq_putc(m, '\n');
4188
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004189 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004190 seq_printf(m, "unevictable=%lu", unevictable_nr);
4191 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004192 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004193 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004194 seq_printf(m, " N%d=%lu", nid, node_nr);
4195 }
4196 seq_putc(m, '\n');
4197 return 0;
4198}
4199#endif /* CONFIG_NUMA */
4200
Paul Menagec64745c2008-04-29 01:00:02 -07004201static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4202 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004203{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004204 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004205 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004206 int i;
4207
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004208 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004209 mem_cgroup_get_local_stat(memcg, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004210
Ying Han406eb0c2011-05-26 16:25:37 -07004211
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004212 for (i = 0; i < NR_MCS_STAT; i++) {
4213 if (i == MCS_SWAP && !do_swap_account)
4214 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004215 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004216 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004217
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004218 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004219 {
4220 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004221 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004222 cb->fill(cb, "hierarchical_memory_limit", limit);
4223 if (do_swap_account)
4224 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4225 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004226
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004227 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004228 mem_cgroup_get_total_stat(memcg, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004229 for (i = 0; i < NR_MCS_STAT; i++) {
4230 if (i == MCS_SWAP && !do_swap_account)
4231 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004232 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004233 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004234
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004235#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004236 {
4237 int nid, zid;
4238 struct mem_cgroup_per_zone *mz;
4239 unsigned long recent_rotated[2] = {0, 0};
4240 unsigned long recent_scanned[2] = {0, 0};
4241
4242 for_each_online_node(nid)
4243 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004244 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004245
4246 recent_rotated[0] +=
4247 mz->reclaim_stat.recent_rotated[0];
4248 recent_rotated[1] +=
4249 mz->reclaim_stat.recent_rotated[1];
4250 recent_scanned[0] +=
4251 mz->reclaim_stat.recent_scanned[0];
4252 recent_scanned[1] +=
4253 mz->reclaim_stat.recent_scanned[1];
4254 }
4255 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4256 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4257 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4258 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4259 }
4260#endif
4261
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004262 return 0;
4263}
4264
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004265static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4266{
4267 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4268
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004269 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004270}
4271
4272static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4273 u64 val)
4274{
4275 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4276 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004277
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004278 if (val > 100)
4279 return -EINVAL;
4280
4281 if (cgrp->parent == NULL)
4282 return -EINVAL;
4283
4284 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004285
4286 cgroup_lock();
4287
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004288 /* If under hierarchy, only empty-root can set this value */
4289 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004290 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4291 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004292 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004293 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004294
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004295 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004296
Li Zefan068b38c2009-01-15 13:51:26 -08004297 cgroup_unlock();
4298
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004299 return 0;
4300}
4301
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004302static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4303{
4304 struct mem_cgroup_threshold_ary *t;
4305 u64 usage;
4306 int i;
4307
4308 rcu_read_lock();
4309 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004310 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004311 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004312 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004313
4314 if (!t)
4315 goto unlock;
4316
4317 usage = mem_cgroup_usage(memcg, swap);
4318
4319 /*
4320 * current_threshold points to threshold just below usage.
4321 * If it's not true, a threshold was crossed after last
4322 * call of __mem_cgroup_threshold().
4323 */
Phil Carmody5407a562010-05-26 14:42:42 -07004324 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004325
4326 /*
4327 * Iterate backward over array of thresholds starting from
4328 * current_threshold and check if a threshold is crossed.
4329 * If none of thresholds below usage is crossed, we read
4330 * only one element of the array here.
4331 */
4332 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4333 eventfd_signal(t->entries[i].eventfd, 1);
4334
4335 /* i = current_threshold + 1 */
4336 i++;
4337
4338 /*
4339 * Iterate forward over array of thresholds starting from
4340 * current_threshold+1 and check if a threshold is crossed.
4341 * If none of thresholds above usage is crossed, we read
4342 * only one element of the array here.
4343 */
4344 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4345 eventfd_signal(t->entries[i].eventfd, 1);
4346
4347 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004348 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004349unlock:
4350 rcu_read_unlock();
4351}
4352
4353static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4354{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004355 while (memcg) {
4356 __mem_cgroup_threshold(memcg, false);
4357 if (do_swap_account)
4358 __mem_cgroup_threshold(memcg, true);
4359
4360 memcg = parent_mem_cgroup(memcg);
4361 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004362}
4363
4364static int compare_thresholds(const void *a, const void *b)
4365{
4366 const struct mem_cgroup_threshold *_a = a;
4367 const struct mem_cgroup_threshold *_b = b;
4368
4369 return _a->threshold - _b->threshold;
4370}
4371
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004372static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004373{
4374 struct mem_cgroup_eventfd_list *ev;
4375
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004376 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004377 eventfd_signal(ev->eventfd, 1);
4378 return 0;
4379}
4380
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004381static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004382{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004383 struct mem_cgroup *iter;
4384
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004385 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004386 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004387}
4388
4389static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4390 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004391{
4392 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004393 struct mem_cgroup_thresholds *thresholds;
4394 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004395 int type = MEMFILE_TYPE(cft->private);
4396 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398
4399 ret = res_counter_memparse_write_strategy(args, &threshold);
4400 if (ret)
4401 return ret;
4402
4403 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004404
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004405 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004406 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004407 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004408 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004409 else
4410 BUG();
4411
4412 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4413
4414 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004415 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004416 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4417
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004418 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004419
4420 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004421 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004422 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004423 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004424 ret = -ENOMEM;
4425 goto unlock;
4426 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004427 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004428
4429 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004430 if (thresholds->primary) {
4431 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004432 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004433 }
4434
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004435 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004436 new->entries[size - 1].eventfd = eventfd;
4437 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004438
4439 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004440 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004441 compare_thresholds, NULL);
4442
4443 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004444 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004445 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004446 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004447 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004448 * new->current_threshold will not be used until
4449 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004450 * it here.
4451 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004452 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004453 }
4454 }
4455
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004456 /* Free old spare buffer and save old primary buffer as spare */
4457 kfree(thresholds->spare);
4458 thresholds->spare = thresholds->primary;
4459
4460 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004461
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004462 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004463 synchronize_rcu();
4464
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004465unlock:
4466 mutex_unlock(&memcg->thresholds_lock);
4467
4468 return ret;
4469}
4470
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004471static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004472 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004473{
4474 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004475 struct mem_cgroup_thresholds *thresholds;
4476 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004477 int type = MEMFILE_TYPE(cft->private);
4478 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004479 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004480
4481 mutex_lock(&memcg->thresholds_lock);
4482 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004483 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004484 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004485 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004486 else
4487 BUG();
4488
Anton Vorontsov371528c2012-02-24 05:14:46 +04004489 if (!thresholds->primary)
4490 goto unlock;
4491
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004492 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4493
4494 /* Check if a threshold crossed before removing */
4495 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4496
4497 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004498 size = 0;
4499 for (i = 0; i < thresholds->primary->size; i++) {
4500 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004501 size++;
4502 }
4503
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004504 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004505
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004506 /* Set thresholds array to NULL if we don't have thresholds */
4507 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004508 kfree(new);
4509 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004510 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511 }
4512
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004513 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004514
4515 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004516 new->current_threshold = -1;
4517 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4518 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004519 continue;
4520
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004521 new->entries[j] = thresholds->primary->entries[i];
4522 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004523 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004524 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004525 * until rcu_assign_pointer(), so it's safe to increment
4526 * it here.
4527 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004528 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004529 }
4530 j++;
4531 }
4532
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004533swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004534 /* Swap primary and spare array */
4535 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004536 /* If all events are unregistered, free the spare array */
4537 if (!new) {
4538 kfree(thresholds->spare);
4539 thresholds->spare = NULL;
4540 }
4541
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004542 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004543
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004544 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004545 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004546unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004547 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004548}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004549
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004550static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4551 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4552{
4553 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4554 struct mem_cgroup_eventfd_list *event;
4555 int type = MEMFILE_TYPE(cft->private);
4556
4557 BUG_ON(type != _OOM_TYPE);
4558 event = kmalloc(sizeof(*event), GFP_KERNEL);
4559 if (!event)
4560 return -ENOMEM;
4561
Michal Hocko1af8efe2011-07-26 16:08:24 -07004562 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004563
4564 event->eventfd = eventfd;
4565 list_add(&event->list, &memcg->oom_notify);
4566
4567 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004568 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004569 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004570 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004571
4572 return 0;
4573}
4574
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004575static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004576 struct cftype *cft, struct eventfd_ctx *eventfd)
4577{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004578 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004579 struct mem_cgroup_eventfd_list *ev, *tmp;
4580 int type = MEMFILE_TYPE(cft->private);
4581
4582 BUG_ON(type != _OOM_TYPE);
4583
Michal Hocko1af8efe2011-07-26 16:08:24 -07004584 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004585
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004586 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004587 if (ev->eventfd == eventfd) {
4588 list_del(&ev->list);
4589 kfree(ev);
4590 }
4591 }
4592
Michal Hocko1af8efe2011-07-26 16:08:24 -07004593 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004594}
4595
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004596static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4597 struct cftype *cft, struct cgroup_map_cb *cb)
4598{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004599 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004600
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004601 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004602
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004603 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004604 cb->fill(cb, "under_oom", 1);
4605 else
4606 cb->fill(cb, "under_oom", 0);
4607 return 0;
4608}
4609
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004610static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4611 struct cftype *cft, u64 val)
4612{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004613 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004614 struct mem_cgroup *parent;
4615
4616 /* cannot set to root cgroup and only 0 and 1 are allowed */
4617 if (!cgrp->parent || !((val == 0) || (val == 1)))
4618 return -EINVAL;
4619
4620 parent = mem_cgroup_from_cont(cgrp->parent);
4621
4622 cgroup_lock();
4623 /* oom-kill-disable is a flag for subhierarchy. */
4624 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004625 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004626 cgroup_unlock();
4627 return -EINVAL;
4628 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004629 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004630 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004631 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004632 cgroup_unlock();
4633 return 0;
4634}
4635
Ying Han406eb0c2011-05-26 16:25:37 -07004636#ifdef CONFIG_NUMA
4637static const struct file_operations mem_control_numa_stat_file_operations = {
4638 .read = seq_read,
4639 .llseek = seq_lseek,
4640 .release = single_release,
4641};
4642
4643static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4644{
4645 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4646
4647 file->f_op = &mem_control_numa_stat_file_operations;
4648 return single_open(file, mem_control_numa_stat_show, cont);
4649}
4650#endif /* CONFIG_NUMA */
4651
Glauber Costae5671df2011-12-11 21:47:01 +00004652#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03004653static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004654{
Glauber Costa1d62e432012-04-09 19:36:33 -03004655 return mem_cgroup_sockets_init(memcg, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004656};
4657
Glauber Costa1d62e432012-04-09 19:36:33 -03004658static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004659{
Glauber Costa1d62e432012-04-09 19:36:33 -03004660 mem_cgroup_sockets_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004661}
Glauber Costae5671df2011-12-11 21:47:01 +00004662#else
Glauber Costacbe128e32012-04-09 19:36:34 -03004663static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004664{
4665 return 0;
4666}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004667
Glauber Costa1d62e432012-04-09 19:36:33 -03004668static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004669{
4670}
Glauber Costae5671df2011-12-11 21:47:01 +00004671#endif
4672
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004673static struct cftype mem_cgroup_files[] = {
4674 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004675 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004676 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004677 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004678 .register_event = mem_cgroup_usage_register_event,
4679 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004680 },
4681 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004682 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004683 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004684 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004685 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004686 },
4687 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004688 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004689 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004690 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004691 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004692 },
4693 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004694 .name = "soft_limit_in_bytes",
4695 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4696 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004697 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07004698 },
4699 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004700 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004701 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004702 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004703 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004704 },
Balbir Singh8697d332008-02-07 00:13:59 -08004705 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004706 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004707 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004708 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004709 {
4710 .name = "force_empty",
4711 .trigger = mem_cgroup_force_empty_write,
4712 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004713 {
4714 .name = "use_hierarchy",
4715 .write_u64 = mem_cgroup_hierarchy_write,
4716 .read_u64 = mem_cgroup_hierarchy_read,
4717 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004718 {
4719 .name = "swappiness",
4720 .read_u64 = mem_cgroup_swappiness_read,
4721 .write_u64 = mem_cgroup_swappiness_write,
4722 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004723 {
4724 .name = "move_charge_at_immigrate",
4725 .read_u64 = mem_cgroup_move_charge_read,
4726 .write_u64 = mem_cgroup_move_charge_write,
4727 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004728 {
4729 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004730 .read_map = mem_cgroup_oom_control_read,
4731 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004732 .register_event = mem_cgroup_oom_register_event,
4733 .unregister_event = mem_cgroup_oom_unregister_event,
4734 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4735 },
Ying Han406eb0c2011-05-26 16:25:37 -07004736#ifdef CONFIG_NUMA
4737 {
4738 .name = "numa_stat",
4739 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004740 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004741 },
4742#endif
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004743#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004744 {
4745 .name = "memsw.usage_in_bytes",
4746 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004747 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004748 .register_event = mem_cgroup_usage_register_event,
4749 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004750 },
4751 {
4752 .name = "memsw.max_usage_in_bytes",
4753 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4754 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004755 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004756 },
4757 {
4758 .name = "memsw.limit_in_bytes",
4759 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4760 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004761 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004762 },
4763 {
4764 .name = "memsw.failcnt",
4765 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4766 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004767 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004768 },
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004769#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004770 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004771};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004772
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004773static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004774{
4775 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004776 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab932012-03-21 16:34:19 -07004777 enum lru_list lru;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004778 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004779 /*
4780 * This routine is called against possible nodes.
4781 * But it's BUG to call kmalloc() against offline node.
4782 *
4783 * TODO: this routine can waste much memory for nodes which will
4784 * never be onlined. It's better to use memory hotplug callback
4785 * function.
4786 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004787 if (!node_state(node, N_NORMAL_MEMORY))
4788 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004789 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004790 if (!pn)
4791 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004792
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004793 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4794 mz = &pn->zoneinfo[zone];
Hugh Dickinsf156ab932012-03-21 16:34:19 -07004795 for_each_lru(lru)
4796 INIT_LIST_HEAD(&mz->lruvec.lists[lru]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004797 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004798 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004799 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004800 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004801 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004802 return 0;
4803}
4804
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004805static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004806{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004807 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004808}
4809
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004810static struct mem_cgroup *mem_cgroup_alloc(void)
4811{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004812 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004813 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004814
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004815 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004816 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004817 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004818 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004819 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004820
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004821 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004822 return NULL;
4823
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004824 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4825 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004826 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004827 spin_lock_init(&memcg->pcp_counter_lock);
4828 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004829
4830out_free:
4831 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004832 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004833 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004834 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004835 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004836}
4837
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004838/*
Hugh Dickins59927fb2012-03-15 15:17:07 -07004839 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4840 * but in process context. The work_freeing structure is overlaid
4841 * on the rcu_freeing structure, which itself is overlaid on memsw.
4842 */
4843static void vfree_work(struct work_struct *work)
4844{
4845 struct mem_cgroup *memcg;
4846
4847 memcg = container_of(work, struct mem_cgroup, work_freeing);
4848 vfree(memcg);
4849}
4850static void vfree_rcu(struct rcu_head *rcu_head)
4851{
4852 struct mem_cgroup *memcg;
4853
4854 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4855 INIT_WORK(&memcg->work_freeing, vfree_work);
4856 schedule_work(&memcg->work_freeing);
4857}
4858
4859/*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004860 * At destroying mem_cgroup, references from swap_cgroup can remain.
4861 * (scanning all at force_empty is too costly...)
4862 *
4863 * Instead of clearing all references at force_empty, we remember
4864 * the number of reference from swap_cgroup and free mem_cgroup when
4865 * it goes down to 0.
4866 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004867 * Removal of cgroup itself succeeds regardless of refs from swap.
4868 */
4869
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004870static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004871{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004872 int node;
4873
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004874 mem_cgroup_remove_from_trees(memcg);
4875 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004876
Bob Liu3ed28fa2012-01-12 17:19:04 -08004877 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004878 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004879
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004880 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004881 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004882 kfree_rcu(memcg, rcu_freeing);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004883 else
Hugh Dickins59927fb2012-03-15 15:17:07 -07004884 call_rcu(&memcg->rcu_freeing, vfree_rcu);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004885}
4886
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004887static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004888{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004889 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004890}
4891
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004892static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004893{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004894 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4895 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4896 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004897 if (parent)
4898 mem_cgroup_put(parent);
4899 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004900}
4901
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004902static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004903{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004904 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004905}
4906
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004907/*
4908 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4909 */
Glauber Costae1aab162011-12-11 21:47:03 +00004910struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004911{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004912 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004913 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004914 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004915}
Glauber Costae1aab162011-12-11 21:47:03 +00004916EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004917
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004918#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4919static void __init enable_swap_cgroup(void)
4920{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004921 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004922 do_swap_account = 1;
4923}
4924#else
4925static void __init enable_swap_cgroup(void)
4926{
4927}
4928#endif
4929
Balbir Singhf64c3f52009-09-23 15:56:37 -07004930static int mem_cgroup_soft_limit_tree_init(void)
4931{
4932 struct mem_cgroup_tree_per_node *rtpn;
4933 struct mem_cgroup_tree_per_zone *rtpz;
4934 int tmp, node, zone;
4935
Bob Liu3ed28fa2012-01-12 17:19:04 -08004936 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004937 tmp = node;
4938 if (!node_state(node, N_NORMAL_MEMORY))
4939 tmp = -1;
4940 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4941 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004942 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004943
4944 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4945
4946 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4947 rtpz = &rtpn->rb_tree_per_zone[zone];
4948 rtpz->rb_root = RB_ROOT;
4949 spin_lock_init(&rtpz->lock);
4950 }
4951 }
4952 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004953
4954err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004955 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004956 if (!soft_limit_tree.rb_tree_per_node[node])
4957 break;
4958 kfree(soft_limit_tree.rb_tree_per_node[node]);
4959 soft_limit_tree.rb_tree_per_node[node] = NULL;
4960 }
4961 return 1;
4962
Balbir Singhf64c3f52009-09-23 15:56:37 -07004963}
4964
Li Zefan0eb253e2009-01-15 13:51:25 -08004965static struct cgroup_subsys_state * __ref
Li Zefan761b3ef52012-01-31 13:47:36 +08004966mem_cgroup_create(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004967{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004968 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004969 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004970 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004971
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004972 memcg = mem_cgroup_alloc();
4973 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004974 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004975
Bob Liu3ed28fa2012-01-12 17:19:04 -08004976 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004977 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004978 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004979
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004980 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004981 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004982 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004983 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004984 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004985 if (mem_cgroup_soft_limit_tree_init())
4986 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004987 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004988 for_each_possible_cpu(cpu) {
4989 struct memcg_stock_pcp *stock =
4990 &per_cpu(memcg_stock, cpu);
4991 INIT_WORK(&stock->work, drain_local_stock);
4992 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004993 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004994 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004995 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004996 memcg->use_hierarchy = parent->use_hierarchy;
4997 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004998 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004999
Balbir Singh18f59ea2009-01-07 18:08:07 -08005000 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005001 res_counter_init(&memcg->res, &parent->res);
5002 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005003 /*
5004 * We increment refcnt of the parent to ensure that we can
5005 * safely access it on res_counter_charge/uncharge.
5006 * This refcnt will be decremented when freeing this
5007 * mem_cgroup(see mem_cgroup_put).
5008 */
5009 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005010 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005011 res_counter_init(&memcg->res, NULL);
5012 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005013 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005014 memcg->last_scanned_node = MAX_NUMNODES;
5015 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08005016
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005017 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005018 memcg->swappiness = mem_cgroup_swappiness(parent);
5019 atomic_set(&memcg->refcnt, 1);
5020 memcg->move_charge_at_immigrate = 0;
5021 mutex_init(&memcg->thresholds_lock);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07005022 spin_lock_init(&memcg->move_lock);
Glauber Costacbe128e32012-04-09 19:36:34 -03005023
5024 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
5025 if (error) {
5026 /*
5027 * We call put now because our (and parent's) refcnts
5028 * are already in place. mem_cgroup_put() will internally
5029 * call __mem_cgroup_free, so return directly
5030 */
5031 mem_cgroup_put(memcg);
5032 return ERR_PTR(error);
5033 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005034 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005035free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005036 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005037 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005038}
5039
Li Zefan761b3ef52012-01-31 13:47:36 +08005040static int mem_cgroup_pre_destroy(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005041{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005042 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005043
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005044 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005045}
5046
Li Zefan761b3ef52012-01-31 13:47:36 +08005047static void mem_cgroup_destroy(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005048{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005049 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005050
Glauber Costa1d62e432012-04-09 19:36:33 -03005051 kmem_cgroup_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005052
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005053 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005054}
5055
Daisuke Nishimura02491442010-03-10 15:22:17 -08005056#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005057/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005058#define PRECHARGE_COUNT_AT_ONCE 256
5059static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005060{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005061 int ret = 0;
5062 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005063 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005064
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005065 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005066 mc.precharge += count;
5067 /* we don't need css_get for root */
5068 return ret;
5069 }
5070 /* try to charge at once */
5071 if (count > 1) {
5072 struct res_counter *dummy;
5073 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005074 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005075 * by cgroup_lock_live_cgroup() that it is not removed and we
5076 * are still under the same cgroup_mutex. So we can postpone
5077 * css_get().
5078 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005079 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005080 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005081 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005082 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005083 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005084 goto one_by_one;
5085 }
5086 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005087 return ret;
5088 }
5089one_by_one:
5090 /* fall back to one by one charge */
5091 while (count--) {
5092 if (signal_pending(current)) {
5093 ret = -EINTR;
5094 break;
5095 }
5096 if (!batch_count--) {
5097 batch_count = PRECHARGE_COUNT_AT_ONCE;
5098 cond_resched();
5099 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005100 ret = __mem_cgroup_try_charge(NULL,
5101 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005102 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005103 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005104 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005105 mc.precharge++;
5106 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005107 return ret;
5108}
5109
5110/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005111 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005112 * @vma: the vma the pte to be checked belongs
5113 * @addr: the address corresponding to the pte to be checked
5114 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005115 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005116 *
5117 * Returns
5118 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5119 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5120 * move charge. if @target is not NULL, the page is stored in target->page
5121 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005122 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5123 * target for charge migration. if @target is not NULL, the entry is stored
5124 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005125 *
5126 * Called with pte lock held.
5127 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005128union mc_target {
5129 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005130 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005131};
5132
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005133enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005134 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005135 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005136 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005137};
5138
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005139static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5140 unsigned long addr, pte_t ptent)
5141{
5142 struct page *page = vm_normal_page(vma, addr, ptent);
5143
5144 if (!page || !page_mapped(page))
5145 return NULL;
5146 if (PageAnon(page)) {
5147 /* we don't move shared anon */
Hugh Dickinsbe22aec2012-03-09 13:37:32 -08005148 if (!move_anon() || page_mapcount(page) > 2)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005149 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005150 } else if (!move_file())
5151 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005152 return NULL;
5153 if (!get_page_unless_zero(page))
5154 return NULL;
5155
5156 return page;
5157}
5158
5159static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5160 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5161{
5162 int usage_count;
5163 struct page *page = NULL;
5164 swp_entry_t ent = pte_to_swp_entry(ptent);
5165
5166 if (!move_anon() || non_swap_entry(ent))
5167 return NULL;
5168 usage_count = mem_cgroup_count_swap_user(ent, &page);
5169 if (usage_count > 1) { /* we don't move shared anon */
5170 if (page)
5171 put_page(page);
5172 return NULL;
5173 }
5174 if (do_swap_account)
5175 entry->val = ent.val;
5176
5177 return page;
5178}
5179
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005180static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5181 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5182{
5183 struct page *page = NULL;
5184 struct inode *inode;
5185 struct address_space *mapping;
5186 pgoff_t pgoff;
5187
5188 if (!vma->vm_file) /* anonymous vma */
5189 return NULL;
5190 if (!move_file())
5191 return NULL;
5192
5193 inode = vma->vm_file->f_path.dentry->d_inode;
5194 mapping = vma->vm_file->f_mapping;
5195 if (pte_none(ptent))
5196 pgoff = linear_page_index(vma, addr);
5197 else /* pte_file(ptent) is true */
5198 pgoff = pte_to_pgoff(ptent);
5199
5200 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005201 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005202
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005203#ifdef CONFIG_SWAP
5204 /* shmem/tmpfs may report page out on swap: account for that too. */
5205 if (radix_tree_exceptional_entry(page)) {
5206 swp_entry_t swap = radix_to_swp_entry(page);
5207 if (do_swap_account)
5208 *entry = swap;
5209 page = find_get_page(&swapper_space, swap.val);
5210 }
5211#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005212 return page;
5213}
5214
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005215static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005216 unsigned long addr, pte_t ptent, union mc_target *target)
5217{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005218 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005219 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005220 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005221 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005222
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005223 if (pte_present(ptent))
5224 page = mc_handle_present_pte(vma, addr, ptent);
5225 else if (is_swap_pte(ptent))
5226 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005227 else if (pte_none(ptent) || pte_file(ptent))
5228 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005229
5230 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005231 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005232 if (page) {
5233 pc = lookup_page_cgroup(page);
5234 /*
5235 * Do only loose check w/o page_cgroup lock.
5236 * mem_cgroup_move_account() checks the pc is valid or not under
5237 * the lock.
5238 */
5239 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5240 ret = MC_TARGET_PAGE;
5241 if (target)
5242 target->page = page;
5243 }
5244 if (!ret || !target)
5245 put_page(page);
5246 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005247 /* There is a swap entry and a page doesn't exist or isn't charged */
5248 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005249 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005250 ret = MC_TARGET_SWAP;
5251 if (target)
5252 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005253 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005254 return ret;
5255}
5256
Naoya Horiguchi12724852012-03-21 16:34:28 -07005257#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5258/*
5259 * We don't consider swapping or file mapped pages because THP does not
5260 * support them for now.
5261 * Caller should make sure that pmd_trans_huge(pmd) is true.
5262 */
5263static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5264 unsigned long addr, pmd_t pmd, union mc_target *target)
5265{
5266 struct page *page = NULL;
5267 struct page_cgroup *pc;
5268 enum mc_target_type ret = MC_TARGET_NONE;
5269
5270 page = pmd_page(pmd);
5271 VM_BUG_ON(!page || !PageHead(page));
5272 if (!move_anon())
5273 return ret;
5274 pc = lookup_page_cgroup(page);
5275 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5276 ret = MC_TARGET_PAGE;
5277 if (target) {
5278 get_page(page);
5279 target->page = page;
5280 }
5281 }
5282 return ret;
5283}
5284#else
5285static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5286 unsigned long addr, pmd_t pmd, union mc_target *target)
5287{
5288 return MC_TARGET_NONE;
5289}
5290#endif
5291
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005292static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5293 unsigned long addr, unsigned long end,
5294 struct mm_walk *walk)
5295{
5296 struct vm_area_struct *vma = walk->private;
5297 pte_t *pte;
5298 spinlock_t *ptl;
5299
Naoya Horiguchi12724852012-03-21 16:34:28 -07005300 if (pmd_trans_huge_lock(pmd, vma) == 1) {
5301 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5302 mc.precharge += HPAGE_PMD_NR;
5303 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005304 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005305 }
Dave Hansen03319322011-03-22 16:32:56 -07005306
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005307 if (pmd_trans_unstable(pmd))
5308 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005309 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5310 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005311 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005312 mc.precharge++; /* increment precharge temporarily */
5313 pte_unmap_unlock(pte - 1, ptl);
5314 cond_resched();
5315
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005316 return 0;
5317}
5318
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005319static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5320{
5321 unsigned long precharge;
5322 struct vm_area_struct *vma;
5323
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005324 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005325 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5326 struct mm_walk mem_cgroup_count_precharge_walk = {
5327 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5328 .mm = mm,
5329 .private = vma,
5330 };
5331 if (is_vm_hugetlb_page(vma))
5332 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005333 walk_page_range(vma->vm_start, vma->vm_end,
5334 &mem_cgroup_count_precharge_walk);
5335 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005336 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005337
5338 precharge = mc.precharge;
5339 mc.precharge = 0;
5340
5341 return precharge;
5342}
5343
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005344static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5345{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005346 unsigned long precharge = mem_cgroup_count_precharge(mm);
5347
5348 VM_BUG_ON(mc.moving_task);
5349 mc.moving_task = current;
5350 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005351}
5352
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005353/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5354static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005355{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005356 struct mem_cgroup *from = mc.from;
5357 struct mem_cgroup *to = mc.to;
5358
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005359 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005360 if (mc.precharge) {
5361 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5362 mc.precharge = 0;
5363 }
5364 /*
5365 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5366 * we must uncharge here.
5367 */
5368 if (mc.moved_charge) {
5369 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5370 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005371 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005372 /* we must fixup refcnts and charges */
5373 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005374 /* uncharge swap account from the old cgroup */
5375 if (!mem_cgroup_is_root(mc.from))
5376 res_counter_uncharge(&mc.from->memsw,
5377 PAGE_SIZE * mc.moved_swap);
5378 __mem_cgroup_put(mc.from, mc.moved_swap);
5379
5380 if (!mem_cgroup_is_root(mc.to)) {
5381 /*
5382 * we charged both to->res and to->memsw, so we should
5383 * uncharge to->res.
5384 */
5385 res_counter_uncharge(&mc.to->res,
5386 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005387 }
5388 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005389 mc.moved_swap = 0;
5390 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005391 memcg_oom_recover(from);
5392 memcg_oom_recover(to);
5393 wake_up_all(&mc.waitq);
5394}
5395
5396static void mem_cgroup_clear_mc(void)
5397{
5398 struct mem_cgroup *from = mc.from;
5399
5400 /*
5401 * we must clear moving_task before waking up waiters at the end of
5402 * task migration.
5403 */
5404 mc.moving_task = NULL;
5405 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005406 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005407 mc.from = NULL;
5408 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005409 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005410 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005411}
5412
Li Zefan761b3ef52012-01-31 13:47:36 +08005413static int mem_cgroup_can_attach(struct cgroup *cgroup,
5414 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005415{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005416 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005417 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005418 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005419
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005420 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005421 struct mm_struct *mm;
5422 struct mem_cgroup *from = mem_cgroup_from_task(p);
5423
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005424 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005425
5426 mm = get_task_mm(p);
5427 if (!mm)
5428 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005429 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005430 if (mm->owner == p) {
5431 VM_BUG_ON(mc.from);
5432 VM_BUG_ON(mc.to);
5433 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005434 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005435 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005436 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005437 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005438 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005439 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005440 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005441 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005442
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005443 ret = mem_cgroup_precharge_mc(mm);
5444 if (ret)
5445 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005446 }
5447 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005448 }
5449 return ret;
5450}
5451
Li Zefan761b3ef52012-01-31 13:47:36 +08005452static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5453 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005454{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005455 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005456}
5457
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005458static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5459 unsigned long addr, unsigned long end,
5460 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005461{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005462 int ret = 0;
5463 struct vm_area_struct *vma = walk->private;
5464 pte_t *pte;
5465 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005466 enum mc_target_type target_type;
5467 union mc_target target;
5468 struct page *page;
5469 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005470
Naoya Horiguchi12724852012-03-21 16:34:28 -07005471 /*
5472 * We don't take compound_lock() here but no race with splitting thp
5473 * happens because:
5474 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5475 * under splitting, which means there's no concurrent thp split,
5476 * - if another thread runs into split_huge_page() just after we
5477 * entered this if-block, the thread must wait for page table lock
5478 * to be unlocked in __split_huge_page_splitting(), where the main
5479 * part of thp split is not executed yet.
5480 */
5481 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005482 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005483 spin_unlock(&vma->vm_mm->page_table_lock);
5484 return 0;
5485 }
5486 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5487 if (target_type == MC_TARGET_PAGE) {
5488 page = target.page;
5489 if (!isolate_lru_page(page)) {
5490 pc = lookup_page_cgroup(page);
5491 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5492 pc, mc.from, mc.to,
5493 false)) {
5494 mc.precharge -= HPAGE_PMD_NR;
5495 mc.moved_charge += HPAGE_PMD_NR;
5496 }
5497 putback_lru_page(page);
5498 }
5499 put_page(page);
5500 }
5501 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005502 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005503 }
5504
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005505 if (pmd_trans_unstable(pmd))
5506 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005507retry:
5508 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5509 for (; addr != end; addr += PAGE_SIZE) {
5510 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005511 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005512
5513 if (!mc.precharge)
5514 break;
5515
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005516 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005517 case MC_TARGET_PAGE:
5518 page = target.page;
5519 if (isolate_lru_page(page))
5520 goto put;
5521 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005522 if (!mem_cgroup_move_account(page, 1, pc,
5523 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005524 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005525 /* we uncharge from mc.from later. */
5526 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005527 }
5528 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005529put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005530 put_page(page);
5531 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005532 case MC_TARGET_SWAP:
5533 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005534 if (!mem_cgroup_move_swap_account(ent,
5535 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005536 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005537 /* we fixup refcnts and charges later. */
5538 mc.moved_swap++;
5539 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005540 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005541 default:
5542 break;
5543 }
5544 }
5545 pte_unmap_unlock(pte - 1, ptl);
5546 cond_resched();
5547
5548 if (addr != end) {
5549 /*
5550 * We have consumed all precharges we got in can_attach().
5551 * We try charge one by one, but don't do any additional
5552 * charges to mc.to if we have failed in charge once in attach()
5553 * phase.
5554 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005555 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005556 if (!ret)
5557 goto retry;
5558 }
5559
5560 return ret;
5561}
5562
5563static void mem_cgroup_move_charge(struct mm_struct *mm)
5564{
5565 struct vm_area_struct *vma;
5566
5567 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005568retry:
5569 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5570 /*
5571 * Someone who are holding the mmap_sem might be waiting in
5572 * waitq. So we cancel all extra charges, wake up all waiters,
5573 * and retry. Because we cancel precharges, we might not be able
5574 * to move enough charges, but moving charge is a best-effort
5575 * feature anyway, so it wouldn't be a big problem.
5576 */
5577 __mem_cgroup_clear_mc();
5578 cond_resched();
5579 goto retry;
5580 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005581 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5582 int ret;
5583 struct mm_walk mem_cgroup_move_charge_walk = {
5584 .pmd_entry = mem_cgroup_move_charge_pte_range,
5585 .mm = mm,
5586 .private = vma,
5587 };
5588 if (is_vm_hugetlb_page(vma))
5589 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005590 ret = walk_page_range(vma->vm_start, vma->vm_end,
5591 &mem_cgroup_move_charge_walk);
5592 if (ret)
5593 /*
5594 * means we have consumed all precharges and failed in
5595 * doing additional charge. Just abandon here.
5596 */
5597 break;
5598 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005599 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005600}
5601
Li Zefan761b3ef52012-01-31 13:47:36 +08005602static void mem_cgroup_move_task(struct cgroup *cont,
5603 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005604{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005605 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005606 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005607
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005608 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005609 if (mc.to)
5610 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005611 mmput(mm);
5612 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005613 if (mc.to)
5614 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005615}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005616#else /* !CONFIG_MMU */
Li Zefan761b3ef52012-01-31 13:47:36 +08005617static int mem_cgroup_can_attach(struct cgroup *cgroup,
5618 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005619{
5620 return 0;
5621}
Li Zefan761b3ef52012-01-31 13:47:36 +08005622static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5623 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005624{
5625}
Li Zefan761b3ef52012-01-31 13:47:36 +08005626static void mem_cgroup_move_task(struct cgroup *cont,
5627 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005628{
5629}
5630#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005631
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005632struct cgroup_subsys mem_cgroup_subsys = {
5633 .name = "memory",
5634 .subsys_id = mem_cgroup_subsys_id,
5635 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005636 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005637 .destroy = mem_cgroup_destroy,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005638 .can_attach = mem_cgroup_can_attach,
5639 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005640 .attach = mem_cgroup_move_task,
Tejun Heo6bc10342012-04-01 12:09:55 -07005641 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005642 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005643 .use_id = 1,
Tejun Heo48ddbe12012-04-01 12:09:56 -07005644 .__DEPRECATED_clear_css_refs = true,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005645};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005646
5647#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005648static int __init enable_swap_account(char *s)
5649{
5650 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005651 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005652 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005653 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005654 really_do_swap_account = 0;
5655 return 1;
5656}
Michal Hockoa2c89902011-05-24 17:12:50 -07005657__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005658
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005659#endif