blob: 795e525afaba8914f3f8863de6a7299d84898ada [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
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070062static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
Andrew Mortonc255a452012-07-31 16:43:02 -070064#ifdef CONFIG_MEMCG_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*/
Andrew Mortonc255a452012-07-31 16:43:02 -070069#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080070static 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
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070076#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080077#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 */
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -070090 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080091 MEM_CGROUP_STAT_NSTATS,
92};
93
Johannes Weineraf7c4b02012-05-29 15:07:08 -070094static const char * const mem_cgroup_stat_names[] = {
95 "cache",
96 "rss",
97 "mapped_file",
98 "swap",
99};
100
Johannes Weinere9f89742011-03-23 16:42:37 -0700101enum mem_cgroup_events_index {
102 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
103 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700104 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
105 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700106 MEM_CGROUP_EVENTS_NSTATS,
107};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108
109static const char * const mem_cgroup_events_names[] = {
110 "pgpgin",
111 "pgpgout",
112 "pgfault",
113 "pgmajfault",
114};
115
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700116/*
117 * Per memcg event counter is incremented at every pagein/pageout. With THP,
118 * it will be incremated by the number of pages. This counter is used for
119 * for trigger some periodic events. This is straightforward and better
120 * than using jiffies etc. to handle periodic memcg event.
121 */
122enum mem_cgroup_events_target {
123 MEM_CGROUP_TARGET_THRESH,
124 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700125 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700126 MEM_CGROUP_NTARGETS,
127};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700128#define THRESHOLDS_EVENTS_TARGET 128
129#define SOFTLIMIT_EVENTS_TARGET 1024
130#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700131
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800132struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700133 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700134 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700135 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700136 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800137};
138
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800139struct mem_cgroup_reclaim_iter {
140 /* css_id of the last scanned hierarchy member */
141 int position;
142 /* scan generation, increased every round-trip */
143 unsigned int generation;
144};
145
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800146/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800147 * per-zone information in memory controller.
148 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800149struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800150 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700151 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800152
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800153 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
154
Balbir Singhf64c3f52009-09-23 15:56:37 -0700155 struct rb_node tree_node; /* RB tree node */
156 unsigned long long usage_in_excess;/* Set to the value by which */
157 /* the soft limit is exceeded*/
158 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700159 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700160 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800161};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800162
163struct mem_cgroup_per_node {
164 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
165};
166
167struct mem_cgroup_lru_info {
168 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
169};
170
171/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700172 * Cgroups above their limits are maintained in a RB-Tree, independent of
173 * their hierarchy representation
174 */
175
176struct mem_cgroup_tree_per_zone {
177 struct rb_root rb_root;
178 spinlock_t lock;
179};
180
181struct mem_cgroup_tree_per_node {
182 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
183};
184
185struct mem_cgroup_tree {
186 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
187};
188
189static struct mem_cgroup_tree soft_limit_tree __read_mostly;
190
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800191struct mem_cgroup_threshold {
192 struct eventfd_ctx *eventfd;
193 u64 threshold;
194};
195
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700196/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800197struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700198 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700199 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800200 /* Size of entries[] */
201 unsigned int size;
202 /* Array of thresholds */
203 struct mem_cgroup_threshold entries[0];
204};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700205
206struct mem_cgroup_thresholds {
207 /* Primary thresholds array */
208 struct mem_cgroup_threshold_ary *primary;
209 /*
210 * Spare threshold array.
211 * This is needed to make mem_cgroup_unregister_event() "never fail".
212 * It must be able to store at least primary->size - 1 entries.
213 */
214 struct mem_cgroup_threshold_ary *spare;
215};
216
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700217/* for OOM */
218struct mem_cgroup_eventfd_list {
219 struct list_head list;
220 struct eventfd_ctx *eventfd;
221};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800222
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700223static void mem_cgroup_threshold(struct mem_cgroup *memcg);
224static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800225
Balbir Singhf64c3f52009-09-23 15:56:37 -0700226/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800227 * The memory controller data structure. The memory controller controls both
228 * page cache and RSS per cgroup. We would eventually like to provide
229 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
230 * to help the administrator determine what knobs to tune.
231 *
232 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800233 * we hit the water mark. May be even add a low water mark, such that
234 * no reclaim occurs from a cgroup at it's low water mark, this is
235 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800236 */
237struct mem_cgroup {
238 struct cgroup_subsys_state css;
239 /*
240 * the counter to account for memory usage
241 */
242 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700243
244 union {
245 /*
246 * the counter to account for mem+swap usage.
247 */
248 struct res_counter memsw;
249
250 /*
251 * rcu_freeing is used only when freeing struct mem_cgroup,
252 * so put it into a union to avoid wasting more memory.
253 * It must be disjoint from the css field. It could be
254 * in a union with the res field, but res plays a much
255 * larger part in mem_cgroup life than memsw, and might
256 * be of interest, even at time of free, when debugging.
257 * So share rcu_head with the less interesting memsw.
258 */
259 struct rcu_head rcu_freeing;
260 /*
Glauber Costa3afe36b2012-05-29 15:07:10 -0700261 * We also need some space for a worker in deferred freeing.
262 * By the time we call it, rcu_freeing is no longer in use.
Hugh Dickins59927fb2012-03-15 15:17:07 -0700263 */
264 struct work_struct work_freeing;
265 };
266
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800267 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800268 * Per cgroup active and inactive list, similar to the
269 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800270 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800271 struct mem_cgroup_lru_info info;
Ying Han889976d2011-05-26 16:25:33 -0700272 int last_scanned_node;
273#if MAX_NUMNODES > 1
274 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700275 atomic_t numainfo_events;
276 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700277#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800278 /*
279 * Should the accounting and control be hierarchical, per subtree?
280 */
281 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700282
283 bool oom_lock;
284 atomic_t under_oom;
285
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800286 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800287
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700288 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700289 /* OOM-Killer disable */
290 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800291
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700292 /* set when res.limit == memsw.limit */
293 bool memsw_is_minimum;
294
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800295 /* protect arrays of thresholds */
296 struct mutex thresholds_lock;
297
298 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700299 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700300
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800301 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700302 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700303
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700304 /* For oom notifier event fd */
305 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700306
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800307 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800308 * Should we move charges of a task when a task is moved into this
309 * mem_cgroup ? And what type of charges should we move ?
310 */
311 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800312 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700313 * set > 0 if pages under this cgroup are moving to other cgroup.
314 */
315 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700316 /* taken only while moving_account > 0 */
317 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700318 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800319 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800320 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700321 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700322 /*
323 * used when a cpu is offlined or other synchronizations
324 * See mem_cgroup_read_stat().
325 */
326 struct mem_cgroup_stat_cpu nocpu_base;
327 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000328
329#ifdef CONFIG_INET
330 struct tcp_memcontrol tcp_mem;
331#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800332};
333
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800334/* Stuffs for move charges at task migration. */
335/*
336 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
337 * left-shifted bitmap of these types.
338 */
339enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800340 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700341 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800342 NR_MOVE_TYPE,
343};
344
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800345/* "mc" and its members are protected by cgroup_mutex */
346static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800347 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800348 struct mem_cgroup *from;
349 struct mem_cgroup *to;
350 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800351 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800352 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800353 struct task_struct *moving_task; /* a task moving charges */
354 wait_queue_head_t waitq; /* a waitq for other context */
355} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700356 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800357 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
358};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800359
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700360static bool move_anon(void)
361{
362 return test_bit(MOVE_CHARGE_TYPE_ANON,
363 &mc.to->move_charge_at_immigrate);
364}
365
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700366static bool move_file(void)
367{
368 return test_bit(MOVE_CHARGE_TYPE_FILE,
369 &mc.to->move_charge_at_immigrate);
370}
371
Balbir Singh4e416952009-09-23 15:56:39 -0700372/*
373 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
374 * limit reclaim to prevent infinite loops, if they ever occur.
375 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700376#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
377#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700378
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800379enum charge_type {
380 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700381 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800382 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700383 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700384 NR_CHARGE_TYPE,
385};
386
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800387/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000388#define _MEM (0)
389#define _MEMSWAP (1)
390#define _OOM_TYPE (2)
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700391#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
392#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800393#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700394/* Used for OOM nofiier */
395#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800396
Balbir Singh75822b42009-09-23 15:56:38 -0700397/*
398 * Reclaim flags for mem_cgroup_hierarchical_reclaim
399 */
400#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
401#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
402#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
403#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
404
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700405static void mem_cgroup_get(struct mem_cgroup *memcg);
406static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000407
Wanpeng Lib2145142012-07-31 16:46:01 -0700408static inline
409struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
410{
411 return container_of(s, struct mem_cgroup, css);
412}
413
Glauber Costae1aab162011-12-11 21:47:03 +0000414/* Writing them here to avoid exposing memcg's inner layout */
Andrew Mortonc255a452012-07-31 16:43:02 -0700415#ifdef CONFIG_MEMCG_KMEM
Glauber Costae1aab162011-12-11 21:47:03 +0000416#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000417#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000418
419static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
420void sock_update_memcg(struct sock *sk)
421{
Glauber Costa376be5f2012-01-20 04:57:14 +0000422 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000423 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700424 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000425
426 BUG_ON(!sk->sk_prot->proto_cgroup);
427
Glauber Costaf3f511e2012-01-05 20:16:39 +0000428 /* Socket cloning can throw us here with sk_cgrp already
429 * filled. It won't however, necessarily happen from
430 * process context. So the test for root memcg given
431 * the current task's memcg won't help us in this case.
432 *
433 * Respecting the original socket's memcg is a better
434 * decision in this case.
435 */
436 if (sk->sk_cgrp) {
437 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
438 mem_cgroup_get(sk->sk_cgrp->memcg);
439 return;
440 }
441
Glauber Costae1aab162011-12-11 21:47:03 +0000442 rcu_read_lock();
443 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700444 cg_proto = sk->sk_prot->proto_cgroup(memcg);
445 if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
Glauber Costae1aab162011-12-11 21:47:03 +0000446 mem_cgroup_get(memcg);
Glauber Costa3f134612012-05-29 15:07:11 -0700447 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000448 }
449 rcu_read_unlock();
450 }
451}
452EXPORT_SYMBOL(sock_update_memcg);
453
454void sock_release_memcg(struct sock *sk)
455{
Glauber Costa376be5f2012-01-20 04:57:14 +0000456 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000457 struct mem_cgroup *memcg;
458 WARN_ON(!sk->sk_cgrp->memcg);
459 memcg = sk->sk_cgrp->memcg;
460 mem_cgroup_put(memcg);
461 }
462}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000463
Glauber Costa319d3b92012-01-15 22:04:39 +0000464#ifdef CONFIG_INET
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000465struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
466{
467 if (!memcg || mem_cgroup_is_root(memcg))
468 return NULL;
469
470 return &memcg->tcp_mem.cg_proto;
471}
472EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000473#endif /* CONFIG_INET */
Andrew Mortonc255a452012-07-31 16:43:02 -0700474#endif /* CONFIG_MEMCG_KMEM */
Glauber Costae1aab162011-12-11 21:47:03 +0000475
Andrew Mortonc255a452012-07-31 16:43:02 -0700476#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costa3f134612012-05-29 15:07:11 -0700477static void disarm_sock_keys(struct mem_cgroup *memcg)
478{
479 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
480 return;
481 static_key_slow_dec(&memcg_socket_limit_enabled);
482}
483#else
484static void disarm_sock_keys(struct mem_cgroup *memcg)
485{
486}
487#endif
488
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700489static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800490
Balbir Singhf64c3f52009-09-23 15:56:37 -0700491static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700492mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700493{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700494 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700495}
496
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700497struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100498{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700499 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100500}
501
Balbir Singhf64c3f52009-09-23 15:56:37 -0700502static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700503page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700504{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700505 int nid = page_to_nid(page);
506 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700507
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700508 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700509}
510
511static struct mem_cgroup_tree_per_zone *
512soft_limit_tree_node_zone(int nid, int zid)
513{
514 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
515}
516
517static struct mem_cgroup_tree_per_zone *
518soft_limit_tree_from_page(struct page *page)
519{
520 int nid = page_to_nid(page);
521 int zid = page_zonenum(page);
522
523 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
524}
525
526static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700527__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700528 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700529 struct mem_cgroup_tree_per_zone *mctz,
530 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700531{
532 struct rb_node **p = &mctz->rb_root.rb_node;
533 struct rb_node *parent = NULL;
534 struct mem_cgroup_per_zone *mz_node;
535
536 if (mz->on_tree)
537 return;
538
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700539 mz->usage_in_excess = new_usage_in_excess;
540 if (!mz->usage_in_excess)
541 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700542 while (*p) {
543 parent = *p;
544 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
545 tree_node);
546 if (mz->usage_in_excess < mz_node->usage_in_excess)
547 p = &(*p)->rb_left;
548 /*
549 * We can't avoid mem cgroups that are over their soft
550 * limit by the same amount
551 */
552 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
553 p = &(*p)->rb_right;
554 }
555 rb_link_node(&mz->tree_node, parent, p);
556 rb_insert_color(&mz->tree_node, &mctz->rb_root);
557 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700558}
559
560static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700561__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700562 struct mem_cgroup_per_zone *mz,
563 struct mem_cgroup_tree_per_zone *mctz)
564{
565 if (!mz->on_tree)
566 return;
567 rb_erase(&mz->tree_node, &mctz->rb_root);
568 mz->on_tree = false;
569}
570
571static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700572mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700573 struct mem_cgroup_per_zone *mz,
574 struct mem_cgroup_tree_per_zone *mctz)
575{
576 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700577 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700578 spin_unlock(&mctz->lock);
579}
580
Balbir Singhf64c3f52009-09-23 15:56:37 -0700581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700582static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700583{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700584 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700585 struct mem_cgroup_per_zone *mz;
586 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700587 int nid = page_to_nid(page);
588 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700589 mctz = soft_limit_tree_from_page(page);
590
591 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700592 * Necessary to update all ancestors when hierarchy is used.
593 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700594 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700595 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
596 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
597 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700598 /*
599 * We have to update the tree if mz is on RB-tree or
600 * mem is over its softlimit.
601 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700602 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700603 spin_lock(&mctz->lock);
604 /* if on-tree, remove it */
605 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700607 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700608 * Insert again. mz->usage_in_excess will be updated.
609 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700610 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700611 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700612 spin_unlock(&mctz->lock);
613 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700614 }
615}
616
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700617static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700618{
619 int node, zone;
620 struct mem_cgroup_per_zone *mz;
621 struct mem_cgroup_tree_per_zone *mctz;
622
Bob Liu3ed28fa2012-01-12 17:19:04 -0800623 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700624 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700625 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700626 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700627 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700628 }
629 }
630}
631
Balbir Singh4e416952009-09-23 15:56:39 -0700632static struct mem_cgroup_per_zone *
633__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
634{
635 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700636 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700637
638retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700639 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700640 rightmost = rb_last(&mctz->rb_root);
641 if (!rightmost)
642 goto done; /* Nothing to reclaim from */
643
644 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
645 /*
646 * Remove the node now but someone else can add it back,
647 * we will to add it back at the end of reclaim to its correct
648 * position in the tree.
649 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700650 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
651 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
652 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700653 goto retry;
654done:
655 return mz;
656}
657
658static struct mem_cgroup_per_zone *
659mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
660{
661 struct mem_cgroup_per_zone *mz;
662
663 spin_lock(&mctz->lock);
664 mz = __mem_cgroup_largest_soft_limit_node(mctz);
665 spin_unlock(&mctz->lock);
666 return mz;
667}
668
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700669/*
670 * Implementation Note: reading percpu statistics for memcg.
671 *
672 * Both of vmstat[] and percpu_counter has threshold and do periodic
673 * synchronization to implement "quick" read. There are trade-off between
674 * reading cost and precision of value. Then, we may have a chance to implement
675 * a periodic synchronizion of counter in memcg's counter.
676 *
677 * But this _read() function is used for user interface now. The user accounts
678 * memory usage by memory cgroup and he _always_ requires exact value because
679 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
680 * have to visit all online cpus and make sum. So, for now, unnecessary
681 * synchronization is not implemented. (just implemented for cpu hotplug)
682 *
683 * If there are kernel internal actions which can make use of some not-exact
684 * value, and reading all cpu value can be performance bottleneck in some
685 * common workload, threashold and synchonization as vmstat[] should be
686 * implemented.
687 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700689 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800690{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700691 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800692 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800693
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700694 get_online_cpus();
695 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700696 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700697#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700698 spin_lock(&memcg->pcp_counter_lock);
699 val += memcg->nocpu_base.count[idx];
700 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700701#endif
702 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800703 return val;
704}
705
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700706static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700707 bool charge)
708{
709 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700710 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700711}
712
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700713static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700714 enum mem_cgroup_events_index idx)
715{
716 unsigned long val = 0;
717 int cpu;
718
719 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700720 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700721#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700722 spin_lock(&memcg->pcp_counter_lock);
723 val += memcg->nocpu_base.events[idx];
724 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700725#endif
726 return val;
727}
728
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700729static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700730 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800731{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800732 preempt_disable();
733
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700734 /*
735 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
736 * counted as CACHE even if it's on ANON LRU.
737 */
738 if (anon)
739 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700740 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800741 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700742 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700744
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800745 /* pagein of a big page is an event. So, ignore page size */
746 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700747 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800748 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700749 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800750 nr_pages = -nr_pages; /* for event */
751 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800752
Johannes Weiner13114712012-05-29 15:07:07 -0700753 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800754
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800755 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800756}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800757
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700758unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700759mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700760{
761 struct mem_cgroup_per_zone *mz;
762
763 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
764 return mz->lru_size[lru];
765}
766
767static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700768mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700769 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700770{
771 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700772 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700773 unsigned long ret = 0;
774
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700775 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700776
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700777 for_each_lru(lru) {
778 if (BIT(lru) & lru_mask)
779 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700780 }
781 return ret;
782}
783
784static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700785mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700786 int nid, unsigned int lru_mask)
787{
Ying Han889976d2011-05-26 16:25:33 -0700788 u64 total = 0;
789 int zid;
790
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700791 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700792 total += mem_cgroup_zone_nr_lru_pages(memcg,
793 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700794
Ying Han889976d2011-05-26 16:25:33 -0700795 return total;
796}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700797
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700798static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700799 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800800{
Ying Han889976d2011-05-26 16:25:33 -0700801 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800802 u64 total = 0;
803
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700804 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700805 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800806 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800807}
808
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800809static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
810 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800811{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700812 unsigned long val, next;
813
Johannes Weiner13114712012-05-29 15:07:07 -0700814 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700815 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700816 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800817 if ((long)next - (long)val < 0) {
818 switch (target) {
819 case MEM_CGROUP_TARGET_THRESH:
820 next = val + THRESHOLDS_EVENTS_TARGET;
821 break;
822 case MEM_CGROUP_TARGET_SOFTLIMIT:
823 next = val + SOFTLIMIT_EVENTS_TARGET;
824 break;
825 case MEM_CGROUP_TARGET_NUMAINFO:
826 next = val + NUMAINFO_EVENTS_TARGET;
827 break;
828 default:
829 break;
830 }
831 __this_cpu_write(memcg->stat->targets[target], next);
832 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700833 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800834 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800835}
836
837/*
838 * Check events in order.
839 *
840 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700841static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800842{
Steven Rostedt47994012011-11-02 13:38:33 -0700843 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800844 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800845 if (unlikely(mem_cgroup_event_ratelimit(memcg,
846 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800847 bool do_softlimit;
848 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800849
850 do_softlimit = mem_cgroup_event_ratelimit(memcg,
851 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700852#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800853 do_numainfo = mem_cgroup_event_ratelimit(memcg,
854 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700855#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800856 preempt_enable();
857
858 mem_cgroup_threshold(memcg);
859 if (unlikely(do_softlimit))
860 mem_cgroup_update_tree(memcg, page);
861#if MAX_NUMNODES > 1
862 if (unlikely(do_numainfo))
863 atomic_inc(&memcg->numainfo_events);
864#endif
865 } else
866 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800867}
868
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000869struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800870{
Wanpeng Lib2145142012-07-31 16:46:01 -0700871 return mem_cgroup_from_css(
872 cgroup_subsys_state(cont, mem_cgroup_subsys_id));
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800873}
874
Balbir Singhcf475ad2008-04-29 01:00:16 -0700875struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800876{
Balbir Singh31a78f22008-09-28 23:09:31 +0100877 /*
878 * mm_update_next_owner() may clear mm->owner to NULL
879 * if it races with swapoff, page migration, etc.
880 * So this can be called with p == NULL.
881 */
882 if (unlikely(!p))
883 return NULL;
884
Wanpeng Lib2145142012-07-31 16:46:01 -0700885 return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800886}
887
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700888struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700890 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700891
892 if (!mm)
893 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800894 /*
895 * Because we have no locks, mm->owner's may be being moved to other
896 * cgroup. We use css_tryget() here even if this looks
897 * pessimistic (rather than adding locks here).
898 */
899 rcu_read_lock();
900 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700901 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
902 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800903 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700904 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800905 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700906 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800907}
908
Johannes Weiner56600482012-01-12 17:17:59 -0800909/**
910 * mem_cgroup_iter - iterate over memory cgroup hierarchy
911 * @root: hierarchy root
912 * @prev: previously returned memcg, NULL on first invocation
913 * @reclaim: cookie for shared reclaim walks, NULL for full walks
914 *
915 * Returns references to children of the hierarchy below @root, or
916 * @root itself, or %NULL after a full round-trip.
917 *
918 * Caller must pass the return value in @prev on subsequent
919 * invocations for reference counting, or use mem_cgroup_iter_break()
920 * to cancel a hierarchy walk before the round-trip is complete.
921 *
922 * Reclaimers can specify a zone and a priority level in @reclaim to
923 * divide up the memcgs in the hierarchy among all concurrent
924 * reclaimers operating on the same zone and priority.
925 */
926struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
927 struct mem_cgroup *prev,
928 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700929{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800930 struct mem_cgroup *memcg = NULL;
931 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700932
Johannes Weiner56600482012-01-12 17:17:59 -0800933 if (mem_cgroup_disabled())
934 return NULL;
935
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700936 if (!root)
937 root = root_mem_cgroup;
938
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800939 if (prev && !reclaim)
940 id = css_id(&prev->css);
941
942 if (prev && prev != root)
943 css_put(&prev->css);
944
945 if (!root->use_hierarchy && root != root_mem_cgroup) {
946 if (prev)
947 return NULL;
948 return root;
949 }
950
951 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800952 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800953 struct cgroup_subsys_state *css;
954
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800955 if (reclaim) {
956 int nid = zone_to_nid(reclaim->zone);
957 int zid = zone_idx(reclaim->zone);
958 struct mem_cgroup_per_zone *mz;
959
960 mz = mem_cgroup_zoneinfo(root, nid, zid);
961 iter = &mz->reclaim_iter[reclaim->priority];
962 if (prev && reclaim->generation != iter->generation)
963 return NULL;
964 id = iter->position;
965 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800966
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700967 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800968 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
969 if (css) {
970 if (css == &root->css || css_tryget(css))
Wanpeng Lib2145142012-07-31 16:46:01 -0700971 memcg = mem_cgroup_from_css(css);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800972 } else
973 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700974 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700975
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800976 if (reclaim) {
977 iter->position = id;
978 if (!css)
979 iter->generation++;
980 else if (!prev && memcg)
981 reclaim->generation = iter->generation;
982 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800983
984 if (prev && !css)
985 return NULL;
986 }
987 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700988}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800989
Johannes Weiner56600482012-01-12 17:17:59 -0800990/**
991 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
992 * @root: hierarchy root
993 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
994 */
995void mem_cgroup_iter_break(struct mem_cgroup *root,
996 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800997{
998 if (!root)
999 root = root_mem_cgroup;
1000 if (prev && prev != root)
1001 css_put(&prev->css);
1002}
1003
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001004/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001005 * Iteration constructs for visiting all cgroups (under a tree). If
1006 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1007 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001008 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001009#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001010 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001011 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001012 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001013
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001014#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001015 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001016 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001017 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001018
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001019static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -07001020{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001021 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001022}
1023
Ying Han456f9982011-05-26 16:25:38 -07001024void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1025{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001026 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001027
1028 if (!mm)
1029 return;
1030
1031 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001032 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1033 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001034 goto out;
1035
1036 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001037 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001038 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1039 break;
1040 case PGMAJFAULT:
1041 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001042 break;
1043 default:
1044 BUG();
1045 }
1046out:
1047 rcu_read_unlock();
1048}
1049EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1050
Johannes Weiner925b7672012-01-12 17:18:15 -08001051/**
1052 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1053 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001054 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001055 *
1056 * Returns the lru list vector holding pages for the given @zone and
1057 * @mem. This can be the global zone lruvec, if the memory controller
1058 * is disabled.
1059 */
1060struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1061 struct mem_cgroup *memcg)
1062{
1063 struct mem_cgroup_per_zone *mz;
1064
1065 if (mem_cgroup_disabled())
1066 return &zone->lruvec;
1067
1068 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1069 return &mz->lruvec;
1070}
1071
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001072/*
1073 * Following LRU functions are allowed to be used without PCG_LOCK.
1074 * Operations are called by routine of global LRU independently from memcg.
1075 * What we have to take care of here is validness of pc->mem_cgroup.
1076 *
1077 * Changes to pc->mem_cgroup happens when
1078 * 1. charge
1079 * 2. moving account
1080 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1081 * It is added to LRU before charge.
1082 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1083 * When moving account, the page is not on LRU. It's isolated.
1084 */
1085
Johannes Weiner925b7672012-01-12 17:18:15 -08001086/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001087 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001088 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001089 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001090 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001091struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001092{
1093 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001094 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001095 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001096
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001097 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001098 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001099
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001100 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001101 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001102
1103 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001104 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001105 * an uncharged page off lru does nothing to secure
1106 * its former mem_cgroup from sudden removal.
1107 *
1108 * Our caller holds lru_lock, and PageCgroupUsed is updated
1109 * under page_cgroup lock: between them, they make all uses
1110 * of pc->mem_cgroup safe.
1111 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001112 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001113 pc->mem_cgroup = memcg = root_mem_cgroup;
1114
Johannes Weiner925b7672012-01-12 17:18:15 -08001115 mz = page_cgroup_zoneinfo(memcg, page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001116 return &mz->lruvec;
1117}
1118
1119/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001120 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1121 * @lruvec: mem_cgroup per zone lru vector
1122 * @lru: index of lru list the page is sitting on
1123 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001124 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001125 * This function must be called when a page is added to or removed from an
1126 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001127 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001128void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1129 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001130{
1131 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001132 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001133
1134 if (mem_cgroup_disabled())
1135 return;
1136
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001137 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1138 lru_size = mz->lru_size + lru;
1139 *lru_size += nr_pages;
1140 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001141}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001142
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001143/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001144 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001145 * hierarchy subtree
1146 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001147bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1148 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001149{
Johannes Weiner91c637342012-05-29 15:06:24 -07001150 if (root_memcg == memcg)
1151 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001152 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001153 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001154 return css_is_ancestor(&memcg->css, &root_memcg->css);
1155}
1156
1157static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1158 struct mem_cgroup *memcg)
1159{
1160 bool ret;
1161
Johannes Weiner91c637342012-05-29 15:06:24 -07001162 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001163 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001164 rcu_read_unlock();
1165 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001166}
1167
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001168int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001169{
1170 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001171 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001172 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001173
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001174 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001175 if (p) {
1176 curr = try_get_mem_cgroup_from_mm(p->mm);
1177 task_unlock(p);
1178 } else {
1179 /*
1180 * All threads may have already detached their mm's, but the oom
1181 * killer still needs to detect if they have already been oom
1182 * killed to prevent needlessly killing additional tasks.
1183 */
1184 task_lock(task);
1185 curr = mem_cgroup_from_task(task);
1186 if (curr)
1187 css_get(&curr->css);
1188 task_unlock(task);
1189 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001190 if (!curr)
1191 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001192 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001193 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001194 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001195 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1196 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001197 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001198 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001199 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001200 return ret;
1201}
1202
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001203int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001204{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001205 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001206 unsigned long inactive;
1207 unsigned long active;
1208 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001209
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001210 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1211 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001212
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001213 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1214 if (gb)
1215 inactive_ratio = int_sqrt(10 * gb);
1216 else
1217 inactive_ratio = 1;
1218
Johannes Weiner9b272972011-11-02 13:38:23 -07001219 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001220}
1221
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001222int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001223{
1224 unsigned long active;
1225 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001226
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001227 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE);
1228 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE);
Rik van Riel56e49d22009-06-16 15:32:28 -07001229
1230 return (active > inactive);
1231}
1232
Balbir Singh6d61ef42009-01-07 18:08:06 -08001233#define mem_cgroup_from_res_counter(counter, member) \
1234 container_of(counter, struct mem_cgroup, member)
1235
Johannes Weiner19942822011-02-01 15:52:43 -08001236/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001237 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001238 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001239 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001240 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001241 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001242 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001243static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001244{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001245 unsigned long long margin;
1246
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001247 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001248 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001249 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001250 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001251}
1252
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001253int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001254{
1255 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001256
1257 /* root ? */
1258 if (cgrp->parent == NULL)
1259 return vm_swappiness;
1260
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001261 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001262}
1263
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001264/*
1265 * memcg->moving_account is used for checking possibility that some thread is
1266 * calling move_account(). When a thread on CPU-A starts moving pages under
1267 * a memcg, other threads should check memcg->moving_account under
1268 * rcu_read_lock(), like this:
1269 *
1270 * CPU-A CPU-B
1271 * rcu_read_lock()
1272 * memcg->moving_account+1 if (memcg->mocing_account)
1273 * take heavy locks.
1274 * synchronize_rcu() update something.
1275 * rcu_read_unlock()
1276 * start move here.
1277 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001278
1279/* for quick checking without looking up memcg */
1280atomic_t memcg_moving __read_mostly;
1281
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001282static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001283{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001284 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001285 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001286 synchronize_rcu();
1287}
1288
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001289static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001290{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001291 /*
1292 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1293 * We check NULL in callee rather than caller.
1294 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001295 if (memcg) {
1296 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001297 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001298 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001299}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001300
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001301/*
1302 * 2 routines for checking "mem" is under move_account() or not.
1303 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001304 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1305 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001306 * pc->mem_cgroup may be overwritten.
1307 *
1308 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1309 * under hierarchy of moving cgroups. This is for
1310 * waiting at hith-memory prressure caused by "move".
1311 */
1312
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001313static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001314{
1315 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001316 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001317}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001318
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001319static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001320{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001321 struct mem_cgroup *from;
1322 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001323 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001324 /*
1325 * Unlike task_move routines, we access mc.to, mc.from not under
1326 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1327 */
1328 spin_lock(&mc.lock);
1329 from = mc.from;
1330 to = mc.to;
1331 if (!from)
1332 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001333
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001334 ret = mem_cgroup_same_or_subtree(memcg, from)
1335 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001336unlock:
1337 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001338 return ret;
1339}
1340
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001341static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001342{
1343 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001344 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001345 DEFINE_WAIT(wait);
1346 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1347 /* moving charge context might have finished. */
1348 if (mc.moving_task)
1349 schedule();
1350 finish_wait(&mc.waitq, &wait);
1351 return true;
1352 }
1353 }
1354 return false;
1355}
1356
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001357/*
1358 * Take this lock when
1359 * - a code tries to modify page's memcg while it's USED.
1360 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001361 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001362 */
1363static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1364 unsigned long *flags)
1365{
1366 spin_lock_irqsave(&memcg->move_lock, *flags);
1367}
1368
1369static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1370 unsigned long *flags)
1371{
1372 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1373}
1374
Balbir Singhe2224322009-04-02 16:57:39 -07001375/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001376 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001377 * @memcg: The memory cgroup that went over limit
1378 * @p: Task that is going to be killed
1379 *
1380 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1381 * enabled
1382 */
1383void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1384{
1385 struct cgroup *task_cgrp;
1386 struct cgroup *mem_cgrp;
1387 /*
1388 * Need a buffer in BSS, can't rely on allocations. The code relies
1389 * on the assumption that OOM is serialized for memory controller.
1390 * If this assumption is broken, revisit this code.
1391 */
1392 static char memcg_name[PATH_MAX];
1393 int ret;
1394
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001395 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001396 return;
1397
Balbir Singhe2224322009-04-02 16:57:39 -07001398 rcu_read_lock();
1399
1400 mem_cgrp = memcg->css.cgroup;
1401 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1402
1403 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1404 if (ret < 0) {
1405 /*
1406 * Unfortunately, we are unable to convert to a useful name
1407 * But we'll still print out the usage information
1408 */
1409 rcu_read_unlock();
1410 goto done;
1411 }
1412 rcu_read_unlock();
1413
1414 printk(KERN_INFO "Task in %s killed", memcg_name);
1415
1416 rcu_read_lock();
1417 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1418 if (ret < 0) {
1419 rcu_read_unlock();
1420 goto done;
1421 }
1422 rcu_read_unlock();
1423
1424 /*
1425 * Continues from above, so we don't need an KERN_ level
1426 */
1427 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1428done:
1429
1430 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1431 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1432 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1433 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1434 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1435 "failcnt %llu\n",
1436 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1437 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1438 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1439}
1440
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001441/*
1442 * This function returns the number of memcg under hierarchy tree. Returns
1443 * 1(self count) if no children.
1444 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001445static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001446{
1447 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001448 struct mem_cgroup *iter;
1449
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001450 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001451 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001452 return num;
1453}
1454
Balbir Singh6d61ef42009-01-07 18:08:06 -08001455/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001456 * Return the memory (and swap, if configured) limit for a memcg.
1457 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001458static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001459{
1460 u64 limit;
1461 u64 memsw;
1462
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001463 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1464 limit += total_swap_pages << PAGE_SHIFT;
1465
David Rientjesa63d83f2010-08-09 17:19:46 -07001466 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1467 /*
1468 * If memsw is finite and limits the amount of swap space available
1469 * to this memcg, return that limit.
1470 */
1471 return min(limit, memsw);
1472}
1473
David Rientjes876aafb2012-07-31 16:43:48 -07001474void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1475 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001476{
1477 struct mem_cgroup *iter;
1478 unsigned long chosen_points = 0;
1479 unsigned long totalpages;
1480 unsigned int points = 0;
1481 struct task_struct *chosen = NULL;
1482
David Rientjes876aafb2012-07-31 16:43:48 -07001483 /*
1484 * If current has a pending SIGKILL, then automatically select it. The
1485 * goal is to allow it to allocate so that it may quickly exit and free
1486 * its memory.
1487 */
1488 if (fatal_signal_pending(current)) {
1489 set_thread_flag(TIF_MEMDIE);
1490 return;
1491 }
1492
1493 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001494 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1495 for_each_mem_cgroup_tree(iter, memcg) {
1496 struct cgroup *cgroup = iter->css.cgroup;
1497 struct cgroup_iter it;
1498 struct task_struct *task;
1499
1500 cgroup_iter_start(cgroup, &it);
1501 while ((task = cgroup_iter_next(cgroup, &it))) {
1502 switch (oom_scan_process_thread(task, totalpages, NULL,
1503 false)) {
1504 case OOM_SCAN_SELECT:
1505 if (chosen)
1506 put_task_struct(chosen);
1507 chosen = task;
1508 chosen_points = ULONG_MAX;
1509 get_task_struct(chosen);
1510 /* fall through */
1511 case OOM_SCAN_CONTINUE:
1512 continue;
1513 case OOM_SCAN_ABORT:
1514 cgroup_iter_end(cgroup, &it);
1515 mem_cgroup_iter_break(memcg, iter);
1516 if (chosen)
1517 put_task_struct(chosen);
1518 return;
1519 case OOM_SCAN_OK:
1520 break;
1521 };
1522 points = oom_badness(task, memcg, NULL, totalpages);
1523 if (points > chosen_points) {
1524 if (chosen)
1525 put_task_struct(chosen);
1526 chosen = task;
1527 chosen_points = points;
1528 get_task_struct(chosen);
1529 }
1530 }
1531 cgroup_iter_end(cgroup, &it);
1532 }
1533
1534 if (!chosen)
1535 return;
1536 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001537 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1538 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001539}
1540
Johannes Weiner56600482012-01-12 17:17:59 -08001541static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1542 gfp_t gfp_mask,
1543 unsigned long flags)
1544{
1545 unsigned long total = 0;
1546 bool noswap = false;
1547 int loop;
1548
1549 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1550 noswap = true;
1551 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1552 noswap = true;
1553
1554 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1555 if (loop)
1556 drain_all_stock_async(memcg);
1557 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1558 /*
1559 * Allow limit shrinkers, which are triggered directly
1560 * by userspace, to catch signals and stop reclaim
1561 * after minimal progress, regardless of the margin.
1562 */
1563 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1564 break;
1565 if (mem_cgroup_margin(memcg))
1566 break;
1567 /*
1568 * If nothing was reclaimed after two attempts, there
1569 * may be no reclaimable pages in this hierarchy.
1570 */
1571 if (loop && !total)
1572 break;
1573 }
1574 return total;
1575}
1576
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001577/**
1578 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001579 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001580 * @nid: the node ID to be checked.
1581 * @noswap : specify true here if the user wants flle only information.
1582 *
1583 * This function returns whether the specified memcg contains any
1584 * reclaimable pages on a node. Returns true if there are any reclaimable
1585 * pages in the node.
1586 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001587static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001588 int nid, bool noswap)
1589{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001590 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001591 return true;
1592 if (noswap || !total_swap_pages)
1593 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001594 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001595 return true;
1596 return false;
1597
1598}
Ying Han889976d2011-05-26 16:25:33 -07001599#if MAX_NUMNODES > 1
1600
1601/*
1602 * Always updating the nodemask is not very good - even if we have an empty
1603 * list or the wrong list here, we can start from some node and traverse all
1604 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1605 *
1606 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001607static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001608{
1609 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001610 /*
1611 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1612 * pagein/pageout changes since the last update.
1613 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001614 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001615 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001616 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001617 return;
1618
Ying Han889976d2011-05-26 16:25:33 -07001619 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001620 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001621
1622 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1623
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001624 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1625 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001626 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001627
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001628 atomic_set(&memcg->numainfo_events, 0);
1629 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001630}
1631
1632/*
1633 * Selecting a node where we start reclaim from. Because what we need is just
1634 * reducing usage counter, start from anywhere is O,K. Considering
1635 * memory reclaim from current node, there are pros. and cons.
1636 *
1637 * Freeing memory from current node means freeing memory from a node which
1638 * we'll use or we've used. So, it may make LRU bad. And if several threads
1639 * hit limits, it will see a contention on a node. But freeing from remote
1640 * node means more costs for memory reclaim because of memory latency.
1641 *
1642 * Now, we use round-robin. Better algorithm is welcomed.
1643 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001644int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001645{
1646 int node;
1647
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001648 mem_cgroup_may_update_nodemask(memcg);
1649 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001650
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001651 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001652 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001653 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001654 /*
1655 * We call this when we hit limit, not when pages are added to LRU.
1656 * No LRU may hold pages because all pages are UNEVICTABLE or
1657 * memcg is too small and all pages are not on LRU. In that case,
1658 * we use curret node.
1659 */
1660 if (unlikely(node == MAX_NUMNODES))
1661 node = numa_node_id();
1662
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001663 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001664 return node;
1665}
1666
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001667/*
1668 * Check all nodes whether it contains reclaimable pages or not.
1669 * For quick scan, we make use of scan_nodes. This will allow us to skip
1670 * unused nodes. But scan_nodes is lazily updated and may not cotain
1671 * enough new information. We need to do double check.
1672 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001673static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001674{
1675 int nid;
1676
1677 /*
1678 * quick check...making use of scan_node.
1679 * We can skip unused nodes.
1680 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001681 if (!nodes_empty(memcg->scan_nodes)) {
1682 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001683 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001684 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001686 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001687 return true;
1688 }
1689 }
1690 /*
1691 * Check rest of nodes.
1692 */
1693 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001694 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001695 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001696 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001697 return true;
1698 }
1699 return false;
1700}
1701
Ying Han889976d2011-05-26 16:25:33 -07001702#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001703int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001704{
1705 return 0;
1706}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001707
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001708static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001709{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001710 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001711}
Ying Han889976d2011-05-26 16:25:33 -07001712#endif
1713
Johannes Weiner56600482012-01-12 17:17:59 -08001714static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1715 struct zone *zone,
1716 gfp_t gfp_mask,
1717 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001718{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001719 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001720 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001721 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001722 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001723 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001724 struct mem_cgroup_reclaim_cookie reclaim = {
1725 .zone = zone,
1726 .priority = 0,
1727 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001728
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001729 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001730
Balbir Singh4e416952009-09-23 15:56:39 -07001731 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001732 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001733 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001734 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001735 if (loop >= 2) {
1736 /*
1737 * If we have not been able to reclaim
1738 * anything, it might because there are
1739 * no reclaimable pages under this hierarchy
1740 */
Johannes Weiner56600482012-01-12 17:17:59 -08001741 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001742 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001743 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001744 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001745 * excess >> 2 is not to excessive so as to
1746 * reclaim too much, nor too less that we keep
1747 * coming back to reclaim from this cgroup
1748 */
1749 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001750 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001751 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001752 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001753 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001754 }
Johannes Weiner56600482012-01-12 17:17:59 -08001755 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001756 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001757 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1758 zone, &nr_scanned);
1759 *total_scanned += nr_scanned;
1760 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001761 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001762 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001763 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001764 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001765}
1766
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001767/*
1768 * Check OOM-Killer is already running under our hierarchy.
1769 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001770 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001771 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001772static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001773{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001774 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001775
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001776 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001777 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001778 /*
1779 * this subtree of our hierarchy is already locked
1780 * so we cannot give a lock.
1781 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001782 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001783 mem_cgroup_iter_break(memcg, iter);
1784 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001785 } else
1786 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001787 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001788
Michal Hocko79dfdac2011-07-26 16:08:23 -07001789 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001790 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001791
1792 /*
1793 * OK, we failed to lock the whole subtree so we have to clean up
1794 * what we set up to the failing subtree
1795 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001796 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001797 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001798 mem_cgroup_iter_break(memcg, iter);
1799 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001800 }
1801 iter->oom_lock = false;
1802 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001803 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001804}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001805
Michal Hocko79dfdac2011-07-26 16:08:23 -07001806/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001807 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001808 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001809static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001810{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001811 struct mem_cgroup *iter;
1812
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001813 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001814 iter->oom_lock = false;
1815 return 0;
1816}
1817
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001818static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001819{
1820 struct mem_cgroup *iter;
1821
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001822 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001823 atomic_inc(&iter->under_oom);
1824}
1825
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001826static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001827{
1828 struct mem_cgroup *iter;
1829
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001830 /*
1831 * When a new child is created while the hierarchy is under oom,
1832 * mem_cgroup_oom_lock() may not be called. We have to use
1833 * atomic_add_unless() here.
1834 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001835 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001836 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001837}
1838
Michal Hocko1af8efe2011-07-26 16:08:24 -07001839static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001840static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1841
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001842struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001843 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001844 wait_queue_t wait;
1845};
1846
1847static int memcg_oom_wake_function(wait_queue_t *wait,
1848 unsigned mode, int sync, void *arg)
1849{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001850 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1851 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001852 struct oom_wait_info *oom_wait_info;
1853
1854 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001855 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001856
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001857 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001858 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001859 * Then we can use css_is_ancestor without taking care of RCU.
1860 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001861 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1862 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001863 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001864 return autoremove_wake_function(wait, mode, sync, arg);
1865}
1866
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001867static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001868{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001869 /* for filtering, pass "memcg" as argument. */
1870 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001871}
1872
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001873static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001874{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001875 if (memcg && atomic_read(&memcg->under_oom))
1876 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001877}
1878
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001879/*
1880 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1881 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001882static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
1883 int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001884{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001885 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001886 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001887
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001888 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001889 owait.wait.flags = 0;
1890 owait.wait.func = memcg_oom_wake_function;
1891 owait.wait.private = current;
1892 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001893 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001894 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001895
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001896 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001897 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001898 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001899 /*
1900 * Even if signal_pending(), we can't quit charge() loop without
1901 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1902 * under OOM is always welcomed, use TASK_KILLABLE here.
1903 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001904 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001905 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001906 need_to_kill = false;
1907 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001908 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001909 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001910
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001911 if (need_to_kill) {
1912 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07001913 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001914 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001915 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001916 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001917 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001918 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001919 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001920 mem_cgroup_oom_unlock(memcg);
1921 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001922 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001923
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001924 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001925
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001926 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1927 return false;
1928 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001929 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001930 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001931}
1932
Balbir Singhd69b0422009-06-17 16:26:34 -07001933/*
1934 * Currently used to update mapped file statistics, but the routine can be
1935 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001936 *
1937 * Notes: Race condition
1938 *
1939 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1940 * it tends to be costly. But considering some conditions, we doesn't need
1941 * to do so _always_.
1942 *
1943 * Considering "charge", lock_page_cgroup() is not required because all
1944 * file-stat operations happen after a page is attached to radix-tree. There
1945 * are no race with "charge".
1946 *
1947 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1948 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1949 * if there are race with "uncharge". Statistics itself is properly handled
1950 * by flags.
1951 *
1952 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001953 * small, we check mm->moving_account and detect there are possibility of race
1954 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001955 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001956
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001957void __mem_cgroup_begin_update_page_stat(struct page *page,
1958 bool *locked, unsigned long *flags)
1959{
1960 struct mem_cgroup *memcg;
1961 struct page_cgroup *pc;
1962
1963 pc = lookup_page_cgroup(page);
1964again:
1965 memcg = pc->mem_cgroup;
1966 if (unlikely(!memcg || !PageCgroupUsed(pc)))
1967 return;
1968 /*
1969 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07001970 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001971 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001972 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001973 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001974 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001975 return;
1976
1977 move_lock_mem_cgroup(memcg, flags);
1978 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
1979 move_unlock_mem_cgroup(memcg, flags);
1980 goto again;
1981 }
1982 *locked = true;
1983}
1984
1985void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
1986{
1987 struct page_cgroup *pc = lookup_page_cgroup(page);
1988
1989 /*
1990 * It's guaranteed that pc->mem_cgroup never changes while
1991 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07001992 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001993 */
1994 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1995}
1996
Greg Thelen2a7106f2011-01-13 15:47:37 -08001997void mem_cgroup_update_page_stat(struct page *page,
1998 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001999{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002000 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002001 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002002 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002003
Johannes Weinercfa44942012-01-12 17:18:38 -08002004 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002005 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002006
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002007 memcg = pc->mem_cgroup;
2008 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002009 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002010
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002011 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08002012 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08002013 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002014 break;
2015 default:
2016 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002017 }
Balbir Singhd69b0422009-06-17 16:26:34 -07002018
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002019 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002020}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002021
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002022/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002023 * size of first charge trial. "32" comes from vmscan.c's magic value.
2024 * TODO: maybe necessary to use big numbers in big irons.
2025 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002026#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002027struct memcg_stock_pcp {
2028 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002029 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002030 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002031 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002032#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002033};
2034static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002035static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002036
2037/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002038 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002039 * from local stock and true is returned. If the stock is 0 or charges from a
2040 * cgroup which is not current target, returns false. This stock will be
2041 * refilled.
2042 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002043static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002044{
2045 struct memcg_stock_pcp *stock;
2046 bool ret = true;
2047
2048 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002049 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002050 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002051 else /* need to call res_counter_charge */
2052 ret = false;
2053 put_cpu_var(memcg_stock);
2054 return ret;
2055}
2056
2057/*
2058 * Returns stocks cached in percpu to res_counter and reset cached information.
2059 */
2060static void drain_stock(struct memcg_stock_pcp *stock)
2061{
2062 struct mem_cgroup *old = stock->cached;
2063
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002064 if (stock->nr_pages) {
2065 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2066
2067 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002068 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002069 res_counter_uncharge(&old->memsw, bytes);
2070 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002071 }
2072 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002073}
2074
2075/*
2076 * This must be called under preempt disabled or must be called by
2077 * a thread which is pinned to local cpu.
2078 */
2079static void drain_local_stock(struct work_struct *dummy)
2080{
2081 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2082 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002083 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002084}
2085
2086/*
2087 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002088 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002089 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002090static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002091{
2092 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2093
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002094 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002095 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002096 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002097 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002098 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002099 put_cpu_var(memcg_stock);
2100}
2101
2102/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002103 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002104 * of the hierarchy under it. sync flag says whether we should block
2105 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002106 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002107static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002108{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002109 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002110
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002111 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002112 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002113 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002114 for_each_online_cpu(cpu) {
2115 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002116 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002117
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002118 memcg = stock->cached;
2119 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002120 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002121 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002122 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002123 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2124 if (cpu == curcpu)
2125 drain_local_stock(&stock->work);
2126 else
2127 schedule_work_on(cpu, &stock->work);
2128 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002129 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002130 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002131
2132 if (!sync)
2133 goto out;
2134
2135 for_each_online_cpu(cpu) {
2136 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002137 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002138 flush_work(&stock->work);
2139 }
2140out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002141 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002142}
2143
2144/*
2145 * Tries to drain stocked charges in other cpus. This function is asynchronous
2146 * and just put a work per cpu for draining localy on each cpu. Caller can
2147 * expects some charges will be back to res_counter later but cannot wait for
2148 * it.
2149 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002150static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002151{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002152 /*
2153 * If someone calls draining, avoid adding more kworker runs.
2154 */
2155 if (!mutex_trylock(&percpu_charge_mutex))
2156 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002157 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002158 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002159}
2160
2161/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002162static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002163{
2164 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002165 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002166 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002167 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002168}
2169
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002170/*
2171 * This function drains percpu counter value from DEAD cpu and
2172 * move it to local cpu. Note that this function can be preempted.
2173 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002174static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002175{
2176 int i;
2177
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002178 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002179 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002180 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002181
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002182 per_cpu(memcg->stat->count[i], cpu) = 0;
2183 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002184 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002185 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002186 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002187
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002188 per_cpu(memcg->stat->events[i], cpu) = 0;
2189 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002190 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002191 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002192}
2193
2194static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002195 unsigned long action,
2196 void *hcpu)
2197{
2198 int cpu = (unsigned long)hcpu;
2199 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002200 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002201
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002202 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002203 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002204
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002205 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002206 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002207
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002208 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002209 mem_cgroup_drain_pcp_counter(iter, cpu);
2210
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002211 stock = &per_cpu(memcg_stock, cpu);
2212 drain_stock(stock);
2213 return NOTIFY_OK;
2214}
2215
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002216
2217/* See __mem_cgroup_try_charge() for details */
2218enum {
2219 CHARGE_OK, /* success */
2220 CHARGE_RETRY, /* need to retry but retry is not bad */
2221 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2222 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2223 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2224};
2225
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002226static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002227 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002228{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002229 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002230 struct mem_cgroup *mem_over_limit;
2231 struct res_counter *fail_res;
2232 unsigned long flags = 0;
2233 int ret;
2234
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002235 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002236
2237 if (likely(!ret)) {
2238 if (!do_swap_account)
2239 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002240 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002241 if (likely(!ret))
2242 return CHARGE_OK;
2243
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002244 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002245 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2246 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2247 } else
2248 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002249 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002250 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2251 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002252 *
2253 * Never reclaim on behalf of optional batching, retry with a
2254 * single page instead.
2255 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002256 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002257 return CHARGE_RETRY;
2258
2259 if (!(gfp_mask & __GFP_WAIT))
2260 return CHARGE_WOULDBLOCK;
2261
Johannes Weiner56600482012-01-12 17:17:59 -08002262 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002263 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002264 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002265 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002266 * Even though the limit is exceeded at this point, reclaim
2267 * may have been able to free some pages. Retry the charge
2268 * before killing the task.
2269 *
2270 * Only for regular pages, though: huge pages are rather
2271 * unlikely to succeed so close to the limit, and we fall back
2272 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002273 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002274 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002275 return CHARGE_RETRY;
2276
2277 /*
2278 * At task move, charge accounts can be doubly counted. So, it's
2279 * better to wait until the end of task_move if something is going on.
2280 */
2281 if (mem_cgroup_wait_acct_move(mem_over_limit))
2282 return CHARGE_RETRY;
2283
2284 /* If we don't need to call oom-killer at el, return immediately */
2285 if (!oom_check)
2286 return CHARGE_NOMEM;
2287 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002288 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002289 return CHARGE_OOM_DIE;
2290
2291 return CHARGE_RETRY;
2292}
2293
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002294/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002295 * __mem_cgroup_try_charge() does
2296 * 1. detect memcg to be charged against from passed *mm and *ptr,
2297 * 2. update res_counter
2298 * 3. call memory reclaim if necessary.
2299 *
2300 * In some special case, if the task is fatal, fatal_signal_pending() or
2301 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2302 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2303 * as possible without any hazards. 2: all pages should have a valid
2304 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2305 * pointer, that is treated as a charge to root_mem_cgroup.
2306 *
2307 * So __mem_cgroup_try_charge() will return
2308 * 0 ... on success, filling *ptr with a valid memcg pointer.
2309 * -ENOMEM ... charge failure because of resource limits.
2310 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2311 *
2312 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2313 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002314 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002315static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002316 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002317 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002318 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002319 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002320{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002321 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002322 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002323 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002324 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002325
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002326 /*
2327 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2328 * in system level. So, allow to go ahead dying process in addition to
2329 * MEMDIE process.
2330 */
2331 if (unlikely(test_thread_flag(TIF_MEMDIE)
2332 || fatal_signal_pending(current)))
2333 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002334
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002335 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002336 * We always charge the cgroup the mm_struct belongs to.
2337 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002338 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002339 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002340 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002341 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002342 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002343again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002344 if (*ptr) { /* css should be a valid one */
2345 memcg = *ptr;
2346 VM_BUG_ON(css_is_removed(&memcg->css));
2347 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002348 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002349 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002350 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002351 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002352 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002353 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002354
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002355 rcu_read_lock();
2356 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002357 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002358 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002359 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002360 * race with swapoff. Then, we have small risk of mis-accouning.
2361 * But such kind of mis-account by race always happens because
2362 * we don't have cgroup_mutex(). It's overkill and we allo that
2363 * small race, here.
2364 * (*) swapoff at el will charge against mm-struct not against
2365 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002366 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002367 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002368 if (!memcg)
2369 memcg = root_mem_cgroup;
2370 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002371 rcu_read_unlock();
2372 goto done;
2373 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002374 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002375 /*
2376 * It seems dagerous to access memcg without css_get().
2377 * But considering how consume_stok works, it's not
2378 * necessary. If consume_stock success, some charges
2379 * from this memcg are cached on this cpu. So, we
2380 * don't need to call css_get()/css_tryget() before
2381 * calling consume_stock().
2382 */
2383 rcu_read_unlock();
2384 goto done;
2385 }
2386 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002387 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002388 rcu_read_unlock();
2389 goto again;
2390 }
2391 rcu_read_unlock();
2392 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002393
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002394 do {
2395 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002396
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002397 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002398 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002399 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002400 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002401 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002402
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002403 oom_check = false;
2404 if (oom && !nr_oom_retries) {
2405 oom_check = true;
2406 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2407 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002408
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002409 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002410 switch (ret) {
2411 case CHARGE_OK:
2412 break;
2413 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002414 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002415 css_put(&memcg->css);
2416 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002417 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002418 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002419 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002420 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002421 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002422 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002423 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002424 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002425 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002426 /* If oom, we never return -ENOMEM */
2427 nr_oom_retries--;
2428 break;
2429 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002430 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002431 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002432 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002433 } while (ret != CHARGE_OK);
2434
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002435 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002436 refill_stock(memcg, batch - nr_pages);
2437 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002438done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002439 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002440 return 0;
2441nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002442 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002443 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002444bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002445 *ptr = root_mem_cgroup;
2446 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002447}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002448
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002449/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002450 * Somemtimes we have to undo a charge we got by try_charge().
2451 * This function is for that and do uncharge, put css's refcnt.
2452 * gotten by try_charge().
2453 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002454static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002455 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002456{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002457 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002458 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002459
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002460 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002461 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002462 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002463 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002464}
2465
2466/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002467 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2468 * This is useful when moving usage to parent cgroup.
2469 */
2470static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2471 unsigned int nr_pages)
2472{
2473 unsigned long bytes = nr_pages * PAGE_SIZE;
2474
2475 if (mem_cgroup_is_root(memcg))
2476 return;
2477
2478 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2479 if (do_swap_account)
2480 res_counter_uncharge_until(&memcg->memsw,
2481 memcg->memsw.parent, bytes);
2482}
2483
2484/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002485 * A helper function to get mem_cgroup from ID. must be called under
2486 * rcu_read_lock(). The caller must check css_is_removed() or some if
2487 * it's concern. (dropping refcnt from swap can be called against removed
2488 * memcg.)
2489 */
2490static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2491{
2492 struct cgroup_subsys_state *css;
2493
2494 /* ID 0 is unused ID */
2495 if (!id)
2496 return NULL;
2497 css = css_lookup(&mem_cgroup_subsys, id);
2498 if (!css)
2499 return NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -07002500 return mem_cgroup_from_css(css);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002501}
2502
Wu Fengguange42d9d52009-12-16 12:19:59 +01002503struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002504{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002505 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002506 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002507 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002508 swp_entry_t ent;
2509
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002510 VM_BUG_ON(!PageLocked(page));
2511
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002512 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002513 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002514 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002515 memcg = pc->mem_cgroup;
2516 if (memcg && !css_tryget(&memcg->css))
2517 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002518 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002519 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002520 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002521 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002522 memcg = mem_cgroup_lookup(id);
2523 if (memcg && !css_tryget(&memcg->css))
2524 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002525 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002526 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002527 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002528 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002529}
2530
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002531static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002532 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002533 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002534 enum charge_type ctype,
2535 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002536{
Johannes Weinerce587e62012-04-24 20:22:33 +02002537 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002538 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002539 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002540 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002541 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002542
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002543 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002544 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002545 /*
2546 * we don't need page_cgroup_lock about tail pages, becase they are not
2547 * accessed by any other context at this point.
2548 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002549
2550 /*
2551 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2552 * may already be on some other mem_cgroup's LRU. Take care of it.
2553 */
2554 if (lrucare) {
2555 zone = page_zone(page);
2556 spin_lock_irq(&zone->lru_lock);
2557 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002558 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002559 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002560 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002561 was_on_lru = true;
2562 }
2563 }
2564
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002565 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002566 /*
2567 * We access a page_cgroup asynchronously without lock_page_cgroup().
2568 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2569 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2570 * before USED bit, we need memory barrier here.
2571 * See mem_cgroup_add_lru_list(), etc.
2572 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002573 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002574 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002575
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002576 if (lrucare) {
2577 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002578 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002579 VM_BUG_ON(PageLRU(page));
2580 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002581 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002582 }
2583 spin_unlock_irq(&zone->lru_lock);
2584 }
2585
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002586 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002587 anon = true;
2588 else
2589 anon = false;
2590
2591 mem_cgroup_charge_statistics(memcg, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002592 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002593
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002594 /*
2595 * "charge_statistics" updated event counter. Then, check it.
2596 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2597 * if they exceeds softlimit.
2598 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002599 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002600}
2601
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002602#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2603
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002604#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002605/*
2606 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002607 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2608 * charge/uncharge will be never happen and move_account() is done under
2609 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002610 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002611void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002612{
2613 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002614 struct page_cgroup *pc;
2615 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002616
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002617 if (mem_cgroup_disabled())
2618 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002619 for (i = 1; i < HPAGE_PMD_NR; i++) {
2620 pc = head_pc + i;
2621 pc->mem_cgroup = head_pc->mem_cgroup;
2622 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002623 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2624 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002625}
Hugh Dickins12d27102012-01-12 17:19:52 -08002626#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002627
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002628/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002629 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002630 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002631 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002632 * @pc: page_cgroup of the page.
2633 * @from: mem_cgroup which the page is moved from.
2634 * @to: mem_cgroup which the page is moved to. @from != @to.
2635 *
2636 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002637 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002638 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002639 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002640 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2641 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002642 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002643static int mem_cgroup_move_account(struct page *page,
2644 unsigned int nr_pages,
2645 struct page_cgroup *pc,
2646 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002647 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002648{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002649 unsigned long flags;
2650 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002651 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002652
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002653 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002654 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002655 /*
2656 * The page is isolated from LRU. So, collapse function
2657 * will not handle this page. But page splitting can happen.
2658 * Do this check under compound_page_lock(). The caller should
2659 * hold it.
2660 */
2661 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002662 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002663 goto out;
2664
2665 lock_page_cgroup(pc);
2666
2667 ret = -EINVAL;
2668 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2669 goto unlock;
2670
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002671 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002672
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002673 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002674 /* Update mapped_file data for mem_cgroup */
2675 preempt_disable();
2676 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2677 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2678 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002679 }
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002680 mem_cgroup_charge_statistics(from, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002681
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002682 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002683 pc->mem_cgroup = to;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002684 mem_cgroup_charge_statistics(to, anon, nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002685 /*
2686 * We charges against "to" which may not have any tasks. Then, "to"
2687 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002688 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002689 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002690 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002691 */
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002692 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002693 ret = 0;
2694unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002695 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002696 /*
2697 * check events
2698 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002699 memcg_check_events(to, page);
2700 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002701out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002702 return ret;
2703}
2704
2705/*
2706 * move charges to its parent.
2707 */
2708
Johannes Weiner5564e882011-03-23 16:42:29 -07002709static int mem_cgroup_move_parent(struct page *page,
2710 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07002711 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002712{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002713 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002714 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002715 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002716 int ret;
2717
2718 /* Is ROOT ? */
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002719 if (mem_cgroup_is_root(child))
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002720 return -EINVAL;
2721
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002722 ret = -EBUSY;
2723 if (!get_page_unless_zero(page))
2724 goto out;
2725 if (isolate_lru_page(page))
2726 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002727
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002728 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002729
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002730 parent = parent_mem_cgroup(child);
2731 /*
2732 * If no parent, move charges to root cgroup.
2733 */
2734 if (!parent)
2735 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002736
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002737 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002738 flags = compound_lock_irqsave(page);
2739
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002740 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002741 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002742 if (!ret)
2743 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002744
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002745 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002746 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002747 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002748put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002749 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002750out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002751 return ret;
2752}
2753
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002754/*
2755 * Charge the memory controller for page usage.
2756 * Return
2757 * 0 if the charge was successful
2758 * < 0 if the cgroup is over its limit
2759 */
2760static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002761 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002762{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002763 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002764 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002765 bool oom = true;
2766 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002767
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002768 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002769 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002770 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002771 /*
2772 * Never OOM-kill a process for a huge page. The
2773 * fault handler will fall back to regular pages.
2774 */
2775 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002776 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002777
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002778 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002779 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002780 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02002781 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002782 return 0;
2783}
2784
2785int mem_cgroup_newpage_charge(struct page *page,
2786 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002787{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002788 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002789 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002790 VM_BUG_ON(page_mapped(page));
2791 VM_BUG_ON(page->mapping && !PageAnon(page));
2792 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002793 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002794 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002795}
2796
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002797/*
2798 * While swap-in, try_charge -> commit or cancel, the page is locked.
2799 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002800 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002801 * "commit()" or removed by "cancel()"
2802 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07002803static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2804 struct page *page,
2805 gfp_t mask,
2806 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002807{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002808 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07002809 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002810 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002811
Johannes Weiner90deb782012-07-31 16:45:47 -07002812 pc = lookup_page_cgroup(page);
2813 /*
2814 * Every swap fault against a single page tries to charge the
2815 * page, bail as early as possible. shmem_unuse() encounters
2816 * already charged pages, too. The USED bit is protected by
2817 * the page lock, which serializes swap cache removal, which
2818 * in turn serializes uncharging.
2819 */
2820 if (PageCgroupUsed(pc))
2821 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002822 if (!do_swap_account)
2823 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002824 memcg = try_get_mem_cgroup_from_page(page);
2825 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002826 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002827 *memcgp = memcg;
2828 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002829 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002830 if (ret == -EINTR)
2831 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002832 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002833charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002834 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2835 if (ret == -EINTR)
2836 ret = 0;
2837 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002838}
2839
Johannes Weiner0435a2f2012-07-31 16:45:43 -07002840int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
2841 gfp_t gfp_mask, struct mem_cgroup **memcgp)
2842{
2843 *memcgp = NULL;
2844 if (mem_cgroup_disabled())
2845 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07002846 /*
2847 * A racing thread's fault, or swapoff, may have already
2848 * updated the pte, and even removed page from swap cache: in
2849 * those cases unuse_pte()'s pte_same() test will fail; but
2850 * there's also a KSM case which does need to charge the page.
2851 */
2852 if (!PageSwapCache(page)) {
2853 int ret;
2854
2855 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
2856 if (ret == -EINTR)
2857 ret = 0;
2858 return ret;
2859 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07002860 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
2861}
2862
Johannes Weiner827a03d2012-07-31 16:45:36 -07002863void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
2864{
2865 if (mem_cgroup_disabled())
2866 return;
2867 if (!memcg)
2868 return;
2869 __mem_cgroup_cancel_charge(memcg, 1);
2870}
2871
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002872static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002873__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002874 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002875{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002876 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002877 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002878 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002879 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002880 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002881
Johannes Weinerce587e62012-04-24 20:22:33 +02002882 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002883 /*
2884 * Now swap is on-memory. This means this page may be
2885 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002886 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2887 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2888 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002889 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002890 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002891 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07002892 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002893 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002894 /*
2895 * At swapin, we may charge account against cgroup which has no tasks.
2896 * So, rmdir()->pre_destroy() can be called while we do this charge.
2897 * In that case, we need to call pre_destroy() again. check it here.
2898 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002899 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002900}
2901
Johannes Weiner72835c82012-01-12 17:18:32 -08002902void mem_cgroup_commit_charge_swapin(struct page *page,
2903 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002904{
Johannes Weiner72835c82012-01-12 17:18:32 -08002905 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002906 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002907}
2908
Johannes Weiner827a03d2012-07-31 16:45:36 -07002909int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2910 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002911{
Johannes Weiner827a03d2012-07-31 16:45:36 -07002912 struct mem_cgroup *memcg = NULL;
2913 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
2914 int ret;
2915
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002916 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07002917 return 0;
2918 if (PageCompound(page))
2919 return 0;
2920
Johannes Weiner827a03d2012-07-31 16:45:36 -07002921 if (!PageSwapCache(page))
2922 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
2923 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07002924 ret = __mem_cgroup_try_charge_swapin(mm, page,
2925 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07002926 if (!ret)
2927 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2928 }
2929 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002930}
2931
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002932static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002933 unsigned int nr_pages,
2934 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002935{
2936 struct memcg_batch_info *batch = NULL;
2937 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002938
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002939 /* If swapout, usage of swap doesn't decrease */
2940 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2941 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002942
2943 batch = &current->memcg_batch;
2944 /*
2945 * In usual, we do css_get() when we remember memcg pointer.
2946 * But in this case, we keep res->usage until end of a series of
2947 * uncharges. Then, it's ok to ignore memcg's refcnt.
2948 */
2949 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002950 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002951 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002952 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002953 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002954 * the same cgroup and we have chance to coalesce uncharges.
2955 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2956 * because we want to do uncharge as soon as possible.
2957 */
2958
2959 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2960 goto direct_uncharge;
2961
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002962 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002963 goto direct_uncharge;
2964
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002965 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002966 * In typical case, batch->memcg == mem. This means we can
2967 * merge a series of uncharges to an uncharge of res_counter.
2968 * If not, we uncharge res_counter ony by one.
2969 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002970 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002971 goto direct_uncharge;
2972 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002973 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002974 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002975 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002976 return;
2977direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002978 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002979 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002980 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2981 if (unlikely(batch->memcg != memcg))
2982 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002983}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002984
Balbir Singh8697d332008-02-07 00:13:59 -08002985/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002986 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002987 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002988static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07002989__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
2990 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002991{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002992 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002993 unsigned int nr_pages = 1;
2994 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002995 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002996
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002997 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002998 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002999
Johannes Weiner0c59b892012-07-31 16:45:31 -07003000 VM_BUG_ON(PageSwapCache(page));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003001
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003002 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003003 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003004 VM_BUG_ON(!PageTransHuge(page));
3005 }
Balbir Singh8697d332008-02-07 00:13:59 -08003006 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08003007 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08003008 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003009 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003010 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003011 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003012
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003013 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003014
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003015 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003016
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003017 if (!PageCgroupUsed(pc))
3018 goto unlock_out;
3019
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003020 anon = PageAnon(page);
3021
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003022 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003023 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07003024 /*
3025 * Generally PageAnon tells if it's the anon statistics to be
3026 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3027 * used before page reached the stage of being marked PageAnon.
3028 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003029 anon = true;
3030 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003031 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003032 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07003033 if (page_mapped(page))
3034 goto unlock_out;
3035 /*
3036 * Pages under migration may not be uncharged. But
3037 * end_migration() /must/ be the one uncharging the
3038 * unused post-migration page and so it has to call
3039 * here with the migration bit still set. See the
3040 * res_counter handling below.
3041 */
3042 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003043 goto unlock_out;
3044 break;
3045 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3046 if (!PageAnon(page)) { /* Shared memory */
3047 if (page->mapping && !page_is_file_cache(page))
3048 goto unlock_out;
3049 } else if (page_mapped(page)) /* Anon */
3050 goto unlock_out;
3051 break;
3052 default:
3053 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003054 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003055
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003056 mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003057
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003058 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003059 /*
3060 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3061 * freed from LRU. This is safe because uncharged page is expected not
3062 * to be reused (freed soon). Exception is SwapCache, it's handled by
3063 * special functions.
3064 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003065
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003066 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003067 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003068 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003069 * will never be freed.
3070 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003071 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003072 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003073 mem_cgroup_swap_statistics(memcg, true);
3074 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003075 }
Johannes Weiner0030f532012-07-31 16:45:25 -07003076 /*
3077 * Migration does not charge the res_counter for the
3078 * replacement page, so leave it alone when phasing out the
3079 * page that is unused after the migration.
3080 */
3081 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003082 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003083
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003084 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003085
3086unlock_out:
3087 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003088 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003089}
3090
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003091void mem_cgroup_uncharge_page(struct page *page)
3092{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003093 /* early check. */
3094 if (page_mapped(page))
3095 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08003096 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner0c59b892012-07-31 16:45:31 -07003097 if (PageSwapCache(page))
3098 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07003099 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003100}
3101
3102void mem_cgroup_uncharge_cache_page(struct page *page)
3103{
3104 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003105 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07003106 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003107}
3108
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003109/*
3110 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3111 * In that cases, pages are freed continuously and we can expect pages
3112 * are in the same memcg. All these calls itself limits the number of
3113 * pages freed at once, then uncharge_start/end() is called properly.
3114 * This may be called prural(2) times in a context,
3115 */
3116
3117void mem_cgroup_uncharge_start(void)
3118{
3119 current->memcg_batch.do_batch++;
3120 /* We can do nest. */
3121 if (current->memcg_batch.do_batch == 1) {
3122 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003123 current->memcg_batch.nr_pages = 0;
3124 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003125 }
3126}
3127
3128void mem_cgroup_uncharge_end(void)
3129{
3130 struct memcg_batch_info *batch = &current->memcg_batch;
3131
3132 if (!batch->do_batch)
3133 return;
3134
3135 batch->do_batch--;
3136 if (batch->do_batch) /* If stacked, do nothing. */
3137 return;
3138
3139 if (!batch->memcg)
3140 return;
3141 /*
3142 * This "batch->memcg" is valid without any css_get/put etc...
3143 * bacause we hide charges behind us.
3144 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003145 if (batch->nr_pages)
3146 res_counter_uncharge(&batch->memcg->res,
3147 batch->nr_pages * PAGE_SIZE);
3148 if (batch->memsw_nr_pages)
3149 res_counter_uncharge(&batch->memcg->memsw,
3150 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003151 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003152 /* forget this pointer (for sanity check) */
3153 batch->memcg = NULL;
3154}
3155
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003156#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003157/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003158 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003159 * memcg information is recorded to swap_cgroup of "ent"
3160 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003161void
3162mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003163{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003164 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003165 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003166
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003167 if (!swapout) /* this was a swap cache but the swap is unused ! */
3168 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3169
Johannes Weiner0030f532012-07-31 16:45:25 -07003170 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003171
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003172 /*
3173 * record memcg information, if swapout && memcg != NULL,
3174 * mem_cgroup_get() was called in uncharge().
3175 */
3176 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003177 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003178}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003179#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003180
Andrew Mortonc255a452012-07-31 16:43:02 -07003181#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003182/*
3183 * called from swap_entry_free(). remove record in swap_cgroup and
3184 * uncharge "memsw" account.
3185 */
3186void mem_cgroup_uncharge_swap(swp_entry_t ent)
3187{
3188 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003189 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003190
3191 if (!do_swap_account)
3192 return;
3193
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003194 id = swap_cgroup_record(ent, 0);
3195 rcu_read_lock();
3196 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003197 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003198 /*
3199 * We uncharge this because swap is freed.
3200 * This memcg can be obsolete one. We avoid calling css_tryget
3201 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003202 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003203 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003204 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003205 mem_cgroup_put(memcg);
3206 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003207 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003208}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003209
3210/**
3211 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3212 * @entry: swap entry to be moved
3213 * @from: mem_cgroup which the entry is moved from
3214 * @to: mem_cgroup which the entry is moved to
3215 *
3216 * It succeeds only when the swap_cgroup's record for this entry is the same
3217 * as the mem_cgroup's id of @from.
3218 *
3219 * Returns 0 on success, -EINVAL on failure.
3220 *
3221 * The caller must have charged to @to, IOW, called res_counter_charge() about
3222 * both res and memsw, and called css_get().
3223 */
3224static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003225 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003226{
3227 unsigned short old_id, new_id;
3228
3229 old_id = css_id(&from->css);
3230 new_id = css_id(&to->css);
3231
3232 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003233 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003234 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003235 /*
3236 * This function is only called from task migration context now.
3237 * It postpones res_counter and refcount handling till the end
3238 * of task migration(mem_cgroup_clear_mc()) for performance
3239 * improvement. But we cannot postpone mem_cgroup_get(to)
3240 * because if the process that has been moved to @to does
3241 * swap-in, the refcount of @to might be decreased to 0.
3242 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003243 mem_cgroup_get(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003244 return 0;
3245 }
3246 return -EINVAL;
3247}
3248#else
3249static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003250 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003251{
3252 return -EINVAL;
3253}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003254#endif
3255
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003256/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003257 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3258 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003259 */
Johannes Weiner0030f532012-07-31 16:45:25 -07003260void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
3261 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003262{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003263 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003264 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003265 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003266
Johannes Weiner72835c82012-01-12 17:18:32 -08003267 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003268
Andrea Arcangeliec168512011-01-13 15:46:56 -08003269 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003270 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07003271 return;
Balbir Singh40779602008-04-04 14:29:59 -07003272
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003273 pc = lookup_page_cgroup(page);
3274 lock_page_cgroup(pc);
3275 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003276 memcg = pc->mem_cgroup;
3277 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003278 /*
3279 * At migrating an anonymous page, its mapcount goes down
3280 * to 0 and uncharge() will be called. But, even if it's fully
3281 * unmapped, migration may fail and this page has to be
3282 * charged again. We set MIGRATION flag here and delay uncharge
3283 * until end_migration() is called
3284 *
3285 * Corner Case Thinking
3286 * A)
3287 * When the old page was mapped as Anon and it's unmap-and-freed
3288 * while migration was ongoing.
3289 * If unmap finds the old page, uncharge() of it will be delayed
3290 * until end_migration(). If unmap finds a new page, it's
3291 * uncharged when it make mapcount to be 1->0. If unmap code
3292 * finds swap_migration_entry, the new page will not be mapped
3293 * and end_migration() will find it(mapcount==0).
3294 *
3295 * B)
3296 * When the old page was mapped but migraion fails, the kernel
3297 * remaps it. A charge for it is kept by MIGRATION flag even
3298 * if mapcount goes down to 0. We can do remap successfully
3299 * without charging it again.
3300 *
3301 * C)
3302 * The "old" page is under lock_page() until the end of
3303 * migration, so, the old page itself will not be swapped-out.
3304 * If the new page is swapped out before end_migraton, our
3305 * hook to usual swap-out path will catch the event.
3306 */
3307 if (PageAnon(page))
3308 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003309 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003310 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003311 /*
3312 * If the page is not charged at this point,
3313 * we return here.
3314 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003315 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07003316 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003317
Johannes Weiner72835c82012-01-12 17:18:32 -08003318 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003319 /*
3320 * We charge new page before it's used/mapped. So, even if unlock_page()
3321 * is called before end_migration, we can catch all events on this new
3322 * page. In the case new page is migrated but not remapped, new page's
3323 * mapcount will be finally 0 and we call uncharge in end_migration().
3324 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003325 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003326 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003327 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07003328 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07003329 /*
3330 * The page is committed to the memcg, but it's not actually
3331 * charged to the res_counter since we plan on replacing the
3332 * old one and only one page is going to be left afterwards.
3333 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003334 __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003335}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003336
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003337/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003338void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003339 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003340{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003341 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003342 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003343 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003344
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003345 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003346 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003347 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003348 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003349 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003350 used = oldpage;
3351 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003352 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003353 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003354 unused = oldpage;
3355 }
Johannes Weiner0030f532012-07-31 16:45:25 -07003356 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07003357 __mem_cgroup_uncharge_common(unused,
3358 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
3359 : MEM_CGROUP_CHARGE_TYPE_CACHE,
3360 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07003361 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003362 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003363 * We disallowed uncharge of pages under migration because mapcount
3364 * of the page goes down to zero, temporarly.
3365 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003366 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003367 pc = lookup_page_cgroup(oldpage);
3368 lock_page_cgroup(pc);
3369 ClearPageCgroupMigration(pc);
3370 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003371
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003372 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003373 * If a page is a file cache, radix-tree replacement is very atomic
3374 * and we can skip this check. When it was an Anon page, its mapcount
3375 * goes down to 0. But because we added MIGRATION flage, it's not
3376 * uncharged yet. There are several case but page->mapcount check
3377 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3378 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003379 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003380 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003381 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003382 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003383 * At migration, we may charge account against cgroup which has no
3384 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003385 * So, rmdir()->pre_destroy() can be called while we do this charge.
3386 * In that case, we need to call pre_destroy() again. check it here.
3387 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003388 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003389}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003390
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003391/*
3392 * At replace page cache, newpage is not under any memcg but it's on
3393 * LRU. So, this function doesn't touch res_counter but handles LRU
3394 * in correct way. Both pages are locked so we cannot race with uncharge.
3395 */
3396void mem_cgroup_replace_page_cache(struct page *oldpage,
3397 struct page *newpage)
3398{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003399 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003400 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003401 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003402
3403 if (mem_cgroup_disabled())
3404 return;
3405
3406 pc = lookup_page_cgroup(oldpage);
3407 /* fix accounting on old pages */
3408 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003409 if (PageCgroupUsed(pc)) {
3410 memcg = pc->mem_cgroup;
3411 mem_cgroup_charge_statistics(memcg, false, -1);
3412 ClearPageCgroupUsed(pc);
3413 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003414 unlock_page_cgroup(pc);
3415
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003416 /*
3417 * When called from shmem_replace_page(), in some cases the
3418 * oldpage has already been charged, and in some cases not.
3419 */
3420 if (!memcg)
3421 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003422 /*
3423 * Even if newpage->mapping was NULL before starting replacement,
3424 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3425 * LRU while we overwrite pc->mem_cgroup.
3426 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003427 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003428}
3429
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003430#ifdef CONFIG_DEBUG_VM
3431static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3432{
3433 struct page_cgroup *pc;
3434
3435 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003436 /*
3437 * Can be NULL while feeding pages into the page allocator for
3438 * the first time, i.e. during boot or memory hotplug;
3439 * or when mem_cgroup_disabled().
3440 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003441 if (likely(pc) && PageCgroupUsed(pc))
3442 return pc;
3443 return NULL;
3444}
3445
3446bool mem_cgroup_bad_page_check(struct page *page)
3447{
3448 if (mem_cgroup_disabled())
3449 return false;
3450
3451 return lookup_page_cgroup_used(page) != NULL;
3452}
3453
3454void mem_cgroup_print_bad_page(struct page *page)
3455{
3456 struct page_cgroup *pc;
3457
3458 pc = lookup_page_cgroup_used(page);
3459 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003460 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003461 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003462 }
3463}
3464#endif
3465
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003466static DEFINE_MUTEX(set_limit_mutex);
3467
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003468static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003469 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003470{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003471 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003472 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003473 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003474 int children = mem_cgroup_count_children(memcg);
3475 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003476 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003477
3478 /*
3479 * For keeping hierarchical_reclaim simple, how long we should retry
3480 * is depends on callers. We set our retry-count to be function
3481 * of # of children which we should visit in this loop.
3482 */
3483 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3484
3485 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003486
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003487 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003488 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003489 if (signal_pending(current)) {
3490 ret = -EINTR;
3491 break;
3492 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003493 /*
3494 * Rather than hide all in some function, I do this in
3495 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07003496 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003497 */
3498 mutex_lock(&set_limit_mutex);
3499 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3500 if (memswlimit < val) {
3501 ret = -EINVAL;
3502 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003503 break;
3504 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003505
3506 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3507 if (memlimit < val)
3508 enlarge = 1;
3509
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003510 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003511 if (!ret) {
3512 if (memswlimit == val)
3513 memcg->memsw_is_minimum = true;
3514 else
3515 memcg->memsw_is_minimum = false;
3516 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003517 mutex_unlock(&set_limit_mutex);
3518
3519 if (!ret)
3520 break;
3521
Johannes Weiner56600482012-01-12 17:17:59 -08003522 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3523 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003524 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3525 /* Usage is reduced ? */
3526 if (curusage >= oldusage)
3527 retry_count--;
3528 else
3529 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003530 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003531 if (!ret && enlarge)
3532 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003533
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003534 return ret;
3535}
3536
Li Zefan338c8432009-06-17 16:27:15 -07003537static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3538 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003539{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003540 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003541 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003542 int children = mem_cgroup_count_children(memcg);
3543 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003544 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003545
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003546 /* see mem_cgroup_resize_res_limit */
3547 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3548 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003549 while (retry_count) {
3550 if (signal_pending(current)) {
3551 ret = -EINTR;
3552 break;
3553 }
3554 /*
3555 * Rather than hide all in some function, I do this in
3556 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07003557 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003558 */
3559 mutex_lock(&set_limit_mutex);
3560 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3561 if (memlimit > val) {
3562 ret = -EINVAL;
3563 mutex_unlock(&set_limit_mutex);
3564 break;
3565 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003566 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3567 if (memswlimit < val)
3568 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003569 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003570 if (!ret) {
3571 if (memlimit == val)
3572 memcg->memsw_is_minimum = true;
3573 else
3574 memcg->memsw_is_minimum = false;
3575 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003576 mutex_unlock(&set_limit_mutex);
3577
3578 if (!ret)
3579 break;
3580
Johannes Weiner56600482012-01-12 17:17:59 -08003581 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3582 MEM_CGROUP_RECLAIM_NOSWAP |
3583 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003584 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003585 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003586 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003587 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003588 else
3589 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003590 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003591 if (!ret && enlarge)
3592 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003593 return ret;
3594}
3595
Balbir Singh4e416952009-09-23 15:56:39 -07003596unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003597 gfp_t gfp_mask,
3598 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003599{
3600 unsigned long nr_reclaimed = 0;
3601 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3602 unsigned long reclaimed;
3603 int loop = 0;
3604 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003605 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003606 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003607
3608 if (order > 0)
3609 return 0;
3610
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003611 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003612 /*
3613 * This loop can run a while, specially if mem_cgroup's continuously
3614 * keep exceeding their soft limit and putting the system under
3615 * pressure
3616 */
3617 do {
3618 if (next_mz)
3619 mz = next_mz;
3620 else
3621 mz = mem_cgroup_largest_soft_limit_node(mctz);
3622 if (!mz)
3623 break;
3624
Ying Han0ae5e892011-05-26 16:25:25 -07003625 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003626 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08003627 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003628 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003629 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003630 spin_lock(&mctz->lock);
3631
3632 /*
3633 * If we failed to reclaim anything from this memory cgroup
3634 * it is time to move on to the next cgroup
3635 */
3636 next_mz = NULL;
3637 if (!reclaimed) {
3638 do {
3639 /*
3640 * Loop until we find yet another one.
3641 *
3642 * By the time we get the soft_limit lock
3643 * again, someone might have aded the
3644 * group back on the RB tree. Iterate to
3645 * make sure we get a different mem.
3646 * mem_cgroup_largest_soft_limit_node returns
3647 * NULL if no other cgroup is present on
3648 * the tree
3649 */
3650 next_mz =
3651 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003652 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003653 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003654 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003655 break;
3656 } while (1);
3657 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003658 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3659 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003660 /*
3661 * One school of thought says that we should not add
3662 * back the node to the tree if reclaim returns 0.
3663 * But our reclaim could return 0, simply because due
3664 * to priority we are exposing a smaller subset of
3665 * memory to reclaim from. Consider this as a longer
3666 * term TODO.
3667 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003668 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003669 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003670 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003671 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003672 loop++;
3673 /*
3674 * Could not reclaim anything and there are no more
3675 * mem cgroups to try or we seem to be looping without
3676 * reclaiming anything.
3677 */
3678 if (!nr_reclaimed &&
3679 (next_mz == NULL ||
3680 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3681 break;
3682 } while (!nr_reclaimed);
3683 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003684 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003685 return nr_reclaimed;
3686}
3687
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003688/*
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003689 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3690 * reclaim the pages page themselves - it just removes the page_cgroups.
3691 * Returns true if some page_cgroups were not freed, indicating that the caller
3692 * must retry this operation.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003693 */
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003694static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003695 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003696{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003697 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003698 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003699 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003700 struct page *busy;
3701 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003702
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003703 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003704 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003705 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003706
Hugh Dickins1eb49272012-03-21 16:34:19 -07003707 loop = mz->lru_size[lru];
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003708 /* give some margin against EBUSY etc...*/
3709 loop += 256;
3710 busy = NULL;
3711 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003712 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003713 struct page *page;
3714
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003715 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003716 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003717 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003718 break;
3719 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003720 page = list_entry(list->prev, struct page, lru);
3721 if (busy == page) {
3722 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003723 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003724 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003725 continue;
3726 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003727 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003728
Johannes Weiner925b7672012-01-12 17:18:15 -08003729 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003730
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003731 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003732 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003733 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003734 cond_resched();
3735 } else
3736 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003737 }
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003738 return !list_empty(list);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003739}
3740
3741/*
3742 * make mem_cgroup's charge to be 0 if there is no task.
3743 * This enables deleting this mem_cgroup.
3744 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003745static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003746{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003747 int ret;
3748 int node, zid, shrink;
3749 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003750 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003751
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003752 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003753
3754 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003755 /* should free all ? */
3756 if (free_all)
3757 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003758move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003759 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003760 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003761 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003762 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003763 /* This is for making all *used* pages to be on LRU. */
3764 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003765 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003766 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003767 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003768 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003769 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003770 enum lru_list lru;
3771 for_each_lru(lru) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003772 ret = mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003773 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003774 if (ret)
3775 break;
3776 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003777 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003778 if (ret)
3779 break;
3780 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003781 mem_cgroup_end_move(memcg);
3782 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003783 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003784 /* "ret" should also be checked to ensure all lists are empty. */
Glauber Costa569530f2012-04-12 12:49:13 -07003785 } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003786out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003787 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003788 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003789
3790try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003791 /* returns EBUSY if there is a task or if we come here twice. */
3792 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003793 ret = -EBUSY;
3794 goto out;
3795 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003796 /* we call try-to-free pages for make this cgroup empty */
3797 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003798 /* try to free all pages in this cgroup */
3799 shrink = 1;
Glauber Costa569530f2012-04-12 12:49:13 -07003800 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003801 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003802
3803 if (signal_pending(current)) {
3804 ret = -EINTR;
3805 goto out;
3806 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003807 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003808 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003809 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003810 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003811 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003812 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003813 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003814
3815 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003816 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003817 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003818 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003819}
3820
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07003821static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003822{
3823 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3824}
3825
3826
Balbir Singh18f59ea2009-01-07 18:08:07 -08003827static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3828{
3829 return mem_cgroup_from_cont(cont)->use_hierarchy;
3830}
3831
3832static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3833 u64 val)
3834{
3835 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003836 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003837 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003838 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003839
3840 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003841 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003842
3843 cgroup_lock();
Glauber Costa567fb432012-07-31 16:43:07 -07003844
3845 if (memcg->use_hierarchy == val)
3846 goto out;
3847
Balbir Singh18f59ea2009-01-07 18:08:07 -08003848 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003849 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003850 * in the child subtrees. If it is unset, then the change can
3851 * occur, provided the current cgroup has no children.
3852 *
3853 * For the root cgroup, parent_mem is NULL, we allow value to be
3854 * set if there are no children.
3855 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003856 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003857 (val == 1 || val == 0)) {
3858 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003859 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003860 else
3861 retval = -EBUSY;
3862 } else
3863 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07003864
3865out:
Balbir Singh18f59ea2009-01-07 18:08:07 -08003866 cgroup_unlock();
3867
3868 return retval;
3869}
3870
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003871
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003872static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003873 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003874{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003875 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003876 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003877
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003878 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003879 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003880 val += mem_cgroup_read_stat(iter, idx);
3881
3882 if (val < 0) /* race ? */
3883 val = 0;
3884 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003885}
3886
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003887static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003888{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003889 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003890
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003891 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003892 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003893 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003894 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003895 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003896 }
3897
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003898 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3899 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003900
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003901 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003902 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003903
3904 return val << PAGE_SHIFT;
3905}
3906
Tejun Heoaf36f902012-04-01 12:09:55 -07003907static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
3908 struct file *file, char __user *buf,
3909 size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003910{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003911 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Tejun Heoaf36f902012-04-01 12:09:55 -07003912 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003913 u64 val;
Tejun Heoaf36f902012-04-01 12:09:55 -07003914 int type, name, len;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003915
3916 type = MEMFILE_TYPE(cft->private);
3917 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003918
3919 if (!do_swap_account && type == _MEMSWAP)
3920 return -EOPNOTSUPP;
3921
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003922 switch (type) {
3923 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003924 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003925 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003926 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003927 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003928 break;
3929 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003930 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003931 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003932 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003933 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003934 break;
3935 default:
3936 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003937 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003938
3939 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3940 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003941}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003942/*
3943 * The user of this function is...
3944 * RES_LIMIT.
3945 */
Paul Menage856c13a2008-07-25 01:47:04 -07003946static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3947 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003948{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003949 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003950 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003951 unsigned long long val;
3952 int ret;
3953
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003954 type = MEMFILE_TYPE(cft->private);
3955 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003956
3957 if (!do_swap_account && type == _MEMSWAP)
3958 return -EOPNOTSUPP;
3959
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003960 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003961 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003962 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3963 ret = -EINVAL;
3964 break;
3965 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003966 /* This function does all necessary parse...reuse it */
3967 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003968 if (ret)
3969 break;
3970 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003971 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003972 else
3973 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003974 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003975 case RES_SOFT_LIMIT:
3976 ret = res_counter_memparse_write_strategy(buffer, &val);
3977 if (ret)
3978 break;
3979 /*
3980 * For memsw, soft limits are hard to implement in terms
3981 * of semantics, for now, we support soft limits for
3982 * control without swap
3983 */
3984 if (type == _MEM)
3985 ret = res_counter_set_soft_limit(&memcg->res, val);
3986 else
3987 ret = -EINVAL;
3988 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003989 default:
3990 ret = -EINVAL; /* should be BUG() ? */
3991 break;
3992 }
3993 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003994}
3995
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003996static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3997 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3998{
3999 struct cgroup *cgroup;
4000 unsigned long long min_limit, min_memsw_limit, tmp;
4001
4002 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4003 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4004 cgroup = memcg->css.cgroup;
4005 if (!memcg->use_hierarchy)
4006 goto out;
4007
4008 while (cgroup->parent) {
4009 cgroup = cgroup->parent;
4010 memcg = mem_cgroup_from_cont(cgroup);
4011 if (!memcg->use_hierarchy)
4012 break;
4013 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4014 min_limit = min(min_limit, tmp);
4015 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4016 min_memsw_limit = min(min_memsw_limit, tmp);
4017 }
4018out:
4019 *mem_limit = min_limit;
4020 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004021}
4022
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004023static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004024{
Tejun Heoaf36f902012-04-01 12:09:55 -07004025 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004026 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004027
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004028 type = MEMFILE_TYPE(event);
4029 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07004030
4031 if (!do_swap_account && type == _MEMSWAP)
4032 return -EOPNOTSUPP;
4033
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004034 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004035 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004036 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004037 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004038 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004039 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004040 break;
4041 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004042 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004043 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004044 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004045 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004046 break;
4047 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004048
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07004049 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004050}
4051
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004052static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4053 struct cftype *cft)
4054{
4055 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4056}
4057
Daisuke Nishimura02491442010-03-10 15:22:17 -08004058#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004059static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4060 struct cftype *cft, u64 val)
4061{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004062 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004063
4064 if (val >= (1 << NR_MOVE_TYPE))
4065 return -EINVAL;
4066 /*
4067 * We check this value several times in both in can_attach() and
4068 * attach(), so we need cgroup lock to prevent this value from being
4069 * inconsistent.
4070 */
4071 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004072 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004073 cgroup_unlock();
4074
4075 return 0;
4076}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004077#else
4078static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4079 struct cftype *cft, u64 val)
4080{
4081 return -ENOSYS;
4082}
4083#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004084
Ying Han406eb0c2011-05-26 16:25:37 -07004085#ifdef CONFIG_NUMA
Wanpeng Liab215882012-07-31 16:43:09 -07004086static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weinerfada52c2012-05-29 15:07:06 -07004087 struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07004088{
4089 int nid;
4090 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4091 unsigned long node_nr;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004092 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07004093
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004094 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004095 seq_printf(m, "total=%lu", total_nr);
4096 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004097 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004098 seq_printf(m, " N%d=%lu", nid, node_nr);
4099 }
4100 seq_putc(m, '\n');
4101
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004102 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004103 seq_printf(m, "file=%lu", file_nr);
4104 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004105 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004106 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004107 seq_printf(m, " N%d=%lu", nid, node_nr);
4108 }
4109 seq_putc(m, '\n');
4110
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004111 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004112 seq_printf(m, "anon=%lu", anon_nr);
4113 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004114 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004115 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004116 seq_printf(m, " N%d=%lu", nid, node_nr);
4117 }
4118 seq_putc(m, '\n');
4119
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004120 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004121 seq_printf(m, "unevictable=%lu", unevictable_nr);
4122 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004123 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004124 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004125 seq_printf(m, " N%d=%lu", nid, node_nr);
4126 }
4127 seq_putc(m, '\n');
4128 return 0;
4129}
4130#endif /* CONFIG_NUMA */
4131
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004132static const char * const mem_cgroup_lru_names[] = {
4133 "inactive_anon",
4134 "active_anon",
4135 "inactive_file",
4136 "active_file",
4137 "unevictable",
4138};
4139
4140static inline void mem_cgroup_lru_names_not_uptodate(void)
4141{
4142 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4143}
4144
Wanpeng Liab215882012-07-31 16:43:09 -07004145static int memcg_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004146 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004147{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004148 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004149 struct mem_cgroup *mi;
4150 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004151
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004152 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004153 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004154 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004155 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4156 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004157 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004158
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004159 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4160 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4161 mem_cgroup_read_events(memcg, i));
4162
4163 for (i = 0; i < NR_LRU_LISTS; i++)
4164 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4165 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4166
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004167 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004168 {
4169 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004170 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004171 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004172 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004173 seq_printf(m, "hierarchical_memsw_limit %llu\n",
4174 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004175 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004176
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004177 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4178 long long val = 0;
4179
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004180 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004181 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004182 for_each_mem_cgroup_tree(mi, memcg)
4183 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4184 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4185 }
4186
4187 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4188 unsigned long long val = 0;
4189
4190 for_each_mem_cgroup_tree(mi, memcg)
4191 val += mem_cgroup_read_events(mi, i);
4192 seq_printf(m, "total_%s %llu\n",
4193 mem_cgroup_events_names[i], val);
4194 }
4195
4196 for (i = 0; i < NR_LRU_LISTS; i++) {
4197 unsigned long long val = 0;
4198
4199 for_each_mem_cgroup_tree(mi, memcg)
4200 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4201 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004202 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004203
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004204#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004205 {
4206 int nid, zid;
4207 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07004208 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004209 unsigned long recent_rotated[2] = {0, 0};
4210 unsigned long recent_scanned[2] = {0, 0};
4211
4212 for_each_online_node(nid)
4213 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004214 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07004215 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004216
Hugh Dickins89abfab2012-05-29 15:06:53 -07004217 recent_rotated[0] += rstat->recent_rotated[0];
4218 recent_rotated[1] += rstat->recent_rotated[1];
4219 recent_scanned[0] += rstat->recent_scanned[0];
4220 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004221 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004222 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4223 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4224 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4225 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004226 }
4227#endif
4228
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004229 return 0;
4230}
4231
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004232static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4233{
4234 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4235
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004236 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004237}
4238
4239static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4240 u64 val)
4241{
4242 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4243 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004244
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004245 if (val > 100)
4246 return -EINVAL;
4247
4248 if (cgrp->parent == NULL)
4249 return -EINVAL;
4250
4251 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004252
4253 cgroup_lock();
4254
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004255 /* If under hierarchy, only empty-root can set this value */
4256 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004257 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4258 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004259 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004260 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004261
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004262 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004263
Li Zefan068b38c2009-01-15 13:51:26 -08004264 cgroup_unlock();
4265
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004266 return 0;
4267}
4268
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004269static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4270{
4271 struct mem_cgroup_threshold_ary *t;
4272 u64 usage;
4273 int i;
4274
4275 rcu_read_lock();
4276 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004277 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004278 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004279 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004280
4281 if (!t)
4282 goto unlock;
4283
4284 usage = mem_cgroup_usage(memcg, swap);
4285
4286 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004287 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004288 * If it's not true, a threshold was crossed after last
4289 * call of __mem_cgroup_threshold().
4290 */
Phil Carmody5407a562010-05-26 14:42:42 -07004291 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004292
4293 /*
4294 * Iterate backward over array of thresholds starting from
4295 * current_threshold and check if a threshold is crossed.
4296 * If none of thresholds below usage is crossed, we read
4297 * only one element of the array here.
4298 */
4299 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4300 eventfd_signal(t->entries[i].eventfd, 1);
4301
4302 /* i = current_threshold + 1 */
4303 i++;
4304
4305 /*
4306 * Iterate forward over array of thresholds starting from
4307 * current_threshold+1 and check if a threshold is crossed.
4308 * If none of thresholds above usage is crossed, we read
4309 * only one element of the array here.
4310 */
4311 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4312 eventfd_signal(t->entries[i].eventfd, 1);
4313
4314 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004315 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004316unlock:
4317 rcu_read_unlock();
4318}
4319
4320static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4321{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004322 while (memcg) {
4323 __mem_cgroup_threshold(memcg, false);
4324 if (do_swap_account)
4325 __mem_cgroup_threshold(memcg, true);
4326
4327 memcg = parent_mem_cgroup(memcg);
4328 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004329}
4330
4331static int compare_thresholds(const void *a, const void *b)
4332{
4333 const struct mem_cgroup_threshold *_a = a;
4334 const struct mem_cgroup_threshold *_b = b;
4335
4336 return _a->threshold - _b->threshold;
4337}
4338
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004339static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004340{
4341 struct mem_cgroup_eventfd_list *ev;
4342
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004343 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004344 eventfd_signal(ev->eventfd, 1);
4345 return 0;
4346}
4347
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004348static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004349{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004350 struct mem_cgroup *iter;
4351
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004352 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004353 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004354}
4355
4356static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4357 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004358{
4359 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004360 struct mem_cgroup_thresholds *thresholds;
4361 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004362 int type = MEMFILE_TYPE(cft->private);
4363 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004364 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004365
4366 ret = res_counter_memparse_write_strategy(args, &threshold);
4367 if (ret)
4368 return ret;
4369
4370 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004371
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004372 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004373 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004374 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004375 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004376 else
4377 BUG();
4378
4379 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4380
4381 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004382 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004383 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4384
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004385 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004386
4387 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004388 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004389 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004390 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004391 ret = -ENOMEM;
4392 goto unlock;
4393 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004394 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004395
4396 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 if (thresholds->primary) {
4398 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004399 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004400 }
4401
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004402 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004403 new->entries[size - 1].eventfd = eventfd;
4404 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004405
4406 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004407 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004408 compare_thresholds, NULL);
4409
4410 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004411 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004412 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004413 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004414 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004415 * new->current_threshold will not be used until
4416 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004417 * it here.
4418 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004419 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004420 } else
4421 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004422 }
4423
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004424 /* Free old spare buffer and save old primary buffer as spare */
4425 kfree(thresholds->spare);
4426 thresholds->spare = thresholds->primary;
4427
4428 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004429
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004430 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004431 synchronize_rcu();
4432
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004433unlock:
4434 mutex_unlock(&memcg->thresholds_lock);
4435
4436 return ret;
4437}
4438
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004439static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004440 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004441{
4442 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004443 struct mem_cgroup_thresholds *thresholds;
4444 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004445 int type = MEMFILE_TYPE(cft->private);
4446 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004447 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004448
4449 mutex_lock(&memcg->thresholds_lock);
4450 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004451 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004452 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004453 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004454 else
4455 BUG();
4456
Anton Vorontsov371528c2012-02-24 05:14:46 +04004457 if (!thresholds->primary)
4458 goto unlock;
4459
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004460 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4461
4462 /* Check if a threshold crossed before removing */
4463 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4464
4465 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004466 size = 0;
4467 for (i = 0; i < thresholds->primary->size; i++) {
4468 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004469 size++;
4470 }
4471
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004472 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004473
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004474 /* Set thresholds array to NULL if we don't have thresholds */
4475 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004476 kfree(new);
4477 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004478 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004479 }
4480
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004481 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004482
4483 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004484 new->current_threshold = -1;
4485 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4486 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004487 continue;
4488
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004489 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004490 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004491 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004492 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004493 * until rcu_assign_pointer(), so it's safe to increment
4494 * it here.
4495 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004496 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004497 }
4498 j++;
4499 }
4500
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004501swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004502 /* Swap primary and spare array */
4503 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004504 /* If all events are unregistered, free the spare array */
4505 if (!new) {
4506 kfree(thresholds->spare);
4507 thresholds->spare = NULL;
4508 }
4509
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004510 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004512 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004513 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004514unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004515 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004516}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004517
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004518static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4519 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4520{
4521 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4522 struct mem_cgroup_eventfd_list *event;
4523 int type = MEMFILE_TYPE(cft->private);
4524
4525 BUG_ON(type != _OOM_TYPE);
4526 event = kmalloc(sizeof(*event), GFP_KERNEL);
4527 if (!event)
4528 return -ENOMEM;
4529
Michal Hocko1af8efe2011-07-26 16:08:24 -07004530 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004531
4532 event->eventfd = eventfd;
4533 list_add(&event->list, &memcg->oom_notify);
4534
4535 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004536 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004537 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004538 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004539
4540 return 0;
4541}
4542
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004543static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004544 struct cftype *cft, struct eventfd_ctx *eventfd)
4545{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004546 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004547 struct mem_cgroup_eventfd_list *ev, *tmp;
4548 int type = MEMFILE_TYPE(cft->private);
4549
4550 BUG_ON(type != _OOM_TYPE);
4551
Michal Hocko1af8efe2011-07-26 16:08:24 -07004552 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004553
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004554 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004555 if (ev->eventfd == eventfd) {
4556 list_del(&ev->list);
4557 kfree(ev);
4558 }
4559 }
4560
Michal Hocko1af8efe2011-07-26 16:08:24 -07004561 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004562}
4563
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004564static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4565 struct cftype *cft, struct cgroup_map_cb *cb)
4566{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004567 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004568
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004569 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004570
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004571 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004572 cb->fill(cb, "under_oom", 1);
4573 else
4574 cb->fill(cb, "under_oom", 0);
4575 return 0;
4576}
4577
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004578static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4579 struct cftype *cft, u64 val)
4580{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004581 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004582 struct mem_cgroup *parent;
4583
4584 /* cannot set to root cgroup and only 0 and 1 are allowed */
4585 if (!cgrp->parent || !((val == 0) || (val == 1)))
4586 return -EINVAL;
4587
4588 parent = mem_cgroup_from_cont(cgrp->parent);
4589
4590 cgroup_lock();
4591 /* oom-kill-disable is a flag for subhierarchy. */
4592 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004593 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004594 cgroup_unlock();
4595 return -EINVAL;
4596 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004597 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004598 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004599 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004600 cgroup_unlock();
4601 return 0;
4602}
4603
Andrew Mortonc255a452012-07-31 16:43:02 -07004604#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03004605static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004606{
Glauber Costa1d62e432012-04-09 19:36:33 -03004607 return mem_cgroup_sockets_init(memcg, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004608};
4609
Glauber Costa1d62e432012-04-09 19:36:33 -03004610static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004611{
Glauber Costa1d62e432012-04-09 19:36:33 -03004612 mem_cgroup_sockets_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004613}
Glauber Costae5671df2011-12-11 21:47:01 +00004614#else
Glauber Costacbe128e32012-04-09 19:36:34 -03004615static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004616{
4617 return 0;
4618}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004619
Glauber Costa1d62e432012-04-09 19:36:33 -03004620static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004621{
4622}
Glauber Costae5671df2011-12-11 21:47:01 +00004623#endif
4624
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004625static struct cftype mem_cgroup_files[] = {
4626 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004627 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004628 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004629 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004630 .register_event = mem_cgroup_usage_register_event,
4631 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004632 },
4633 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004634 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004635 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004636 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004637 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004638 },
4639 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004640 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004641 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004642 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004643 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004644 },
4645 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004646 .name = "soft_limit_in_bytes",
4647 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4648 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004649 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07004650 },
4651 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004652 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004653 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004654 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004655 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004656 },
Balbir Singh8697d332008-02-07 00:13:59 -08004657 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004658 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07004659 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004660 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004661 {
4662 .name = "force_empty",
4663 .trigger = mem_cgroup_force_empty_write,
4664 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004665 {
4666 .name = "use_hierarchy",
4667 .write_u64 = mem_cgroup_hierarchy_write,
4668 .read_u64 = mem_cgroup_hierarchy_read,
4669 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004670 {
4671 .name = "swappiness",
4672 .read_u64 = mem_cgroup_swappiness_read,
4673 .write_u64 = mem_cgroup_swappiness_write,
4674 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004675 {
4676 .name = "move_charge_at_immigrate",
4677 .read_u64 = mem_cgroup_move_charge_read,
4678 .write_u64 = mem_cgroup_move_charge_write,
4679 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004680 {
4681 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004682 .read_map = mem_cgroup_oom_control_read,
4683 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004684 .register_event = mem_cgroup_oom_register_event,
4685 .unregister_event = mem_cgroup_oom_unregister_event,
4686 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4687 },
Ying Han406eb0c2011-05-26 16:25:37 -07004688#ifdef CONFIG_NUMA
4689 {
4690 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07004691 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004692 },
4693#endif
Andrew Mortonc255a452012-07-31 16:43:02 -07004694#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004695 {
4696 .name = "memsw.usage_in_bytes",
4697 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004698 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004699 .register_event = mem_cgroup_usage_register_event,
4700 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004701 },
4702 {
4703 .name = "memsw.max_usage_in_bytes",
4704 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4705 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004706 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004707 },
4708 {
4709 .name = "memsw.limit_in_bytes",
4710 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4711 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004712 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004713 },
4714 {
4715 .name = "memsw.failcnt",
4716 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4717 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004718 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004719 },
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004720#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004721 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004722};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004723
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004724static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004725{
4726 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004727 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004728 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004729 /*
4730 * This routine is called against possible nodes.
4731 * But it's BUG to call kmalloc() against offline node.
4732 *
4733 * TODO: this routine can waste much memory for nodes which will
4734 * never be onlined. It's better to use memory hotplug callback
4735 * function.
4736 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004737 if (!node_state(node, N_NORMAL_MEMORY))
4738 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004739 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004740 if (!pn)
4741 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004742
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004743 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4744 mz = &pn->zoneinfo[zone];
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -07004745 lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004746 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004747 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004748 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004749 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004750 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004751 return 0;
4752}
4753
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004754static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004755{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004756 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004757}
4758
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004759static struct mem_cgroup *mem_cgroup_alloc(void)
4760{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004761 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004762 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004763
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004764 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004765 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004766 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004767 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004768 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004769
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004770 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004771 return NULL;
4772
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004773 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4774 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004775 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004776 spin_lock_init(&memcg->pcp_counter_lock);
4777 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004778
4779out_free:
4780 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004781 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004782 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004783 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004784 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004785}
4786
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004787/*
Glauber Costa3afe36b2012-05-29 15:07:10 -07004788 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
Hugh Dickins59927fb2012-03-15 15:17:07 -07004789 * but in process context. The work_freeing structure is overlaid
4790 * on the rcu_freeing structure, which itself is overlaid on memsw.
4791 */
Glauber Costa3afe36b2012-05-29 15:07:10 -07004792static void free_work(struct work_struct *work)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004793{
4794 struct mem_cgroup *memcg;
Glauber Costa3afe36b2012-05-29 15:07:10 -07004795 int size = sizeof(struct mem_cgroup);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004796
4797 memcg = container_of(work, struct mem_cgroup, work_freeing);
Glauber Costa3f134612012-05-29 15:07:11 -07004798 /*
4799 * We need to make sure that (at least for now), the jump label
4800 * destruction code runs outside of the cgroup lock. This is because
4801 * get_online_cpus(), which is called from the static_branch update,
4802 * can't be called inside the cgroup_lock. cpusets are the ones
4803 * enforcing this dependency, so if they ever change, we might as well.
4804 *
4805 * schedule_work() will guarantee this happens. Be careful if you need
4806 * to move this code around, and make sure it is outside
4807 * the cgroup_lock.
4808 */
4809 disarm_sock_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004810 if (size < PAGE_SIZE)
4811 kfree(memcg);
4812 else
4813 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004814}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004815
4816static void free_rcu(struct rcu_head *rcu_head)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004817{
4818 struct mem_cgroup *memcg;
4819
4820 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004821 INIT_WORK(&memcg->work_freeing, free_work);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004822 schedule_work(&memcg->work_freeing);
4823}
4824
4825/*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004826 * At destroying mem_cgroup, references from swap_cgroup can remain.
4827 * (scanning all at force_empty is too costly...)
4828 *
4829 * Instead of clearing all references at force_empty, we remember
4830 * the number of reference from swap_cgroup and free mem_cgroup when
4831 * it goes down to 0.
4832 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004833 * Removal of cgroup itself succeeds regardless of refs from swap.
4834 */
4835
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004836static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004837{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004838 int node;
4839
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004840 mem_cgroup_remove_from_trees(memcg);
4841 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004842
Bob Liu3ed28fa2012-01-12 17:19:04 -08004843 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004844 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004845
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004846 free_percpu(memcg->stat);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004847 call_rcu(&memcg->rcu_freeing, free_rcu);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004848}
4849
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004850static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004851{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004852 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004853}
4854
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004855static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004856{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004857 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4858 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4859 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004860 if (parent)
4861 mem_cgroup_put(parent);
4862 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004863}
4864
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004865static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004866{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004867 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004868}
4869
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004870/*
4871 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4872 */
Glauber Costae1aab162011-12-11 21:47:03 +00004873struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004874{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004875 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004876 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004877 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004878}
Glauber Costae1aab162011-12-11 21:47:03 +00004879EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004880
Andrew Mortonc255a452012-07-31 16:43:02 -07004881#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004882static void __init enable_swap_cgroup(void)
4883{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004884 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004885 do_swap_account = 1;
4886}
4887#else
4888static void __init enable_swap_cgroup(void)
4889{
4890}
4891#endif
4892
Balbir Singhf64c3f52009-09-23 15:56:37 -07004893static int mem_cgroup_soft_limit_tree_init(void)
4894{
4895 struct mem_cgroup_tree_per_node *rtpn;
4896 struct mem_cgroup_tree_per_zone *rtpz;
4897 int tmp, node, zone;
4898
Bob Liu3ed28fa2012-01-12 17:19:04 -08004899 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004900 tmp = node;
4901 if (!node_state(node, N_NORMAL_MEMORY))
4902 tmp = -1;
4903 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4904 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004905 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004906
4907 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4908
4909 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4910 rtpz = &rtpn->rb_tree_per_zone[zone];
4911 rtpz->rb_root = RB_ROOT;
4912 spin_lock_init(&rtpz->lock);
4913 }
4914 }
4915 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004916
4917err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004918 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004919 if (!soft_limit_tree.rb_tree_per_node[node])
4920 break;
4921 kfree(soft_limit_tree.rb_tree_per_node[node]);
4922 soft_limit_tree.rb_tree_per_node[node] = NULL;
4923 }
4924 return 1;
4925
Balbir Singhf64c3f52009-09-23 15:56:37 -07004926}
4927
Li Zefan0eb253e2009-01-15 13:51:25 -08004928static struct cgroup_subsys_state * __ref
Li Zefan761b3ef2012-01-31 13:47:36 +08004929mem_cgroup_create(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004930{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004931 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004932 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004933 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004934
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004935 memcg = mem_cgroup_alloc();
4936 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004937 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004938
Bob Liu3ed28fa2012-01-12 17:19:04 -08004939 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004940 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004941 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004942
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004943 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004944 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004945 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004946 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004947 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004948 if (mem_cgroup_soft_limit_tree_init())
4949 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004950 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004951 for_each_possible_cpu(cpu) {
4952 struct memcg_stock_pcp *stock =
4953 &per_cpu(memcg_stock, cpu);
4954 INIT_WORK(&stock->work, drain_local_stock);
4955 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004956 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004957 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004958 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004959 memcg->use_hierarchy = parent->use_hierarchy;
4960 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004961 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004962
Balbir Singh18f59ea2009-01-07 18:08:07 -08004963 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004964 res_counter_init(&memcg->res, &parent->res);
4965 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004966 /*
4967 * We increment refcnt of the parent to ensure that we can
4968 * safely access it on res_counter_charge/uncharge.
4969 * This refcnt will be decremented when freeing this
4970 * mem_cgroup(see mem_cgroup_put).
4971 */
4972 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004973 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004974 res_counter_init(&memcg->res, NULL);
4975 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004976 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004977 memcg->last_scanned_node = MAX_NUMNODES;
4978 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08004979
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004980 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004981 memcg->swappiness = mem_cgroup_swappiness(parent);
4982 atomic_set(&memcg->refcnt, 1);
4983 memcg->move_charge_at_immigrate = 0;
4984 mutex_init(&memcg->thresholds_lock);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07004985 spin_lock_init(&memcg->move_lock);
Glauber Costacbe128e32012-04-09 19:36:34 -03004986
4987 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
4988 if (error) {
4989 /*
4990 * We call put now because our (and parent's) refcnts
4991 * are already in place. mem_cgroup_put() will internally
4992 * call __mem_cgroup_free, so return directly
4993 */
4994 mem_cgroup_put(memcg);
4995 return ERR_PTR(error);
4996 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004997 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004998free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004999 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005000 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005001}
5002
Li Zefan761b3ef2012-01-31 13:47:36 +08005003static int mem_cgroup_pre_destroy(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005004{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005005 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005006
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005007 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005008}
5009
Li Zefan761b3ef2012-01-31 13:47:36 +08005010static void mem_cgroup_destroy(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005011{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005012 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005013
Glauber Costa1d62e432012-04-09 19:36:33 -03005014 kmem_cgroup_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005015
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005016 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005017}
5018
Daisuke Nishimura02491442010-03-10 15:22:17 -08005019#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005020/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005021#define PRECHARGE_COUNT_AT_ONCE 256
5022static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005023{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005024 int ret = 0;
5025 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005026 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005027
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005028 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005029 mc.precharge += count;
5030 /* we don't need css_get for root */
5031 return ret;
5032 }
5033 /* try to charge at once */
5034 if (count > 1) {
5035 struct res_counter *dummy;
5036 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005037 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005038 * by cgroup_lock_live_cgroup() that it is not removed and we
5039 * are still under the same cgroup_mutex. So we can postpone
5040 * css_get().
5041 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005042 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005043 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005044 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005045 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005046 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005047 goto one_by_one;
5048 }
5049 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005050 return ret;
5051 }
5052one_by_one:
5053 /* fall back to one by one charge */
5054 while (count--) {
5055 if (signal_pending(current)) {
5056 ret = -EINTR;
5057 break;
5058 }
5059 if (!batch_count--) {
5060 batch_count = PRECHARGE_COUNT_AT_ONCE;
5061 cond_resched();
5062 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005063 ret = __mem_cgroup_try_charge(NULL,
5064 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005065 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005066 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005067 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005068 mc.precharge++;
5069 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005070 return ret;
5071}
5072
5073/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005074 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005075 * @vma: the vma the pte to be checked belongs
5076 * @addr: the address corresponding to the pte to be checked
5077 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005078 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005079 *
5080 * Returns
5081 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5082 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5083 * move charge. if @target is not NULL, the page is stored in target->page
5084 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005085 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5086 * target for charge migration. if @target is not NULL, the entry is stored
5087 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005088 *
5089 * Called with pte lock held.
5090 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005091union mc_target {
5092 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005093 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005094};
5095
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005096enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005097 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005098 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005099 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005100};
5101
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005102static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5103 unsigned long addr, pte_t ptent)
5104{
5105 struct page *page = vm_normal_page(vma, addr, ptent);
5106
5107 if (!page || !page_mapped(page))
5108 return NULL;
5109 if (PageAnon(page)) {
5110 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005111 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005112 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005113 } else if (!move_file())
5114 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005115 return NULL;
5116 if (!get_page_unless_zero(page))
5117 return NULL;
5118
5119 return page;
5120}
5121
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005122#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005123static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5124 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5125{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005126 struct page *page = NULL;
5127 swp_entry_t ent = pte_to_swp_entry(ptent);
5128
5129 if (!move_anon() || non_swap_entry(ent))
5130 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005131 /*
5132 * Because lookup_swap_cache() updates some statistics counter,
5133 * we call find_get_page() with swapper_space directly.
5134 */
5135 page = find_get_page(&swapper_space, ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005136 if (do_swap_account)
5137 entry->val = ent.val;
5138
5139 return page;
5140}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005141#else
5142static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5143 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5144{
5145 return NULL;
5146}
5147#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005148
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005149static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5150 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5151{
5152 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005153 struct address_space *mapping;
5154 pgoff_t pgoff;
5155
5156 if (!vma->vm_file) /* anonymous vma */
5157 return NULL;
5158 if (!move_file())
5159 return NULL;
5160
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005161 mapping = vma->vm_file->f_mapping;
5162 if (pte_none(ptent))
5163 pgoff = linear_page_index(vma, addr);
5164 else /* pte_file(ptent) is true */
5165 pgoff = pte_to_pgoff(ptent);
5166
5167 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005168 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005169
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005170#ifdef CONFIG_SWAP
5171 /* shmem/tmpfs may report page out on swap: account for that too. */
5172 if (radix_tree_exceptional_entry(page)) {
5173 swp_entry_t swap = radix_to_swp_entry(page);
5174 if (do_swap_account)
5175 *entry = swap;
5176 page = find_get_page(&swapper_space, swap.val);
5177 }
5178#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005179 return page;
5180}
5181
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005182static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005183 unsigned long addr, pte_t ptent, union mc_target *target)
5184{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005185 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005186 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005187 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005188 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005189
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005190 if (pte_present(ptent))
5191 page = mc_handle_present_pte(vma, addr, ptent);
5192 else if (is_swap_pte(ptent))
5193 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005194 else if (pte_none(ptent) || pte_file(ptent))
5195 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005196
5197 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005198 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005199 if (page) {
5200 pc = lookup_page_cgroup(page);
5201 /*
5202 * Do only loose check w/o page_cgroup lock.
5203 * mem_cgroup_move_account() checks the pc is valid or not under
5204 * the lock.
5205 */
5206 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5207 ret = MC_TARGET_PAGE;
5208 if (target)
5209 target->page = page;
5210 }
5211 if (!ret || !target)
5212 put_page(page);
5213 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005214 /* There is a swap entry and a page doesn't exist or isn't charged */
5215 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005216 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005217 ret = MC_TARGET_SWAP;
5218 if (target)
5219 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005220 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005221 return ret;
5222}
5223
Naoya Horiguchi12724852012-03-21 16:34:28 -07005224#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5225/*
5226 * We don't consider swapping or file mapped pages because THP does not
5227 * support them for now.
5228 * Caller should make sure that pmd_trans_huge(pmd) is true.
5229 */
5230static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5231 unsigned long addr, pmd_t pmd, union mc_target *target)
5232{
5233 struct page *page = NULL;
5234 struct page_cgroup *pc;
5235 enum mc_target_type ret = MC_TARGET_NONE;
5236
5237 page = pmd_page(pmd);
5238 VM_BUG_ON(!page || !PageHead(page));
5239 if (!move_anon())
5240 return ret;
5241 pc = lookup_page_cgroup(page);
5242 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5243 ret = MC_TARGET_PAGE;
5244 if (target) {
5245 get_page(page);
5246 target->page = page;
5247 }
5248 }
5249 return ret;
5250}
5251#else
5252static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5253 unsigned long addr, pmd_t pmd, union mc_target *target)
5254{
5255 return MC_TARGET_NONE;
5256}
5257#endif
5258
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005259static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5260 unsigned long addr, unsigned long end,
5261 struct mm_walk *walk)
5262{
5263 struct vm_area_struct *vma = walk->private;
5264 pte_t *pte;
5265 spinlock_t *ptl;
5266
Naoya Horiguchi12724852012-03-21 16:34:28 -07005267 if (pmd_trans_huge_lock(pmd, vma) == 1) {
5268 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5269 mc.precharge += HPAGE_PMD_NR;
5270 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005271 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005272 }
Dave Hansen03319322011-03-22 16:32:56 -07005273
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005274 if (pmd_trans_unstable(pmd))
5275 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005276 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5277 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005278 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005279 mc.precharge++; /* increment precharge temporarily */
5280 pte_unmap_unlock(pte - 1, ptl);
5281 cond_resched();
5282
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005283 return 0;
5284}
5285
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005286static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5287{
5288 unsigned long precharge;
5289 struct vm_area_struct *vma;
5290
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005291 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005292 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5293 struct mm_walk mem_cgroup_count_precharge_walk = {
5294 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5295 .mm = mm,
5296 .private = vma,
5297 };
5298 if (is_vm_hugetlb_page(vma))
5299 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005300 walk_page_range(vma->vm_start, vma->vm_end,
5301 &mem_cgroup_count_precharge_walk);
5302 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005303 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005304
5305 precharge = mc.precharge;
5306 mc.precharge = 0;
5307
5308 return precharge;
5309}
5310
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005311static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5312{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005313 unsigned long precharge = mem_cgroup_count_precharge(mm);
5314
5315 VM_BUG_ON(mc.moving_task);
5316 mc.moving_task = current;
5317 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005318}
5319
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005320/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5321static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005322{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005323 struct mem_cgroup *from = mc.from;
5324 struct mem_cgroup *to = mc.to;
5325
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005326 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005327 if (mc.precharge) {
5328 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5329 mc.precharge = 0;
5330 }
5331 /*
5332 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5333 * we must uncharge here.
5334 */
5335 if (mc.moved_charge) {
5336 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5337 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005338 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005339 /* we must fixup refcnts and charges */
5340 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005341 /* uncharge swap account from the old cgroup */
5342 if (!mem_cgroup_is_root(mc.from))
5343 res_counter_uncharge(&mc.from->memsw,
5344 PAGE_SIZE * mc.moved_swap);
5345 __mem_cgroup_put(mc.from, mc.moved_swap);
5346
5347 if (!mem_cgroup_is_root(mc.to)) {
5348 /*
5349 * we charged both to->res and to->memsw, so we should
5350 * uncharge to->res.
5351 */
5352 res_counter_uncharge(&mc.to->res,
5353 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005354 }
5355 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005356 mc.moved_swap = 0;
5357 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005358 memcg_oom_recover(from);
5359 memcg_oom_recover(to);
5360 wake_up_all(&mc.waitq);
5361}
5362
5363static void mem_cgroup_clear_mc(void)
5364{
5365 struct mem_cgroup *from = mc.from;
5366
5367 /*
5368 * we must clear moving_task before waking up waiters at the end of
5369 * task migration.
5370 */
5371 mc.moving_task = NULL;
5372 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005373 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005374 mc.from = NULL;
5375 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005376 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005377 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005378}
5379
Li Zefan761b3ef2012-01-31 13:47:36 +08005380static int mem_cgroup_can_attach(struct cgroup *cgroup,
5381 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005382{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005383 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005384 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005385 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005386
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005387 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005388 struct mm_struct *mm;
5389 struct mem_cgroup *from = mem_cgroup_from_task(p);
5390
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005391 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005392
5393 mm = get_task_mm(p);
5394 if (!mm)
5395 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005396 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005397 if (mm->owner == p) {
5398 VM_BUG_ON(mc.from);
5399 VM_BUG_ON(mc.to);
5400 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005401 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005402 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005403 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005404 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005405 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005406 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005407 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005408 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005409
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005410 ret = mem_cgroup_precharge_mc(mm);
5411 if (ret)
5412 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005413 }
5414 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005415 }
5416 return ret;
5417}
5418
Li Zefan761b3ef2012-01-31 13:47:36 +08005419static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5420 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005421{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005422 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005423}
5424
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005425static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5426 unsigned long addr, unsigned long end,
5427 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005428{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005429 int ret = 0;
5430 struct vm_area_struct *vma = walk->private;
5431 pte_t *pte;
5432 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005433 enum mc_target_type target_type;
5434 union mc_target target;
5435 struct page *page;
5436 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005437
Naoya Horiguchi12724852012-03-21 16:34:28 -07005438 /*
5439 * We don't take compound_lock() here but no race with splitting thp
5440 * happens because:
5441 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5442 * under splitting, which means there's no concurrent thp split,
5443 * - if another thread runs into split_huge_page() just after we
5444 * entered this if-block, the thread must wait for page table lock
5445 * to be unlocked in __split_huge_page_splitting(), where the main
5446 * part of thp split is not executed yet.
5447 */
5448 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005449 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005450 spin_unlock(&vma->vm_mm->page_table_lock);
5451 return 0;
5452 }
5453 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5454 if (target_type == MC_TARGET_PAGE) {
5455 page = target.page;
5456 if (!isolate_lru_page(page)) {
5457 pc = lookup_page_cgroup(page);
5458 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07005459 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005460 mc.precharge -= HPAGE_PMD_NR;
5461 mc.moved_charge += HPAGE_PMD_NR;
5462 }
5463 putback_lru_page(page);
5464 }
5465 put_page(page);
5466 }
5467 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005468 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005469 }
5470
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005471 if (pmd_trans_unstable(pmd))
5472 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005473retry:
5474 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5475 for (; addr != end; addr += PAGE_SIZE) {
5476 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005477 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005478
5479 if (!mc.precharge)
5480 break;
5481
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005482 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005483 case MC_TARGET_PAGE:
5484 page = target.page;
5485 if (isolate_lru_page(page))
5486 goto put;
5487 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005488 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07005489 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005490 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005491 /* we uncharge from mc.from later. */
5492 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005493 }
5494 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005495put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005496 put_page(page);
5497 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005498 case MC_TARGET_SWAP:
5499 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07005500 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005501 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005502 /* we fixup refcnts and charges later. */
5503 mc.moved_swap++;
5504 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005505 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005506 default:
5507 break;
5508 }
5509 }
5510 pte_unmap_unlock(pte - 1, ptl);
5511 cond_resched();
5512
5513 if (addr != end) {
5514 /*
5515 * We have consumed all precharges we got in can_attach().
5516 * We try charge one by one, but don't do any additional
5517 * charges to mc.to if we have failed in charge once in attach()
5518 * phase.
5519 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005520 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005521 if (!ret)
5522 goto retry;
5523 }
5524
5525 return ret;
5526}
5527
5528static void mem_cgroup_move_charge(struct mm_struct *mm)
5529{
5530 struct vm_area_struct *vma;
5531
5532 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005533retry:
5534 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5535 /*
5536 * Someone who are holding the mmap_sem might be waiting in
5537 * waitq. So we cancel all extra charges, wake up all waiters,
5538 * and retry. Because we cancel precharges, we might not be able
5539 * to move enough charges, but moving charge is a best-effort
5540 * feature anyway, so it wouldn't be a big problem.
5541 */
5542 __mem_cgroup_clear_mc();
5543 cond_resched();
5544 goto retry;
5545 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005546 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5547 int ret;
5548 struct mm_walk mem_cgroup_move_charge_walk = {
5549 .pmd_entry = mem_cgroup_move_charge_pte_range,
5550 .mm = mm,
5551 .private = vma,
5552 };
5553 if (is_vm_hugetlb_page(vma))
5554 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005555 ret = walk_page_range(vma->vm_start, vma->vm_end,
5556 &mem_cgroup_move_charge_walk);
5557 if (ret)
5558 /*
5559 * means we have consumed all precharges and failed in
5560 * doing additional charge. Just abandon here.
5561 */
5562 break;
5563 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005564 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005565}
5566
Li Zefan761b3ef2012-01-31 13:47:36 +08005567static void mem_cgroup_move_task(struct cgroup *cont,
5568 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005569{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005570 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005571 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005572
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005573 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005574 if (mc.to)
5575 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005576 mmput(mm);
5577 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005578 if (mc.to)
5579 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005580}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005581#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08005582static int mem_cgroup_can_attach(struct cgroup *cgroup,
5583 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005584{
5585 return 0;
5586}
Li Zefan761b3ef2012-01-31 13:47:36 +08005587static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5588 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005589{
5590}
Li Zefan761b3ef2012-01-31 13:47:36 +08005591static void mem_cgroup_move_task(struct cgroup *cont,
5592 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005593{
5594}
5595#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005596
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005597struct cgroup_subsys mem_cgroup_subsys = {
5598 .name = "memory",
5599 .subsys_id = mem_cgroup_subsys_id,
5600 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005601 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005602 .destroy = mem_cgroup_destroy,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005603 .can_attach = mem_cgroup_can_attach,
5604 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005605 .attach = mem_cgroup_move_task,
Tejun Heo6bc10342012-04-01 12:09:55 -07005606 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005607 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005608 .use_id = 1,
Tejun Heo48ddbe12012-04-01 12:09:56 -07005609 .__DEPRECATED_clear_css_refs = true,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005610};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005611
Andrew Mortonc255a452012-07-31 16:43:02 -07005612#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005613static int __init enable_swap_account(char *s)
5614{
5615 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005616 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005617 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005618 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005619 really_do_swap_account = 0;
5620 return 1;
5621}
Michal Hockoa2c89902011-05-24 17:12:50 -07005622__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005623
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005624#endif