blob: f72b5e52451a7d8e62648dbfe211e5ff11c7c34f [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
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070065/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080066int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080067
68/* for remember boot option*/
69#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70static int really_do_swap_account __initdata = 1;
71#else
72static int really_do_swap_account __initdata = 0;
73#endif
74
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075#else
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 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070090 MEM_CGROUP_STAT_SWAPOUT, /* # 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,
381 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700382 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700383 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800384 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700385 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700386 NR_CHARGE_TYPE,
387};
388
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800389/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000390#define _MEM (0)
391#define _MEMSWAP (1)
392#define _OOM_TYPE (2)
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700393#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
394#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800395#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700396/* Used for OOM nofiier */
397#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800398
Balbir Singh75822b42009-09-23 15:56:38 -0700399/*
400 * Reclaim flags for mem_cgroup_hierarchical_reclaim
401 */
402#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
403#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
404#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
405#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
406
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700407static void mem_cgroup_get(struct mem_cgroup *memcg);
408static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000409
410/* Writing them here to avoid exposing memcg's inner layout */
411#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae1aab162011-12-11 21:47:03 +0000412#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000413#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000414
415static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
416void sock_update_memcg(struct sock *sk)
417{
Glauber Costa376be5f2012-01-20 04:57:14 +0000418 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000419 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700420 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000421
422 BUG_ON(!sk->sk_prot->proto_cgroup);
423
Glauber Costaf3f511e2012-01-05 20:16:39 +0000424 /* Socket cloning can throw us here with sk_cgrp already
425 * filled. It won't however, necessarily happen from
426 * process context. So the test for root memcg given
427 * the current task's memcg won't help us in this case.
428 *
429 * Respecting the original socket's memcg is a better
430 * decision in this case.
431 */
432 if (sk->sk_cgrp) {
433 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
434 mem_cgroup_get(sk->sk_cgrp->memcg);
435 return;
436 }
437
Glauber Costae1aab162011-12-11 21:47:03 +0000438 rcu_read_lock();
439 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700440 cg_proto = sk->sk_prot->proto_cgroup(memcg);
441 if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
Glauber Costae1aab162011-12-11 21:47:03 +0000442 mem_cgroup_get(memcg);
Glauber Costa3f134612012-05-29 15:07:11 -0700443 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000444 }
445 rcu_read_unlock();
446 }
447}
448EXPORT_SYMBOL(sock_update_memcg);
449
450void sock_release_memcg(struct sock *sk)
451{
Glauber Costa376be5f2012-01-20 04:57:14 +0000452 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000453 struct mem_cgroup *memcg;
454 WARN_ON(!sk->sk_cgrp->memcg);
455 memcg = sk->sk_cgrp->memcg;
456 mem_cgroup_put(memcg);
457 }
458}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000459
Glauber Costa319d3b92012-01-15 22:04:39 +0000460#ifdef CONFIG_INET
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000461struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
462{
463 if (!memcg || mem_cgroup_is_root(memcg))
464 return NULL;
465
466 return &memcg->tcp_mem.cg_proto;
467}
468EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000469#endif /* CONFIG_INET */
470#endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
471
Glauber Costa3f134612012-05-29 15:07:11 -0700472#if defined(CONFIG_INET) && defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM)
473static void disarm_sock_keys(struct mem_cgroup *memcg)
474{
475 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
476 return;
477 static_key_slow_dec(&memcg_socket_limit_enabled);
478}
479#else
480static void disarm_sock_keys(struct mem_cgroup *memcg)
481{
482}
483#endif
484
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700485static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800486
Balbir Singhf64c3f52009-09-23 15:56:37 -0700487static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700488mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700489{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700490 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700491}
492
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700493struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100494{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700495 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100496}
497
Balbir Singhf64c3f52009-09-23 15:56:37 -0700498static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700499page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700500{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700501 int nid = page_to_nid(page);
502 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700503
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700504 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700505}
506
507static struct mem_cgroup_tree_per_zone *
508soft_limit_tree_node_zone(int nid, int zid)
509{
510 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
511}
512
513static struct mem_cgroup_tree_per_zone *
514soft_limit_tree_from_page(struct page *page)
515{
516 int nid = page_to_nid(page);
517 int zid = page_zonenum(page);
518
519 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
520}
521
522static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700523__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700524 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700525 struct mem_cgroup_tree_per_zone *mctz,
526 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700527{
528 struct rb_node **p = &mctz->rb_root.rb_node;
529 struct rb_node *parent = NULL;
530 struct mem_cgroup_per_zone *mz_node;
531
532 if (mz->on_tree)
533 return;
534
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700535 mz->usage_in_excess = new_usage_in_excess;
536 if (!mz->usage_in_excess)
537 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700538 while (*p) {
539 parent = *p;
540 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
541 tree_node);
542 if (mz->usage_in_excess < mz_node->usage_in_excess)
543 p = &(*p)->rb_left;
544 /*
545 * We can't avoid mem cgroups that are over their soft
546 * limit by the same amount
547 */
548 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
549 p = &(*p)->rb_right;
550 }
551 rb_link_node(&mz->tree_node, parent, p);
552 rb_insert_color(&mz->tree_node, &mctz->rb_root);
553 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700554}
555
556static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700557__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700558 struct mem_cgroup_per_zone *mz,
559 struct mem_cgroup_tree_per_zone *mctz)
560{
561 if (!mz->on_tree)
562 return;
563 rb_erase(&mz->tree_node, &mctz->rb_root);
564 mz->on_tree = false;
565}
566
567static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700568mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700569 struct mem_cgroup_per_zone *mz,
570 struct mem_cgroup_tree_per_zone *mctz)
571{
572 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700573 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700574 spin_unlock(&mctz->lock);
575}
576
Balbir Singhf64c3f52009-09-23 15:56:37 -0700577
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700578static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700579{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700580 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700581 struct mem_cgroup_per_zone *mz;
582 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700583 int nid = page_to_nid(page);
584 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700585 mctz = soft_limit_tree_from_page(page);
586
587 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700588 * Necessary to update all ancestors when hierarchy is used.
589 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700590 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700591 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
592 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
593 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700594 /*
595 * We have to update the tree if mz is on RB-tree or
596 * mem is over its softlimit.
597 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700598 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700599 spin_lock(&mctz->lock);
600 /* if on-tree, remove it */
601 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700602 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700603 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700604 * Insert again. mz->usage_in_excess will be updated.
605 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700606 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700607 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700608 spin_unlock(&mctz->lock);
609 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700610 }
611}
612
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700613static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700614{
615 int node, zone;
616 struct mem_cgroup_per_zone *mz;
617 struct mem_cgroup_tree_per_zone *mctz;
618
Bob Liu3ed28fa2012-01-12 17:19:04 -0800619 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700620 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700621 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700622 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700623 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700624 }
625 }
626}
627
Balbir Singh4e416952009-09-23 15:56:39 -0700628static struct mem_cgroup_per_zone *
629__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
630{
631 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700632 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700633
634retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700635 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700636 rightmost = rb_last(&mctz->rb_root);
637 if (!rightmost)
638 goto done; /* Nothing to reclaim from */
639
640 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
641 /*
642 * Remove the node now but someone else can add it back,
643 * we will to add it back at the end of reclaim to its correct
644 * position in the tree.
645 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700646 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
647 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
648 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700649 goto retry;
650done:
651 return mz;
652}
653
654static struct mem_cgroup_per_zone *
655mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
656{
657 struct mem_cgroup_per_zone *mz;
658
659 spin_lock(&mctz->lock);
660 mz = __mem_cgroup_largest_soft_limit_node(mctz);
661 spin_unlock(&mctz->lock);
662 return mz;
663}
664
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700665/*
666 * Implementation Note: reading percpu statistics for memcg.
667 *
668 * Both of vmstat[] and percpu_counter has threshold and do periodic
669 * synchronization to implement "quick" read. There are trade-off between
670 * reading cost and precision of value. Then, we may have a chance to implement
671 * a periodic synchronizion of counter in memcg's counter.
672 *
673 * But this _read() function is used for user interface now. The user accounts
674 * memory usage by memory cgroup and he _always_ requires exact value because
675 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
676 * have to visit all online cpus and make sum. So, for now, unnecessary
677 * synchronization is not implemented. (just implemented for cpu hotplug)
678 *
679 * If there are kernel internal actions which can make use of some not-exact
680 * value, and reading all cpu value can be performance bottleneck in some
681 * common workload, threashold and synchonization as vmstat[] should be
682 * implemented.
683 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700684static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700685 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800686{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700687 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800688 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800689
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700690 get_online_cpus();
691 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700692 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700693#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700694 spin_lock(&memcg->pcp_counter_lock);
695 val += memcg->nocpu_base.count[idx];
696 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700697#endif
698 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800699 return val;
700}
701
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700702static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700703 bool charge)
704{
705 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700706 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700707}
708
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700709static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700710 enum mem_cgroup_events_index idx)
711{
712 unsigned long val = 0;
713 int cpu;
714
715 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700716 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700717#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700718 spin_lock(&memcg->pcp_counter_lock);
719 val += memcg->nocpu_base.events[idx];
720 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700721#endif
722 return val;
723}
724
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700725static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700726 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800727{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800728 preempt_disable();
729
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700730 /*
731 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
732 * counted as CACHE even if it's on ANON LRU.
733 */
734 if (anon)
735 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700736 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800737 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700738 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700739 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700740
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800741 /* pagein of a big page is an event. So, ignore page size */
742 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800744 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700745 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800746 nr_pages = -nr_pages; /* for event */
747 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800748
Johannes Weiner13114712012-05-29 15:07:07 -0700749 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800750
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800751 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800752}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800753
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700754unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700755mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700756{
757 struct mem_cgroup_per_zone *mz;
758
759 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
760 return mz->lru_size[lru];
761}
762
763static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700765 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700766{
767 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700768 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700769 unsigned long ret = 0;
770
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700771 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700772
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700773 for_each_lru(lru) {
774 if (BIT(lru) & lru_mask)
775 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700776 }
777 return ret;
778}
779
780static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700781mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700782 int nid, unsigned int lru_mask)
783{
Ying Han889976d2011-05-26 16:25:33 -0700784 u64 total = 0;
785 int zid;
786
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700787 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700788 total += mem_cgroup_zone_nr_lru_pages(memcg,
789 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700790
Ying Han889976d2011-05-26 16:25:33 -0700791 return total;
792}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700793
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700794static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700795 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800796{
Ying Han889976d2011-05-26 16:25:33 -0700797 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800798 u64 total = 0;
799
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700800 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700801 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800802 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800803}
804
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800805static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
806 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800807{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700808 unsigned long val, next;
809
Johannes Weiner13114712012-05-29 15:07:07 -0700810 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700811 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700812 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800813 if ((long)next - (long)val < 0) {
814 switch (target) {
815 case MEM_CGROUP_TARGET_THRESH:
816 next = val + THRESHOLDS_EVENTS_TARGET;
817 break;
818 case MEM_CGROUP_TARGET_SOFTLIMIT:
819 next = val + SOFTLIMIT_EVENTS_TARGET;
820 break;
821 case MEM_CGROUP_TARGET_NUMAINFO:
822 next = val + NUMAINFO_EVENTS_TARGET;
823 break;
824 default:
825 break;
826 }
827 __this_cpu_write(memcg->stat->targets[target], next);
828 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700829 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800830 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800831}
832
833/*
834 * Check events in order.
835 *
836 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700837static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800838{
Steven Rostedt47994012011-11-02 13:38:33 -0700839 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800840 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800841 if (unlikely(mem_cgroup_event_ratelimit(memcg,
842 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800843 bool do_softlimit;
844 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800845
846 do_softlimit = mem_cgroup_event_ratelimit(memcg,
847 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700848#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800849 do_numainfo = mem_cgroup_event_ratelimit(memcg,
850 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700851#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800852 preempt_enable();
853
854 mem_cgroup_threshold(memcg);
855 if (unlikely(do_softlimit))
856 mem_cgroup_update_tree(memcg, page);
857#if MAX_NUMNODES > 1
858 if (unlikely(do_numainfo))
859 atomic_inc(&memcg->numainfo_events);
860#endif
861 } else
862 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800863}
864
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000865struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800866{
867 return container_of(cgroup_subsys_state(cont,
868 mem_cgroup_subsys_id), struct mem_cgroup,
869 css);
870}
871
Balbir Singhcf475ad2008-04-29 01:00:16 -0700872struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800873{
Balbir Singh31a78f22008-09-28 23:09:31 +0100874 /*
875 * mm_update_next_owner() may clear mm->owner to NULL
876 * if it races with swapoff, page migration, etc.
877 * So this can be called with p == NULL.
878 */
879 if (unlikely(!p))
880 return NULL;
881
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800882 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
883 struct mem_cgroup, css);
884}
885
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700886struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800887{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700888 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700889
890 if (!mm)
891 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800892 /*
893 * Because we have no locks, mm->owner's may be being moved to other
894 * cgroup. We use css_tryget() here even if this looks
895 * pessimistic (rather than adding locks here).
896 */
897 rcu_read_lock();
898 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700899 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
900 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800901 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700902 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800903 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700904 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800905}
906
Johannes Weiner56600482012-01-12 17:17:59 -0800907/**
908 * mem_cgroup_iter - iterate over memory cgroup hierarchy
909 * @root: hierarchy root
910 * @prev: previously returned memcg, NULL on first invocation
911 * @reclaim: cookie for shared reclaim walks, NULL for full walks
912 *
913 * Returns references to children of the hierarchy below @root, or
914 * @root itself, or %NULL after a full round-trip.
915 *
916 * Caller must pass the return value in @prev on subsequent
917 * invocations for reference counting, or use mem_cgroup_iter_break()
918 * to cancel a hierarchy walk before the round-trip is complete.
919 *
920 * Reclaimers can specify a zone and a priority level in @reclaim to
921 * divide up the memcgs in the hierarchy among all concurrent
922 * reclaimers operating on the same zone and priority.
923 */
924struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
925 struct mem_cgroup *prev,
926 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700927{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800928 struct mem_cgroup *memcg = NULL;
929 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700930
Johannes Weiner56600482012-01-12 17:17:59 -0800931 if (mem_cgroup_disabled())
932 return NULL;
933
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700934 if (!root)
935 root = root_mem_cgroup;
936
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800937 if (prev && !reclaim)
938 id = css_id(&prev->css);
939
940 if (prev && prev != root)
941 css_put(&prev->css);
942
943 if (!root->use_hierarchy && root != root_mem_cgroup) {
944 if (prev)
945 return NULL;
946 return root;
947 }
948
949 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800950 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800951 struct cgroup_subsys_state *css;
952
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800953 if (reclaim) {
954 int nid = zone_to_nid(reclaim->zone);
955 int zid = zone_idx(reclaim->zone);
956 struct mem_cgroup_per_zone *mz;
957
958 mz = mem_cgroup_zoneinfo(root, nid, zid);
959 iter = &mz->reclaim_iter[reclaim->priority];
960 if (prev && reclaim->generation != iter->generation)
961 return NULL;
962 id = iter->position;
963 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800964
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700965 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800966 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
967 if (css) {
968 if (css == &root->css || css_tryget(css))
969 memcg = container_of(css,
970 struct mem_cgroup, css);
971 } else
972 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700973 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700974
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800975 if (reclaim) {
976 iter->position = id;
977 if (!css)
978 iter->generation++;
979 else if (!prev && memcg)
980 reclaim->generation = iter->generation;
981 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800982
983 if (prev && !css)
984 return NULL;
985 }
986 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700987}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800988
Johannes Weiner56600482012-01-12 17:17:59 -0800989/**
990 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
991 * @root: hierarchy root
992 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
993 */
994void mem_cgroup_iter_break(struct mem_cgroup *root,
995 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800996{
997 if (!root)
998 root = root_mem_cgroup;
999 if (prev && prev != root)
1000 css_put(&prev->css);
1001}
1002
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001003/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001004 * Iteration constructs for visiting all cgroups (under a tree). If
1005 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1006 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001007 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001008#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001009 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001010 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001011 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001012
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001013#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001014 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001015 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001016 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001017
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001018static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -07001019{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001020 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001021}
1022
Ying Han456f9982011-05-26 16:25:38 -07001023void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1024{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001025 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001026
1027 if (!mm)
1028 return;
1029
1030 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001031 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1032 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001033 goto out;
1034
1035 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001036 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001037 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1038 break;
1039 case PGMAJFAULT:
1040 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001041 break;
1042 default:
1043 BUG();
1044 }
1045out:
1046 rcu_read_unlock();
1047}
1048EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1049
Johannes Weiner925b7672012-01-12 17:18:15 -08001050/**
1051 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1052 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001053 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001054 *
1055 * Returns the lru list vector holding pages for the given @zone and
1056 * @mem. This can be the global zone lruvec, if the memory controller
1057 * is disabled.
1058 */
1059struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1060 struct mem_cgroup *memcg)
1061{
1062 struct mem_cgroup_per_zone *mz;
1063
1064 if (mem_cgroup_disabled())
1065 return &zone->lruvec;
1066
1067 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1068 return &mz->lruvec;
1069}
1070
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001071/*
1072 * Following LRU functions are allowed to be used without PCG_LOCK.
1073 * Operations are called by routine of global LRU independently from memcg.
1074 * What we have to take care of here is validness of pc->mem_cgroup.
1075 *
1076 * Changes to pc->mem_cgroup happens when
1077 * 1. charge
1078 * 2. moving account
1079 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1080 * It is added to LRU before charge.
1081 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1082 * When moving account, the page is not on LRU. It's isolated.
1083 */
1084
Johannes Weiner925b7672012-01-12 17:18:15 -08001085/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001086 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001087 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001088 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001089 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001090struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001091{
1092 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001093 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001094 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001095
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001096 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001097 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001098
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001099 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001100 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001101
1102 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001103 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001104 * an uncharged page off lru does nothing to secure
1105 * its former mem_cgroup from sudden removal.
1106 *
1107 * Our caller holds lru_lock, and PageCgroupUsed is updated
1108 * under page_cgroup lock: between them, they make all uses
1109 * of pc->mem_cgroup safe.
1110 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001111 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001112 pc->mem_cgroup = memcg = root_mem_cgroup;
1113
Johannes Weiner925b7672012-01-12 17:18:15 -08001114 mz = page_cgroup_zoneinfo(memcg, page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001115 return &mz->lruvec;
1116}
1117
1118/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001119 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1120 * @lruvec: mem_cgroup per zone lru vector
1121 * @lru: index of lru list the page is sitting on
1122 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001123 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001124 * This function must be called when a page is added to or removed from an
1125 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001126 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001127void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1128 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001129{
1130 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001131 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001132
1133 if (mem_cgroup_disabled())
1134 return;
1135
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001136 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1137 lru_size = mz->lru_size + lru;
1138 *lru_size += nr_pages;
1139 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001140}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001141
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001142/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001143 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001144 * hierarchy subtree
1145 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001146bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1147 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001148{
Johannes Weiner91c637342012-05-29 15:06:24 -07001149 if (root_memcg == memcg)
1150 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001151 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001152 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001153 return css_is_ancestor(&memcg->css, &root_memcg->css);
1154}
1155
1156static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1157 struct mem_cgroup *memcg)
1158{
1159 bool ret;
1160
Johannes Weiner91c637342012-05-29 15:06:24 -07001161 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001162 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001163 rcu_read_unlock();
1164 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001165}
1166
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001167int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001168{
1169 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001170 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001171 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001172
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001173 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001174 if (p) {
1175 curr = try_get_mem_cgroup_from_mm(p->mm);
1176 task_unlock(p);
1177 } else {
1178 /*
1179 * All threads may have already detached their mm's, but the oom
1180 * killer still needs to detect if they have already been oom
1181 * killed to prevent needlessly killing additional tasks.
1182 */
1183 task_lock(task);
1184 curr = mem_cgroup_from_task(task);
1185 if (curr)
1186 css_get(&curr->css);
1187 task_unlock(task);
1188 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001189 if (!curr)
1190 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001191 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001192 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001193 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001194 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1195 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001196 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001197 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001198 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001199 return ret;
1200}
1201
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001202int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001203{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001204 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001205 unsigned long inactive;
1206 unsigned long active;
1207 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001208
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001209 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1210 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001211
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001212 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1213 if (gb)
1214 inactive_ratio = int_sqrt(10 * gb);
1215 else
1216 inactive_ratio = 1;
1217
Johannes Weiner9b272972011-11-02 13:38:23 -07001218 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001219}
1220
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001221int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001222{
1223 unsigned long active;
1224 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001225
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001226 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE);
1227 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE);
Rik van Riel56e49d22009-06-16 15:32:28 -07001228
1229 return (active > inactive);
1230}
1231
Balbir Singh6d61ef42009-01-07 18:08:06 -08001232#define mem_cgroup_from_res_counter(counter, member) \
1233 container_of(counter, struct mem_cgroup, member)
1234
Johannes Weiner19942822011-02-01 15:52:43 -08001235/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001236 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001237 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001238 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001239 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001240 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001241 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001242static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001243{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001244 unsigned long long margin;
1245
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001246 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001247 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001248 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001249 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001250}
1251
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001252int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001253{
1254 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001255
1256 /* root ? */
1257 if (cgrp->parent == NULL)
1258 return vm_swappiness;
1259
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001260 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001261}
1262
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001263/*
1264 * memcg->moving_account is used for checking possibility that some thread is
1265 * calling move_account(). When a thread on CPU-A starts moving pages under
1266 * a memcg, other threads should check memcg->moving_account under
1267 * rcu_read_lock(), like this:
1268 *
1269 * CPU-A CPU-B
1270 * rcu_read_lock()
1271 * memcg->moving_account+1 if (memcg->mocing_account)
1272 * take heavy locks.
1273 * synchronize_rcu() update something.
1274 * rcu_read_unlock()
1275 * start move here.
1276 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001277
1278/* for quick checking without looking up memcg */
1279atomic_t memcg_moving __read_mostly;
1280
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001281static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001282{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001283 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001284 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001285 synchronize_rcu();
1286}
1287
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001289{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001290 /*
1291 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1292 * We check NULL in callee rather than caller.
1293 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001294 if (memcg) {
1295 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001296 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001297 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001298}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001299
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001300/*
1301 * 2 routines for checking "mem" is under move_account() or not.
1302 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001303 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1304 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001305 * pc->mem_cgroup may be overwritten.
1306 *
1307 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1308 * under hierarchy of moving cgroups. This is for
1309 * waiting at hith-memory prressure caused by "move".
1310 */
1311
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001312static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001313{
1314 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001315 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001316}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001317
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001318static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001319{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001320 struct mem_cgroup *from;
1321 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001322 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001323 /*
1324 * Unlike task_move routines, we access mc.to, mc.from not under
1325 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1326 */
1327 spin_lock(&mc.lock);
1328 from = mc.from;
1329 to = mc.to;
1330 if (!from)
1331 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001332
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001333 ret = mem_cgroup_same_or_subtree(memcg, from)
1334 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001335unlock:
1336 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001337 return ret;
1338}
1339
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001340static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001341{
1342 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001343 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001344 DEFINE_WAIT(wait);
1345 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1346 /* moving charge context might have finished. */
1347 if (mc.moving_task)
1348 schedule();
1349 finish_wait(&mc.waitq, &wait);
1350 return true;
1351 }
1352 }
1353 return false;
1354}
1355
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001356/*
1357 * Take this lock when
1358 * - a code tries to modify page's memcg while it's USED.
1359 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001360 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001361 */
1362static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1363 unsigned long *flags)
1364{
1365 spin_lock_irqsave(&memcg->move_lock, *flags);
1366}
1367
1368static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1369 unsigned long *flags)
1370{
1371 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1372}
1373
Balbir Singhe2224322009-04-02 16:57:39 -07001374/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001375 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001376 * @memcg: The memory cgroup that went over limit
1377 * @p: Task that is going to be killed
1378 *
1379 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1380 * enabled
1381 */
1382void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1383{
1384 struct cgroup *task_cgrp;
1385 struct cgroup *mem_cgrp;
1386 /*
1387 * Need a buffer in BSS, can't rely on allocations. The code relies
1388 * on the assumption that OOM is serialized for memory controller.
1389 * If this assumption is broken, revisit this code.
1390 */
1391 static char memcg_name[PATH_MAX];
1392 int ret;
1393
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001394 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001395 return;
1396
Balbir Singhe2224322009-04-02 16:57:39 -07001397 rcu_read_lock();
1398
1399 mem_cgrp = memcg->css.cgroup;
1400 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1401
1402 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1403 if (ret < 0) {
1404 /*
1405 * Unfortunately, we are unable to convert to a useful name
1406 * But we'll still print out the usage information
1407 */
1408 rcu_read_unlock();
1409 goto done;
1410 }
1411 rcu_read_unlock();
1412
1413 printk(KERN_INFO "Task in %s killed", memcg_name);
1414
1415 rcu_read_lock();
1416 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1417 if (ret < 0) {
1418 rcu_read_unlock();
1419 goto done;
1420 }
1421 rcu_read_unlock();
1422
1423 /*
1424 * Continues from above, so we don't need an KERN_ level
1425 */
1426 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1427done:
1428
1429 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1430 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1431 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1432 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1433 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1434 "failcnt %llu\n",
1435 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1436 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1437 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1438}
1439
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001440/*
1441 * This function returns the number of memcg under hierarchy tree. Returns
1442 * 1(self count) if no children.
1443 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001444static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001445{
1446 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001447 struct mem_cgroup *iter;
1448
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001449 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001450 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001451 return num;
1452}
1453
Balbir Singh6d61ef42009-01-07 18:08:06 -08001454/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001455 * Return the memory (and swap, if configured) limit for a memcg.
1456 */
1457u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1458{
1459 u64 limit;
1460 u64 memsw;
1461
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001462 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1463 limit += total_swap_pages << PAGE_SHIFT;
1464
David Rientjesa63d83f2010-08-09 17:19:46 -07001465 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1466 /*
1467 * If memsw is finite and limits the amount of swap space available
1468 * to this memcg, return that limit.
1469 */
1470 return min(limit, memsw);
1471}
1472
Johannes Weiner56600482012-01-12 17:17:59 -08001473static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1474 gfp_t gfp_mask,
1475 unsigned long flags)
1476{
1477 unsigned long total = 0;
1478 bool noswap = false;
1479 int loop;
1480
1481 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1482 noswap = true;
1483 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1484 noswap = true;
1485
1486 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1487 if (loop)
1488 drain_all_stock_async(memcg);
1489 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1490 /*
1491 * Allow limit shrinkers, which are triggered directly
1492 * by userspace, to catch signals and stop reclaim
1493 * after minimal progress, regardless of the margin.
1494 */
1495 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1496 break;
1497 if (mem_cgroup_margin(memcg))
1498 break;
1499 /*
1500 * If nothing was reclaimed after two attempts, there
1501 * may be no reclaimable pages in this hierarchy.
1502 */
1503 if (loop && !total)
1504 break;
1505 }
1506 return total;
1507}
1508
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001509/**
1510 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001511 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001512 * @nid: the node ID to be checked.
1513 * @noswap : specify true here if the user wants flle only information.
1514 *
1515 * This function returns whether the specified memcg contains any
1516 * reclaimable pages on a node. Returns true if there are any reclaimable
1517 * pages in the node.
1518 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001519static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001520 int nid, bool noswap)
1521{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001522 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001523 return true;
1524 if (noswap || !total_swap_pages)
1525 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001526 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001527 return true;
1528 return false;
1529
1530}
Ying Han889976d2011-05-26 16:25:33 -07001531#if MAX_NUMNODES > 1
1532
1533/*
1534 * Always updating the nodemask is not very good - even if we have an empty
1535 * list or the wrong list here, we can start from some node and traverse all
1536 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1537 *
1538 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001539static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001540{
1541 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001542 /*
1543 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1544 * pagein/pageout changes since the last update.
1545 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001546 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001547 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001548 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001549 return;
1550
Ying Han889976d2011-05-26 16:25:33 -07001551 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001552 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001553
1554 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1555
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001556 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1557 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001558 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001559
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001560 atomic_set(&memcg->numainfo_events, 0);
1561 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001562}
1563
1564/*
1565 * Selecting a node where we start reclaim from. Because what we need is just
1566 * reducing usage counter, start from anywhere is O,K. Considering
1567 * memory reclaim from current node, there are pros. and cons.
1568 *
1569 * Freeing memory from current node means freeing memory from a node which
1570 * we'll use or we've used. So, it may make LRU bad. And if several threads
1571 * hit limits, it will see a contention on a node. But freeing from remote
1572 * node means more costs for memory reclaim because of memory latency.
1573 *
1574 * Now, we use round-robin. Better algorithm is welcomed.
1575 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001576int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001577{
1578 int node;
1579
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001580 mem_cgroup_may_update_nodemask(memcg);
1581 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001582
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001584 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001585 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001586 /*
1587 * We call this when we hit limit, not when pages are added to LRU.
1588 * No LRU may hold pages because all pages are UNEVICTABLE or
1589 * memcg is too small and all pages are not on LRU. In that case,
1590 * we use curret node.
1591 */
1592 if (unlikely(node == MAX_NUMNODES))
1593 node = numa_node_id();
1594
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001595 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001596 return node;
1597}
1598
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001599/*
1600 * Check all nodes whether it contains reclaimable pages or not.
1601 * For quick scan, we make use of scan_nodes. This will allow us to skip
1602 * unused nodes. But scan_nodes is lazily updated and may not cotain
1603 * enough new information. We need to do double check.
1604 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001605static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001606{
1607 int nid;
1608
1609 /*
1610 * quick check...making use of scan_node.
1611 * We can skip unused nodes.
1612 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001613 if (!nodes_empty(memcg->scan_nodes)) {
1614 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001615 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001616 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001617
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001618 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001619 return true;
1620 }
1621 }
1622 /*
1623 * Check rest of nodes.
1624 */
1625 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001626 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001627 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001628 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001629 return true;
1630 }
1631 return false;
1632}
1633
Ying Han889976d2011-05-26 16:25:33 -07001634#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001635int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001636{
1637 return 0;
1638}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001639
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001640static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001641{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001642 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001643}
Ying Han889976d2011-05-26 16:25:33 -07001644#endif
1645
Johannes Weiner56600482012-01-12 17:17:59 -08001646static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1647 struct zone *zone,
1648 gfp_t gfp_mask,
1649 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001650{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001651 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001652 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001653 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001654 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001655 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001656 struct mem_cgroup_reclaim_cookie reclaim = {
1657 .zone = zone,
1658 .priority = 0,
1659 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001660
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001661 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001662
Balbir Singh4e416952009-09-23 15:56:39 -07001663 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001664 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001665 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001666 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001667 if (loop >= 2) {
1668 /*
1669 * If we have not been able to reclaim
1670 * anything, it might because there are
1671 * no reclaimable pages under this hierarchy
1672 */
Johannes Weiner56600482012-01-12 17:17:59 -08001673 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001674 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001675 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001676 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001677 * excess >> 2 is not to excessive so as to
1678 * reclaim too much, nor too less that we keep
1679 * coming back to reclaim from this cgroup
1680 */
1681 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001682 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001683 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001684 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001685 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001686 }
Johannes Weiner56600482012-01-12 17:17:59 -08001687 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001688 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001689 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1690 zone, &nr_scanned);
1691 *total_scanned += nr_scanned;
1692 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001693 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001694 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001695 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001696 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001697}
1698
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001699/*
1700 * Check OOM-Killer is already running under our hierarchy.
1701 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001702 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001703 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001704static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001705{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001706 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001707
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001708 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001709 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001710 /*
1711 * this subtree of our hierarchy is already locked
1712 * so we cannot give a lock.
1713 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001714 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001715 mem_cgroup_iter_break(memcg, iter);
1716 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001717 } else
1718 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001719 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001720
Michal Hocko79dfdac2011-07-26 16:08:23 -07001721 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001722 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001723
1724 /*
1725 * OK, we failed to lock the whole subtree so we have to clean up
1726 * what we set up to the failing subtree
1727 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001728 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001729 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001730 mem_cgroup_iter_break(memcg, iter);
1731 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001732 }
1733 iter->oom_lock = false;
1734 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001735 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001736}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001737
Michal Hocko79dfdac2011-07-26 16:08:23 -07001738/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001739 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001740 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001741static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001742{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001743 struct mem_cgroup *iter;
1744
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001745 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001746 iter->oom_lock = false;
1747 return 0;
1748}
1749
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001750static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001751{
1752 struct mem_cgroup *iter;
1753
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001754 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001755 atomic_inc(&iter->under_oom);
1756}
1757
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001758static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001759{
1760 struct mem_cgroup *iter;
1761
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001762 /*
1763 * When a new child is created while the hierarchy is under oom,
1764 * mem_cgroup_oom_lock() may not be called. We have to use
1765 * atomic_add_unless() here.
1766 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001767 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001768 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001769}
1770
Michal Hocko1af8efe2011-07-26 16:08:24 -07001771static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001772static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1773
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001774struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001775 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001776 wait_queue_t wait;
1777};
1778
1779static int memcg_oom_wake_function(wait_queue_t *wait,
1780 unsigned mode, int sync, void *arg)
1781{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001782 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1783 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001784 struct oom_wait_info *oom_wait_info;
1785
1786 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001787 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001788
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001789 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001790 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001791 * Then we can use css_is_ancestor without taking care of RCU.
1792 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001793 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1794 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001795 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001796 return autoremove_wake_function(wait, mode, sync, arg);
1797}
1798
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001799static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001800{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001801 /* for filtering, pass "memcg" as argument. */
1802 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001803}
1804
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001805static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001806{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001807 if (memcg && atomic_read(&memcg->under_oom))
1808 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001809}
1810
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001811/*
1812 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1813 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001814static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
1815 int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001816{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001817 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001818 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001819
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001820 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001821 owait.wait.flags = 0;
1822 owait.wait.func = memcg_oom_wake_function;
1823 owait.wait.private = current;
1824 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001825 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001826 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001827
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001828 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001829 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001830 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001831 /*
1832 * Even if signal_pending(), we can't quit charge() loop without
1833 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1834 * under OOM is always welcomed, use TASK_KILLABLE here.
1835 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001836 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001837 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001838 need_to_kill = false;
1839 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001840 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001841 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001842
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001843 if (need_to_kill) {
1844 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07001845 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001846 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001847 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001848 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001849 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001850 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001851 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001852 mem_cgroup_oom_unlock(memcg);
1853 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001854 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001855
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001856 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001857
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001858 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1859 return false;
1860 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001861 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001862 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001863}
1864
Balbir Singhd69b0422009-06-17 16:26:34 -07001865/*
1866 * Currently used to update mapped file statistics, but the routine can be
1867 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001868 *
1869 * Notes: Race condition
1870 *
1871 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1872 * it tends to be costly. But considering some conditions, we doesn't need
1873 * to do so _always_.
1874 *
1875 * Considering "charge", lock_page_cgroup() is not required because all
1876 * file-stat operations happen after a page is attached to radix-tree. There
1877 * are no race with "charge".
1878 *
1879 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1880 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1881 * if there are race with "uncharge". Statistics itself is properly handled
1882 * by flags.
1883 *
1884 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001885 * small, we check mm->moving_account and detect there are possibility of race
1886 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001887 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001888
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001889void __mem_cgroup_begin_update_page_stat(struct page *page,
1890 bool *locked, unsigned long *flags)
1891{
1892 struct mem_cgroup *memcg;
1893 struct page_cgroup *pc;
1894
1895 pc = lookup_page_cgroup(page);
1896again:
1897 memcg = pc->mem_cgroup;
1898 if (unlikely(!memcg || !PageCgroupUsed(pc)))
1899 return;
1900 /*
1901 * If this memory cgroup is not under account moving, we don't
1902 * need to take move_lock_page_cgroup(). Because we already hold
1903 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001904 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001905 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001906 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001907 return;
1908
1909 move_lock_mem_cgroup(memcg, flags);
1910 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
1911 move_unlock_mem_cgroup(memcg, flags);
1912 goto again;
1913 }
1914 *locked = true;
1915}
1916
1917void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
1918{
1919 struct page_cgroup *pc = lookup_page_cgroup(page);
1920
1921 /*
1922 * It's guaranteed that pc->mem_cgroup never changes while
1923 * lock is held because a routine modifies pc->mem_cgroup
1924 * should take move_lock_page_cgroup().
1925 */
1926 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1927}
1928
Greg Thelen2a7106f2011-01-13 15:47:37 -08001929void mem_cgroup_update_page_stat(struct page *page,
1930 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001931{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001932 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001933 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001934 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001935
Johannes Weinercfa44942012-01-12 17:18:38 -08001936 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001937 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001938
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001939 memcg = pc->mem_cgroup;
1940 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001941 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001942
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001943 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001944 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08001945 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001946 break;
1947 default:
1948 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001949 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001950
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001951 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07001952}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001953
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001954/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001955 * size of first charge trial. "32" comes from vmscan.c's magic value.
1956 * TODO: maybe necessary to use big numbers in big irons.
1957 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001958#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001959struct memcg_stock_pcp {
1960 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001961 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001962 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001963 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001964#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001965};
1966static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001967static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001968
1969/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001970 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001971 * from local stock and true is returned. If the stock is 0 or charges from a
1972 * cgroup which is not current target, returns false. This stock will be
1973 * refilled.
1974 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001975static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001976{
1977 struct memcg_stock_pcp *stock;
1978 bool ret = true;
1979
1980 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001981 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001982 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001983 else /* need to call res_counter_charge */
1984 ret = false;
1985 put_cpu_var(memcg_stock);
1986 return ret;
1987}
1988
1989/*
1990 * Returns stocks cached in percpu to res_counter and reset cached information.
1991 */
1992static void drain_stock(struct memcg_stock_pcp *stock)
1993{
1994 struct mem_cgroup *old = stock->cached;
1995
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001996 if (stock->nr_pages) {
1997 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
1998
1999 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002000 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002001 res_counter_uncharge(&old->memsw, bytes);
2002 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002003 }
2004 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002005}
2006
2007/*
2008 * This must be called under preempt disabled or must be called by
2009 * a thread which is pinned to local cpu.
2010 */
2011static void drain_local_stock(struct work_struct *dummy)
2012{
2013 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2014 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002015 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002016}
2017
2018/*
2019 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002020 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002021 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002022static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002023{
2024 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2025
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002026 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002027 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002028 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002029 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002030 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002031 put_cpu_var(memcg_stock);
2032}
2033
2034/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002035 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002036 * of the hierarchy under it. sync flag says whether we should block
2037 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002038 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002039static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002040{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002041 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002042
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002043 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002044 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002045 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002046 for_each_online_cpu(cpu) {
2047 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002048 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002049
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002050 memcg = stock->cached;
2051 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002052 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002053 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002054 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002055 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2056 if (cpu == curcpu)
2057 drain_local_stock(&stock->work);
2058 else
2059 schedule_work_on(cpu, &stock->work);
2060 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002061 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002062 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002063
2064 if (!sync)
2065 goto out;
2066
2067 for_each_online_cpu(cpu) {
2068 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002069 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002070 flush_work(&stock->work);
2071 }
2072out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002073 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002074}
2075
2076/*
2077 * Tries to drain stocked charges in other cpus. This function is asynchronous
2078 * and just put a work per cpu for draining localy on each cpu. Caller can
2079 * expects some charges will be back to res_counter later but cannot wait for
2080 * it.
2081 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002082static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002083{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002084 /*
2085 * If someone calls draining, avoid adding more kworker runs.
2086 */
2087 if (!mutex_trylock(&percpu_charge_mutex))
2088 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002089 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002090 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002091}
2092
2093/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002094static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002095{
2096 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002097 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002098 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002099 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002100}
2101
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002102/*
2103 * This function drains percpu counter value from DEAD cpu and
2104 * move it to local cpu. Note that this function can be preempted.
2105 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002106static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002107{
2108 int i;
2109
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002110 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002111 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002112 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002113
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002114 per_cpu(memcg->stat->count[i], cpu) = 0;
2115 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002116 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002117 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002118 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002119
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002120 per_cpu(memcg->stat->events[i], cpu) = 0;
2121 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002122 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002123 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002124}
2125
2126static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002127 unsigned long action,
2128 void *hcpu)
2129{
2130 int cpu = (unsigned long)hcpu;
2131 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002132 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002133
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002134 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002135 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002136
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002137 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002138 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002139
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002140 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002141 mem_cgroup_drain_pcp_counter(iter, cpu);
2142
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002143 stock = &per_cpu(memcg_stock, cpu);
2144 drain_stock(stock);
2145 return NOTIFY_OK;
2146}
2147
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002148
2149/* See __mem_cgroup_try_charge() for details */
2150enum {
2151 CHARGE_OK, /* success */
2152 CHARGE_RETRY, /* need to retry but retry is not bad */
2153 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2154 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2155 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2156};
2157
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002158static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002159 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002160{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002161 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002162 struct mem_cgroup *mem_over_limit;
2163 struct res_counter *fail_res;
2164 unsigned long flags = 0;
2165 int ret;
2166
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002167 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002168
2169 if (likely(!ret)) {
2170 if (!do_swap_account)
2171 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002172 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002173 if (likely(!ret))
2174 return CHARGE_OK;
2175
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002176 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002177 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2178 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2179 } else
2180 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002181 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002182 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2183 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002184 *
2185 * Never reclaim on behalf of optional batching, retry with a
2186 * single page instead.
2187 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002188 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002189 return CHARGE_RETRY;
2190
2191 if (!(gfp_mask & __GFP_WAIT))
2192 return CHARGE_WOULDBLOCK;
2193
Johannes Weiner56600482012-01-12 17:17:59 -08002194 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002195 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002196 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002197 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002198 * Even though the limit is exceeded at this point, reclaim
2199 * may have been able to free some pages. Retry the charge
2200 * before killing the task.
2201 *
2202 * Only for regular pages, though: huge pages are rather
2203 * unlikely to succeed so close to the limit, and we fall back
2204 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002205 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002206 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002207 return CHARGE_RETRY;
2208
2209 /*
2210 * At task move, charge accounts can be doubly counted. So, it's
2211 * better to wait until the end of task_move if something is going on.
2212 */
2213 if (mem_cgroup_wait_acct_move(mem_over_limit))
2214 return CHARGE_RETRY;
2215
2216 /* If we don't need to call oom-killer at el, return immediately */
2217 if (!oom_check)
2218 return CHARGE_NOMEM;
2219 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002220 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002221 return CHARGE_OOM_DIE;
2222
2223 return CHARGE_RETRY;
2224}
2225
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002226/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002227 * __mem_cgroup_try_charge() does
2228 * 1. detect memcg to be charged against from passed *mm and *ptr,
2229 * 2. update res_counter
2230 * 3. call memory reclaim if necessary.
2231 *
2232 * In some special case, if the task is fatal, fatal_signal_pending() or
2233 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2234 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2235 * as possible without any hazards. 2: all pages should have a valid
2236 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2237 * pointer, that is treated as a charge to root_mem_cgroup.
2238 *
2239 * So __mem_cgroup_try_charge() will return
2240 * 0 ... on success, filling *ptr with a valid memcg pointer.
2241 * -ENOMEM ... charge failure because of resource limits.
2242 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2243 *
2244 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2245 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002246 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002247static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002248 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002249 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002250 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002251 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002252{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002253 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002254 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002255 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002256 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002257
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002258 /*
2259 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2260 * in system level. So, allow to go ahead dying process in addition to
2261 * MEMDIE process.
2262 */
2263 if (unlikely(test_thread_flag(TIF_MEMDIE)
2264 || fatal_signal_pending(current)))
2265 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002266
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002267 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002268 * We always charge the cgroup the mm_struct belongs to.
2269 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002270 * thread group leader migrates. It's possible that mm is not
2271 * set, if so charge the init_mm (happens for pagecache usage).
2272 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002273 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002274 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002275again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002276 if (*ptr) { /* css should be a valid one */
2277 memcg = *ptr;
2278 VM_BUG_ON(css_is_removed(&memcg->css));
2279 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002280 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002281 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002282 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002283 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002284 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002285 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002286
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002287 rcu_read_lock();
2288 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002289 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002290 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002291 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002292 * race with swapoff. Then, we have small risk of mis-accouning.
2293 * But such kind of mis-account by race always happens because
2294 * we don't have cgroup_mutex(). It's overkill and we allo that
2295 * small race, here.
2296 * (*) swapoff at el will charge against mm-struct not against
2297 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002298 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002299 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002300 if (!memcg)
2301 memcg = root_mem_cgroup;
2302 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002303 rcu_read_unlock();
2304 goto done;
2305 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002306 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002307 /*
2308 * It seems dagerous to access memcg without css_get().
2309 * But considering how consume_stok works, it's not
2310 * necessary. If consume_stock success, some charges
2311 * from this memcg are cached on this cpu. So, we
2312 * don't need to call css_get()/css_tryget() before
2313 * calling consume_stock().
2314 */
2315 rcu_read_unlock();
2316 goto done;
2317 }
2318 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002319 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002320 rcu_read_unlock();
2321 goto again;
2322 }
2323 rcu_read_unlock();
2324 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002325
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002326 do {
2327 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002328
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002329 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002330 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002331 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002332 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002333 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002334
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002335 oom_check = false;
2336 if (oom && !nr_oom_retries) {
2337 oom_check = true;
2338 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2339 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002340
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002341 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002342 switch (ret) {
2343 case CHARGE_OK:
2344 break;
2345 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002346 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002347 css_put(&memcg->css);
2348 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002349 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002350 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002351 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002352 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002353 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002354 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002355 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002356 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002357 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002358 /* If oom, we never return -ENOMEM */
2359 nr_oom_retries--;
2360 break;
2361 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002362 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002363 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002364 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002365 } while (ret != CHARGE_OK);
2366
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002367 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002368 refill_stock(memcg, batch - nr_pages);
2369 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002370done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002371 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002372 return 0;
2373nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002374 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002375 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002376bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002377 *ptr = root_mem_cgroup;
2378 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002379}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002380
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002381/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002382 * Somemtimes we have to undo a charge we got by try_charge().
2383 * This function is for that and do uncharge, put css's refcnt.
2384 * gotten by try_charge().
2385 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002386static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002387 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002388{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002389 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002390 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002391
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002392 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002393 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002394 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002395 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002396}
2397
2398/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002399 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2400 * This is useful when moving usage to parent cgroup.
2401 */
2402static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2403 unsigned int nr_pages)
2404{
2405 unsigned long bytes = nr_pages * PAGE_SIZE;
2406
2407 if (mem_cgroup_is_root(memcg))
2408 return;
2409
2410 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2411 if (do_swap_account)
2412 res_counter_uncharge_until(&memcg->memsw,
2413 memcg->memsw.parent, bytes);
2414}
2415
2416/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002417 * A helper function to get mem_cgroup from ID. must be called under
2418 * rcu_read_lock(). The caller must check css_is_removed() or some if
2419 * it's concern. (dropping refcnt from swap can be called against removed
2420 * memcg.)
2421 */
2422static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2423{
2424 struct cgroup_subsys_state *css;
2425
2426 /* ID 0 is unused ID */
2427 if (!id)
2428 return NULL;
2429 css = css_lookup(&mem_cgroup_subsys, id);
2430 if (!css)
2431 return NULL;
2432 return container_of(css, struct mem_cgroup, css);
2433}
2434
Wu Fengguange42d9d52009-12-16 12:19:59 +01002435struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002436{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002437 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002438 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002439 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002440 swp_entry_t ent;
2441
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002442 VM_BUG_ON(!PageLocked(page));
2443
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002444 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002445 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002446 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002447 memcg = pc->mem_cgroup;
2448 if (memcg && !css_tryget(&memcg->css))
2449 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002450 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002451 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002452 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002453 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002454 memcg = mem_cgroup_lookup(id);
2455 if (memcg && !css_tryget(&memcg->css))
2456 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002457 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002458 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002459 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002460 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002461}
2462
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002463static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002464 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002465 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002466 enum charge_type ctype,
2467 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002468{
Johannes Weinerce587e62012-04-24 20:22:33 +02002469 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002470 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002471 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002472 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002473 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002474
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002475 lock_page_cgroup(pc);
2476 if (unlikely(PageCgroupUsed(pc))) {
2477 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002478 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002479 return;
2480 }
2481 /*
2482 * we don't need page_cgroup_lock about tail pages, becase they are not
2483 * accessed by any other context at this point.
2484 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002485
2486 /*
2487 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2488 * may already be on some other mem_cgroup's LRU. Take care of it.
2489 */
2490 if (lrucare) {
2491 zone = page_zone(page);
2492 spin_lock_irq(&zone->lru_lock);
2493 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002494 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002495 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002496 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002497 was_on_lru = true;
2498 }
2499 }
2500
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002501 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002502 /*
2503 * We access a page_cgroup asynchronously without lock_page_cgroup().
2504 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2505 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2506 * before USED bit, we need memory barrier here.
2507 * See mem_cgroup_add_lru_list(), etc.
2508 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002509 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002510 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002511
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002512 if (lrucare) {
2513 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002514 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002515 VM_BUG_ON(PageLRU(page));
2516 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002517 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002518 }
2519 spin_unlock_irq(&zone->lru_lock);
2520 }
2521
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002522 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2523 anon = true;
2524 else
2525 anon = false;
2526
2527 mem_cgroup_charge_statistics(memcg, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002528 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002529
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002530 /*
2531 * "charge_statistics" updated event counter. Then, check it.
2532 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2533 * if they exceeds softlimit.
2534 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002535 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002536}
2537
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002538#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2539
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002540#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002541/*
2542 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002543 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2544 * charge/uncharge will be never happen and move_account() is done under
2545 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002546 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002547void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002548{
2549 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002550 struct page_cgroup *pc;
2551 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002552
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002553 if (mem_cgroup_disabled())
2554 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002555 for (i = 1; i < HPAGE_PMD_NR; i++) {
2556 pc = head_pc + i;
2557 pc->mem_cgroup = head_pc->mem_cgroup;
2558 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002559 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2560 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002561}
Hugh Dickins12d27102012-01-12 17:19:52 -08002562#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002563
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002564/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002565 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002566 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002567 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002568 * @pc: page_cgroup of the page.
2569 * @from: mem_cgroup which the page is moved from.
2570 * @to: mem_cgroup which the page is moved to. @from != @to.
2571 *
2572 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002573 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002574 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002575 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002576 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2577 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002578 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002579static int mem_cgroup_move_account(struct page *page,
2580 unsigned int nr_pages,
2581 struct page_cgroup *pc,
2582 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002583 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002584{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002585 unsigned long flags;
2586 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002587 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002588
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002589 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002590 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002591 /*
2592 * The page is isolated from LRU. So, collapse function
2593 * will not handle this page. But page splitting can happen.
2594 * Do this check under compound_page_lock(). The caller should
2595 * hold it.
2596 */
2597 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002598 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002599 goto out;
2600
2601 lock_page_cgroup(pc);
2602
2603 ret = -EINVAL;
2604 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2605 goto unlock;
2606
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002607 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002608
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002609 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002610 /* Update mapped_file data for mem_cgroup */
2611 preempt_disable();
2612 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2613 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2614 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002615 }
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002616 mem_cgroup_charge_statistics(from, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002617
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002618 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002619 pc->mem_cgroup = to;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002620 mem_cgroup_charge_statistics(to, anon, nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002621 /*
2622 * We charges against "to" which may not have any tasks. Then, "to"
2623 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002624 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002625 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002626 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002627 */
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002628 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002629 ret = 0;
2630unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002631 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002632 /*
2633 * check events
2634 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002635 memcg_check_events(to, page);
2636 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002637out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002638 return ret;
2639}
2640
2641/*
2642 * move charges to its parent.
2643 */
2644
Johannes Weiner5564e882011-03-23 16:42:29 -07002645static int mem_cgroup_move_parent(struct page *page,
2646 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002647 struct mem_cgroup *child,
2648 gfp_t gfp_mask)
2649{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002650 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002651 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002652 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002653 int ret;
2654
2655 /* Is ROOT ? */
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002656 if (mem_cgroup_is_root(child))
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002657 return -EINVAL;
2658
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002659 ret = -EBUSY;
2660 if (!get_page_unless_zero(page))
2661 goto out;
2662 if (isolate_lru_page(page))
2663 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002664
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002665 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002666
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002667 parent = parent_mem_cgroup(child);
2668 /*
2669 * If no parent, move charges to root cgroup.
2670 */
2671 if (!parent)
2672 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002673
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002674 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002675 flags = compound_lock_irqsave(page);
2676
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002677 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07002678 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07002679 if (!ret)
2680 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002681
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002682 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002683 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002684 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002685put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002686 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002687out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002688 return ret;
2689}
2690
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002691/*
2692 * Charge the memory controller for page usage.
2693 * Return
2694 * 0 if the charge was successful
2695 * < 0 if the cgroup is over its limit
2696 */
2697static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002698 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002699{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002700 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002701 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002702 bool oom = true;
2703 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002704
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002705 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002706 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002707 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002708 /*
2709 * Never OOM-kill a process for a huge page. The
2710 * fault handler will fall back to regular pages.
2711 */
2712 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002713 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002714
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002715 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002716 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002717 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02002718 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002719 return 0;
2720}
2721
2722int mem_cgroup_newpage_charge(struct page *page,
2723 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002724{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002725 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002726 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002727 VM_BUG_ON(page_mapped(page));
2728 VM_BUG_ON(page->mapping && !PageAnon(page));
2729 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002730 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002731 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002732}
2733
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002734static void
2735__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2736 enum charge_type ctype);
2737
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002738int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2739 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002740{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002741 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002742 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002743 int ret;
2744
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002745 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002746 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002747 if (PageCompound(page))
2748 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002749
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002750 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002751 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002752 if (!page_is_file_cache(page))
2753 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002754
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002755 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002756 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002757 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002758 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002759 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002760 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2761 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002762 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002763}
2764
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002765/*
2766 * While swap-in, try_charge -> commit or cancel, the page is locked.
2767 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002768 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002769 * "commit()" or removed by "cancel()"
2770 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002771int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2772 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002773 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002774{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002775 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002776 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002777
Johannes Weiner72835c82012-01-12 17:18:32 -08002778 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002779
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002780 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002781 return 0;
2782
2783 if (!do_swap_account)
2784 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002785 /*
2786 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002787 * the pte, and even removed page from swap cache: in those cases
2788 * do_swap_page()'s pte_same() test will fail; but there's also a
2789 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002790 */
2791 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002792 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002793 memcg = try_get_mem_cgroup_from_page(page);
2794 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002795 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002796 *memcgp = memcg;
2797 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002798 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002799 if (ret == -EINTR)
2800 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002801 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002802charge_cur_mm:
2803 if (unlikely(!mm))
2804 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002805 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2806 if (ret == -EINTR)
2807 ret = 0;
2808 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002809}
2810
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002811static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002812__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002813 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002814{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002815 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002816 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002817 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002818 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002819 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002820
Johannes Weinerce587e62012-04-24 20:22:33 +02002821 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002822 /*
2823 * Now swap is on-memory. This means this page may be
2824 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002825 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2826 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2827 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002828 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002829 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002830 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07002831 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002832 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002833 /*
2834 * At swapin, we may charge account against cgroup which has no tasks.
2835 * So, rmdir()->pre_destroy() can be called while we do this charge.
2836 * In that case, we need to call pre_destroy() again. check it here.
2837 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002838 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002839}
2840
Johannes Weiner72835c82012-01-12 17:18:32 -08002841void mem_cgroup_commit_charge_swapin(struct page *page,
2842 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002843{
Johannes Weiner72835c82012-01-12 17:18:32 -08002844 __mem_cgroup_commit_charge_swapin(page, memcg,
2845 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002846}
2847
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002848void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002849{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002850 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002851 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002852 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002853 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002854 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002855}
2856
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002857static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002858 unsigned int nr_pages,
2859 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002860{
2861 struct memcg_batch_info *batch = NULL;
2862 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002863
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002864 /* If swapout, usage of swap doesn't decrease */
2865 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2866 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002867
2868 batch = &current->memcg_batch;
2869 /*
2870 * In usual, we do css_get() when we remember memcg pointer.
2871 * But in this case, we keep res->usage until end of a series of
2872 * uncharges. Then, it's ok to ignore memcg's refcnt.
2873 */
2874 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002875 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002876 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002877 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002878 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002879 * the same cgroup and we have chance to coalesce uncharges.
2880 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2881 * because we want to do uncharge as soon as possible.
2882 */
2883
2884 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2885 goto direct_uncharge;
2886
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002887 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002888 goto direct_uncharge;
2889
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002890 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002891 * In typical case, batch->memcg == mem. This means we can
2892 * merge a series of uncharges to an uncharge of res_counter.
2893 * If not, we uncharge res_counter ony by one.
2894 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002895 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002896 goto direct_uncharge;
2897 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002898 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002899 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002900 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002901 return;
2902direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002903 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002904 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002905 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2906 if (unlikely(batch->memcg != memcg))
2907 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002908}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002909
Balbir Singh8697d332008-02-07 00:13:59 -08002910/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002911 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002912 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002913static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002914__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002915{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002916 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002917 unsigned int nr_pages = 1;
2918 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002919 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002920
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002921 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002922 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002923
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002924 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002925 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002926
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002927 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002928 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002929 VM_BUG_ON(!PageTransHuge(page));
2930 }
Balbir Singh8697d332008-02-07 00:13:59 -08002931 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002932 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002933 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002934 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002935 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002936 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002937
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002938 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002939
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002940 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002941
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002942 if (!PageCgroupUsed(pc))
2943 goto unlock_out;
2944
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002945 anon = PageAnon(page);
2946
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002947 switch (ctype) {
2948 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002949 /*
2950 * Generally PageAnon tells if it's the anon statistics to be
2951 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2952 * used before page reached the stage of being marked PageAnon.
2953 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002954 anon = true;
2955 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002956 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07002957 /* See mem_cgroup_prepare_migration() */
2958 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002959 goto unlock_out;
2960 break;
2961 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2962 if (!PageAnon(page)) { /* Shared memory */
2963 if (page->mapping && !page_is_file_cache(page))
2964 goto unlock_out;
2965 } else if (page_mapped(page)) /* Anon */
2966 goto unlock_out;
2967 break;
2968 default:
2969 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002970 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002971
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002972 mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002973
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002974 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002975 /*
2976 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2977 * freed from LRU. This is safe because uncharged page is expected not
2978 * to be reused (freed soon). Exception is SwapCache, it's handled by
2979 * special functions.
2980 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08002981
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002982 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002983 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002984 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002985 * will never be freed.
2986 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002987 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002988 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002989 mem_cgroup_swap_statistics(memcg, true);
2990 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002991 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002992 if (!mem_cgroup_is_root(memcg))
2993 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002994
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002995 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002996
2997unlock_out:
2998 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002999 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003000}
3001
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003002void mem_cgroup_uncharge_page(struct page *page)
3003{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003004 /* early check. */
3005 if (page_mapped(page))
3006 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08003007 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003008 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3009}
3010
3011void mem_cgroup_uncharge_cache_page(struct page *page)
3012{
3013 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003014 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003015 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3016}
3017
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003018/*
3019 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3020 * In that cases, pages are freed continuously and we can expect pages
3021 * are in the same memcg. All these calls itself limits the number of
3022 * pages freed at once, then uncharge_start/end() is called properly.
3023 * This may be called prural(2) times in a context,
3024 */
3025
3026void mem_cgroup_uncharge_start(void)
3027{
3028 current->memcg_batch.do_batch++;
3029 /* We can do nest. */
3030 if (current->memcg_batch.do_batch == 1) {
3031 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003032 current->memcg_batch.nr_pages = 0;
3033 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003034 }
3035}
3036
3037void mem_cgroup_uncharge_end(void)
3038{
3039 struct memcg_batch_info *batch = &current->memcg_batch;
3040
3041 if (!batch->do_batch)
3042 return;
3043
3044 batch->do_batch--;
3045 if (batch->do_batch) /* If stacked, do nothing. */
3046 return;
3047
3048 if (!batch->memcg)
3049 return;
3050 /*
3051 * This "batch->memcg" is valid without any css_get/put etc...
3052 * bacause we hide charges behind us.
3053 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003054 if (batch->nr_pages)
3055 res_counter_uncharge(&batch->memcg->res,
3056 batch->nr_pages * PAGE_SIZE);
3057 if (batch->memsw_nr_pages)
3058 res_counter_uncharge(&batch->memcg->memsw,
3059 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003060 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003061 /* forget this pointer (for sanity check) */
3062 batch->memcg = NULL;
3063}
3064
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003065#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003066/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003067 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003068 * memcg information is recorded to swap_cgroup of "ent"
3069 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003070void
3071mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003072{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003073 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003074 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003075
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003076 if (!swapout) /* this was a swap cache but the swap is unused ! */
3077 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3078
3079 memcg = __mem_cgroup_uncharge_common(page, ctype);
3080
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003081 /*
3082 * record memcg information, if swapout && memcg != NULL,
3083 * mem_cgroup_get() was called in uncharge().
3084 */
3085 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003086 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003087}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003088#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003089
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003090#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3091/*
3092 * called from swap_entry_free(). remove record in swap_cgroup and
3093 * uncharge "memsw" account.
3094 */
3095void mem_cgroup_uncharge_swap(swp_entry_t ent)
3096{
3097 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003098 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003099
3100 if (!do_swap_account)
3101 return;
3102
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003103 id = swap_cgroup_record(ent, 0);
3104 rcu_read_lock();
3105 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003106 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003107 /*
3108 * We uncharge this because swap is freed.
3109 * This memcg can be obsolete one. We avoid calling css_tryget
3110 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003111 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003112 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003113 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003114 mem_cgroup_put(memcg);
3115 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003116 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003117}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003118
3119/**
3120 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3121 * @entry: swap entry to be moved
3122 * @from: mem_cgroup which the entry is moved from
3123 * @to: mem_cgroup which the entry is moved to
3124 *
3125 * It succeeds only when the swap_cgroup's record for this entry is the same
3126 * as the mem_cgroup's id of @from.
3127 *
3128 * Returns 0 on success, -EINVAL on failure.
3129 *
3130 * The caller must have charged to @to, IOW, called res_counter_charge() about
3131 * both res and memsw, and called css_get().
3132 */
3133static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003134 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003135{
3136 unsigned short old_id, new_id;
3137
3138 old_id = css_id(&from->css);
3139 new_id = css_id(&to->css);
3140
3141 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003142 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003143 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003144 /*
3145 * This function is only called from task migration context now.
3146 * It postpones res_counter and refcount handling till the end
3147 * of task migration(mem_cgroup_clear_mc()) for performance
3148 * improvement. But we cannot postpone mem_cgroup_get(to)
3149 * because if the process that has been moved to @to does
3150 * swap-in, the refcount of @to might be decreased to 0.
3151 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003152 mem_cgroup_get(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003153 return 0;
3154 }
3155 return -EINVAL;
3156}
3157#else
3158static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003159 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003160{
3161 return -EINVAL;
3162}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003163#endif
3164
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003165/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003166 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3167 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003168 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003169int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003170 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003171{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003172 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003173 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003174 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003175 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003176
Johannes Weiner72835c82012-01-12 17:18:32 -08003177 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003178
Andrea Arcangeliec168512011-01-13 15:46:56 -08003179 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003180 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003181 return 0;
3182
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003183 pc = lookup_page_cgroup(page);
3184 lock_page_cgroup(pc);
3185 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003186 memcg = pc->mem_cgroup;
3187 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003188 /*
3189 * At migrating an anonymous page, its mapcount goes down
3190 * to 0 and uncharge() will be called. But, even if it's fully
3191 * unmapped, migration may fail and this page has to be
3192 * charged again. We set MIGRATION flag here and delay uncharge
3193 * until end_migration() is called
3194 *
3195 * Corner Case Thinking
3196 * A)
3197 * When the old page was mapped as Anon and it's unmap-and-freed
3198 * while migration was ongoing.
3199 * If unmap finds the old page, uncharge() of it will be delayed
3200 * until end_migration(). If unmap finds a new page, it's
3201 * uncharged when it make mapcount to be 1->0. If unmap code
3202 * finds swap_migration_entry, the new page will not be mapped
3203 * and end_migration() will find it(mapcount==0).
3204 *
3205 * B)
3206 * When the old page was mapped but migraion fails, the kernel
3207 * remaps it. A charge for it is kept by MIGRATION flag even
3208 * if mapcount goes down to 0. We can do remap successfully
3209 * without charging it again.
3210 *
3211 * C)
3212 * The "old" page is under lock_page() until the end of
3213 * migration, so, the old page itself will not be swapped-out.
3214 * If the new page is swapped out before end_migraton, our
3215 * hook to usual swap-out path will catch the event.
3216 */
3217 if (PageAnon(page))
3218 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003219 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003220 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003221 /*
3222 * If the page is not charged at this point,
3223 * we return here.
3224 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003225 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003226 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003227
Johannes Weiner72835c82012-01-12 17:18:32 -08003228 *memcgp = memcg;
3229 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003230 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003231 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003232 if (PageAnon(page)) {
3233 lock_page_cgroup(pc);
3234 ClearPageCgroupMigration(pc);
3235 unlock_page_cgroup(pc);
3236 /*
3237 * The old page may be fully unmapped while we kept it.
3238 */
3239 mem_cgroup_uncharge_page(page);
3240 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003241 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003242 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003243 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003244 /*
3245 * We charge new page before it's used/mapped. So, even if unlock_page()
3246 * is called before end_migration, we can catch all events on this new
3247 * page. In the case new page is migrated but not remapped, new page's
3248 * mapcount will be finally 0 and we call uncharge in end_migration().
3249 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003250 if (PageAnon(page))
3251 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3252 else if (page_is_file_cache(page))
3253 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3254 else
3255 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Johannes Weinerce587e62012-04-24 20:22:33 +02003256 __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003257 return ret;
3258}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003259
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003260/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003261void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003262 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003263{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003264 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003265 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003266 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003267
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003268 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003269 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003270 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003271 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003272 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003273 used = oldpage;
3274 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003275 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003276 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003277 unused = oldpage;
3278 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003279 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003280 * We disallowed uncharge of pages under migration because mapcount
3281 * of the page goes down to zero, temporarly.
3282 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003283 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003284 pc = lookup_page_cgroup(oldpage);
3285 lock_page_cgroup(pc);
3286 ClearPageCgroupMigration(pc);
3287 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003288 anon = PageAnon(used);
3289 __mem_cgroup_uncharge_common(unused,
3290 anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED
3291 : MEM_CGROUP_CHARGE_TYPE_CACHE);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003292
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003293 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003294 * If a page is a file cache, radix-tree replacement is very atomic
3295 * and we can skip this check. When it was an Anon page, its mapcount
3296 * goes down to 0. But because we added MIGRATION flage, it's not
3297 * uncharged yet. There are several case but page->mapcount check
3298 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3299 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003300 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003301 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003302 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003303 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003304 * At migration, we may charge account against cgroup which has no
3305 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003306 * So, rmdir()->pre_destroy() can be called while we do this charge.
3307 * In that case, we need to call pre_destroy() again. check it here.
3308 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003309 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003310}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003311
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003312/*
3313 * At replace page cache, newpage is not under any memcg but it's on
3314 * LRU. So, this function doesn't touch res_counter but handles LRU
3315 * in correct way. Both pages are locked so we cannot race with uncharge.
3316 */
3317void mem_cgroup_replace_page_cache(struct page *oldpage,
3318 struct page *newpage)
3319{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003320 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003321 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003322 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003323
3324 if (mem_cgroup_disabled())
3325 return;
3326
3327 pc = lookup_page_cgroup(oldpage);
3328 /* fix accounting on old pages */
3329 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003330 if (PageCgroupUsed(pc)) {
3331 memcg = pc->mem_cgroup;
3332 mem_cgroup_charge_statistics(memcg, false, -1);
3333 ClearPageCgroupUsed(pc);
3334 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003335 unlock_page_cgroup(pc);
3336
Hugh Dickinsbde05d12012-05-29 15:06:38 -07003337 /*
3338 * When called from shmem_replace_page(), in some cases the
3339 * oldpage has already been charged, and in some cases not.
3340 */
3341 if (!memcg)
3342 return;
3343
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003344 if (PageSwapBacked(oldpage))
3345 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3346
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003347 /*
3348 * Even if newpage->mapping was NULL before starting replacement,
3349 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3350 * LRU while we overwrite pc->mem_cgroup.
3351 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003352 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003353}
3354
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003355#ifdef CONFIG_DEBUG_VM
3356static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3357{
3358 struct page_cgroup *pc;
3359
3360 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003361 /*
3362 * Can be NULL while feeding pages into the page allocator for
3363 * the first time, i.e. during boot or memory hotplug;
3364 * or when mem_cgroup_disabled().
3365 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003366 if (likely(pc) && PageCgroupUsed(pc))
3367 return pc;
3368 return NULL;
3369}
3370
3371bool mem_cgroup_bad_page_check(struct page *page)
3372{
3373 if (mem_cgroup_disabled())
3374 return false;
3375
3376 return lookup_page_cgroup_used(page) != NULL;
3377}
3378
3379void mem_cgroup_print_bad_page(struct page *page)
3380{
3381 struct page_cgroup *pc;
3382
3383 pc = lookup_page_cgroup_used(page);
3384 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003385 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003386 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003387 }
3388}
3389#endif
3390
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003391static DEFINE_MUTEX(set_limit_mutex);
3392
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003393static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003394 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003395{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003396 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003397 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003398 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003399 int children = mem_cgroup_count_children(memcg);
3400 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003401 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003402
3403 /*
3404 * For keeping hierarchical_reclaim simple, how long we should retry
3405 * is depends on callers. We set our retry-count to be function
3406 * of # of children which we should visit in this loop.
3407 */
3408 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3409
3410 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003411
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003412 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003413 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003414 if (signal_pending(current)) {
3415 ret = -EINTR;
3416 break;
3417 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003418 /*
3419 * Rather than hide all in some function, I do this in
3420 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003421 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003422 */
3423 mutex_lock(&set_limit_mutex);
3424 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3425 if (memswlimit < val) {
3426 ret = -EINVAL;
3427 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003428 break;
3429 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003430
3431 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3432 if (memlimit < val)
3433 enlarge = 1;
3434
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003435 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003436 if (!ret) {
3437 if (memswlimit == val)
3438 memcg->memsw_is_minimum = true;
3439 else
3440 memcg->memsw_is_minimum = false;
3441 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003442 mutex_unlock(&set_limit_mutex);
3443
3444 if (!ret)
3445 break;
3446
Johannes Weiner56600482012-01-12 17:17:59 -08003447 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3448 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003449 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3450 /* Usage is reduced ? */
3451 if (curusage >= oldusage)
3452 retry_count--;
3453 else
3454 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003455 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003456 if (!ret && enlarge)
3457 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003458
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003459 return ret;
3460}
3461
Li Zefan338c8432009-06-17 16:27:15 -07003462static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3463 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003464{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003465 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003466 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003467 int children = mem_cgroup_count_children(memcg);
3468 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003469 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003470
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003471 /* see mem_cgroup_resize_res_limit */
3472 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3473 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003474 while (retry_count) {
3475 if (signal_pending(current)) {
3476 ret = -EINTR;
3477 break;
3478 }
3479 /*
3480 * Rather than hide all in some function, I do this in
3481 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003482 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003483 */
3484 mutex_lock(&set_limit_mutex);
3485 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3486 if (memlimit > val) {
3487 ret = -EINVAL;
3488 mutex_unlock(&set_limit_mutex);
3489 break;
3490 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003491 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3492 if (memswlimit < val)
3493 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003494 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003495 if (!ret) {
3496 if (memlimit == val)
3497 memcg->memsw_is_minimum = true;
3498 else
3499 memcg->memsw_is_minimum = false;
3500 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003501 mutex_unlock(&set_limit_mutex);
3502
3503 if (!ret)
3504 break;
3505
Johannes Weiner56600482012-01-12 17:17:59 -08003506 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3507 MEM_CGROUP_RECLAIM_NOSWAP |
3508 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003509 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003510 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003511 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003512 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003513 else
3514 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003515 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003516 if (!ret && enlarge)
3517 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003518 return ret;
3519}
3520
Balbir Singh4e416952009-09-23 15:56:39 -07003521unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003522 gfp_t gfp_mask,
3523 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003524{
3525 unsigned long nr_reclaimed = 0;
3526 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3527 unsigned long reclaimed;
3528 int loop = 0;
3529 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003530 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003531 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003532
3533 if (order > 0)
3534 return 0;
3535
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003536 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003537 /*
3538 * This loop can run a while, specially if mem_cgroup's continuously
3539 * keep exceeding their soft limit and putting the system under
3540 * pressure
3541 */
3542 do {
3543 if (next_mz)
3544 mz = next_mz;
3545 else
3546 mz = mem_cgroup_largest_soft_limit_node(mctz);
3547 if (!mz)
3548 break;
3549
Ying Han0ae5e892011-05-26 16:25:25 -07003550 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003551 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08003552 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003553 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003554 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003555 spin_lock(&mctz->lock);
3556
3557 /*
3558 * If we failed to reclaim anything from this memory cgroup
3559 * it is time to move on to the next cgroup
3560 */
3561 next_mz = NULL;
3562 if (!reclaimed) {
3563 do {
3564 /*
3565 * Loop until we find yet another one.
3566 *
3567 * By the time we get the soft_limit lock
3568 * again, someone might have aded the
3569 * group back on the RB tree. Iterate to
3570 * make sure we get a different mem.
3571 * mem_cgroup_largest_soft_limit_node returns
3572 * NULL if no other cgroup is present on
3573 * the tree
3574 */
3575 next_mz =
3576 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003577 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003578 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003579 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003580 break;
3581 } while (1);
3582 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003583 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3584 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003585 /*
3586 * One school of thought says that we should not add
3587 * back the node to the tree if reclaim returns 0.
3588 * But our reclaim could return 0, simply because due
3589 * to priority we are exposing a smaller subset of
3590 * memory to reclaim from. Consider this as a longer
3591 * term TODO.
3592 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003593 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003594 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003595 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003596 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003597 loop++;
3598 /*
3599 * Could not reclaim anything and there are no more
3600 * mem cgroups to try or we seem to be looping without
3601 * reclaiming anything.
3602 */
3603 if (!nr_reclaimed &&
3604 (next_mz == NULL ||
3605 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3606 break;
3607 } while (!nr_reclaimed);
3608 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003609 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003610 return nr_reclaimed;
3611}
3612
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003613/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003614 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003615 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3616 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003617static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003618 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003619{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003620 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003621 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003622 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003623 struct page *busy;
3624 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003625 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003626
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003627 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003628 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003629 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003630
Hugh Dickins1eb49272012-03-21 16:34:19 -07003631 loop = mz->lru_size[lru];
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003632 /* give some margin against EBUSY etc...*/
3633 loop += 256;
3634 busy = NULL;
3635 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003636 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003637 struct page *page;
3638
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003639 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003640 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003641 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003642 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003643 break;
3644 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003645 page = list_entry(list->prev, struct page, lru);
3646 if (busy == page) {
3647 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003648 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003649 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003650 continue;
3651 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003652 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003653
Johannes Weiner925b7672012-01-12 17:18:15 -08003654 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003655
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003656 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003657 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003658 break;
3659
3660 if (ret == -EBUSY || ret == -EINVAL) {
3661 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003662 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003663 cond_resched();
3664 } else
3665 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003666 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003667
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003668 if (!ret && !list_empty(list))
3669 return -EBUSY;
3670 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003671}
3672
3673/*
3674 * make mem_cgroup's charge to be 0 if there is no task.
3675 * This enables deleting this mem_cgroup.
3676 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003677static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003678{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003679 int ret;
3680 int node, zid, shrink;
3681 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003682 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003683
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003684 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003685
3686 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003687 /* should free all ? */
3688 if (free_all)
3689 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003690move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003691 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003692 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003693 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003694 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003695 ret = -EINTR;
3696 if (signal_pending(current))
3697 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003698 /* This is for making all *used* pages to be on LRU. */
3699 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003700 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003701 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003702 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003703 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003704 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003705 enum lru_list lru;
3706 for_each_lru(lru) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003707 ret = mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003708 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003709 if (ret)
3710 break;
3711 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003712 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003713 if (ret)
3714 break;
3715 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003716 mem_cgroup_end_move(memcg);
3717 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003718 /* it seems parent cgroup doesn't have enough mem */
3719 if (ret == -ENOMEM)
3720 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003721 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003722 /* "ret" should also be checked to ensure all lists are empty. */
Glauber Costa569530f2012-04-12 12:49:13 -07003723 } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003724out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003725 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003726 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003727
3728try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003729 /* returns EBUSY if there is a task or if we come here twice. */
3730 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003731 ret = -EBUSY;
3732 goto out;
3733 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003734 /* we call try-to-free pages for make this cgroup empty */
3735 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003736 /* try to free all pages in this cgroup */
3737 shrink = 1;
Glauber Costa569530f2012-04-12 12:49:13 -07003738 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003739 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003740
3741 if (signal_pending(current)) {
3742 ret = -EINTR;
3743 goto out;
3744 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003745 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003746 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003747 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003748 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003749 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003750 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003751 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003752
3753 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003754 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003755 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003756 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003757}
3758
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07003759static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003760{
3761 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3762}
3763
3764
Balbir Singh18f59ea2009-01-07 18:08:07 -08003765static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3766{
3767 return mem_cgroup_from_cont(cont)->use_hierarchy;
3768}
3769
3770static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3771 u64 val)
3772{
3773 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003774 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003775 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003776 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003777
3778 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003779 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003780
3781 cgroup_lock();
3782 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003783 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003784 * in the child subtrees. If it is unset, then the change can
3785 * occur, provided the current cgroup has no children.
3786 *
3787 * For the root cgroup, parent_mem is NULL, we allow value to be
3788 * set if there are no children.
3789 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003790 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003791 (val == 1 || val == 0)) {
3792 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003793 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003794 else
3795 retval = -EBUSY;
3796 } else
3797 retval = -EINVAL;
3798 cgroup_unlock();
3799
3800 return retval;
3801}
3802
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003803
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003804static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003805 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003806{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003807 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003808 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003809
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003810 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003811 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003812 val += mem_cgroup_read_stat(iter, idx);
3813
3814 if (val < 0) /* race ? */
3815 val = 0;
3816 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003817}
3818
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003819static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003820{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003821 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003822
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003823 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003824 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003825 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003826 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003827 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003828 }
3829
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003830 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3831 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003832
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003833 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003834 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003835
3836 return val << PAGE_SHIFT;
3837}
3838
Tejun Heoaf36f902012-04-01 12:09:55 -07003839static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
3840 struct file *file, char __user *buf,
3841 size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003842{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003843 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Tejun Heoaf36f902012-04-01 12:09:55 -07003844 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003845 u64 val;
Tejun Heoaf36f902012-04-01 12:09:55 -07003846 int type, name, len;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003847
3848 type = MEMFILE_TYPE(cft->private);
3849 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003850
3851 if (!do_swap_account && type == _MEMSWAP)
3852 return -EOPNOTSUPP;
3853
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003854 switch (type) {
3855 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003856 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003857 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003858 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003859 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003860 break;
3861 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003862 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003863 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003864 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003865 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003866 break;
3867 default:
3868 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003869 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003870
3871 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3872 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003873}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003874/*
3875 * The user of this function is...
3876 * RES_LIMIT.
3877 */
Paul Menage856c13a2008-07-25 01:47:04 -07003878static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3879 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003880{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003881 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003882 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003883 unsigned long long val;
3884 int ret;
3885
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003886 type = MEMFILE_TYPE(cft->private);
3887 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07003888
3889 if (!do_swap_account && type == _MEMSWAP)
3890 return -EOPNOTSUPP;
3891
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003892 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003893 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003894 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3895 ret = -EINVAL;
3896 break;
3897 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003898 /* This function does all necessary parse...reuse it */
3899 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003900 if (ret)
3901 break;
3902 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003903 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003904 else
3905 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003906 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003907 case RES_SOFT_LIMIT:
3908 ret = res_counter_memparse_write_strategy(buffer, &val);
3909 if (ret)
3910 break;
3911 /*
3912 * For memsw, soft limits are hard to implement in terms
3913 * of semantics, for now, we support soft limits for
3914 * control without swap
3915 */
3916 if (type == _MEM)
3917 ret = res_counter_set_soft_limit(&memcg->res, val);
3918 else
3919 ret = -EINVAL;
3920 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003921 default:
3922 ret = -EINVAL; /* should be BUG() ? */
3923 break;
3924 }
3925 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003926}
3927
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003928static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3929 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3930{
3931 struct cgroup *cgroup;
3932 unsigned long long min_limit, min_memsw_limit, tmp;
3933
3934 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3935 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3936 cgroup = memcg->css.cgroup;
3937 if (!memcg->use_hierarchy)
3938 goto out;
3939
3940 while (cgroup->parent) {
3941 cgroup = cgroup->parent;
3942 memcg = mem_cgroup_from_cont(cgroup);
3943 if (!memcg->use_hierarchy)
3944 break;
3945 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3946 min_limit = min(min_limit, tmp);
3947 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3948 min_memsw_limit = min(min_memsw_limit, tmp);
3949 }
3950out:
3951 *mem_limit = min_limit;
3952 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003953}
3954
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003955static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003956{
Tejun Heoaf36f902012-04-01 12:09:55 -07003957 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003958 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003959
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003960 type = MEMFILE_TYPE(event);
3961 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07003962
3963 if (!do_swap_account && type == _MEMSWAP)
3964 return -EOPNOTSUPP;
3965
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003966 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003967 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003968 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003969 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003970 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003971 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003972 break;
3973 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003974 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003975 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003976 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003977 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003978 break;
3979 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003980
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07003981 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003982}
3983
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003984static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3985 struct cftype *cft)
3986{
3987 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3988}
3989
Daisuke Nishimura02491442010-03-10 15:22:17 -08003990#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003991static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3992 struct cftype *cft, u64 val)
3993{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003994 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003995
3996 if (val >= (1 << NR_MOVE_TYPE))
3997 return -EINVAL;
3998 /*
3999 * We check this value several times in both in can_attach() and
4000 * attach(), so we need cgroup lock to prevent this value from being
4001 * inconsistent.
4002 */
4003 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004004 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004005 cgroup_unlock();
4006
4007 return 0;
4008}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004009#else
4010static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4011 struct cftype *cft, u64 val)
4012{
4013 return -ENOSYS;
4014}
4015#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004016
Ying Han406eb0c2011-05-26 16:25:37 -07004017#ifdef CONFIG_NUMA
Johannes Weinerfada52c2012-05-29 15:07:06 -07004018static int mem_control_numa_stat_show(struct cgroup *cont, struct cftype *cft,
4019 struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07004020{
4021 int nid;
4022 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4023 unsigned long node_nr;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004024 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07004025
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004026 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004027 seq_printf(m, "total=%lu", total_nr);
4028 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004029 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004030 seq_printf(m, " N%d=%lu", nid, node_nr);
4031 }
4032 seq_putc(m, '\n');
4033
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004034 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004035 seq_printf(m, "file=%lu", file_nr);
4036 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004037 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004038 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004039 seq_printf(m, " N%d=%lu", nid, node_nr);
4040 }
4041 seq_putc(m, '\n');
4042
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004043 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004044 seq_printf(m, "anon=%lu", anon_nr);
4045 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004046 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004047 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004048 seq_printf(m, " N%d=%lu", nid, node_nr);
4049 }
4050 seq_putc(m, '\n');
4051
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004052 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004053 seq_printf(m, "unevictable=%lu", unevictable_nr);
4054 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004055 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004056 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004057 seq_printf(m, " N%d=%lu", nid, node_nr);
4058 }
4059 seq_putc(m, '\n');
4060 return 0;
4061}
4062#endif /* CONFIG_NUMA */
4063
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004064static const char * const mem_cgroup_lru_names[] = {
4065 "inactive_anon",
4066 "active_anon",
4067 "inactive_file",
4068 "active_file",
4069 "unevictable",
4070};
4071
4072static inline void mem_cgroup_lru_names_not_uptodate(void)
4073{
4074 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4075}
4076
Paul Menagec64745c2008-04-29 01:00:02 -07004077static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004078 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004079{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004080 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004081 struct mem_cgroup *mi;
4082 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004083
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004084 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4085 if (i == MEM_CGROUP_STAT_SWAPOUT && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004086 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004087 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4088 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004089 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004090
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004091 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4092 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4093 mem_cgroup_read_events(memcg, i));
4094
4095 for (i = 0; i < NR_LRU_LISTS; i++)
4096 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4097 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4098
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004099 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004100 {
4101 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004102 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004103 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004104 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004105 seq_printf(m, "hierarchical_memsw_limit %llu\n",
4106 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004107 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004108
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004109 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4110 long long val = 0;
4111
4112 if (i == MEM_CGROUP_STAT_SWAPOUT && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004113 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004114 for_each_mem_cgroup_tree(mi, memcg)
4115 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4116 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4117 }
4118
4119 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4120 unsigned long long val = 0;
4121
4122 for_each_mem_cgroup_tree(mi, memcg)
4123 val += mem_cgroup_read_events(mi, i);
4124 seq_printf(m, "total_%s %llu\n",
4125 mem_cgroup_events_names[i], val);
4126 }
4127
4128 for (i = 0; i < NR_LRU_LISTS; i++) {
4129 unsigned long long val = 0;
4130
4131 for_each_mem_cgroup_tree(mi, memcg)
4132 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4133 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004134 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004135
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004136#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004137 {
4138 int nid, zid;
4139 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07004140 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004141 unsigned long recent_rotated[2] = {0, 0};
4142 unsigned long recent_scanned[2] = {0, 0};
4143
4144 for_each_online_node(nid)
4145 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004146 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07004147 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004148
Hugh Dickins89abfab2012-05-29 15:06:53 -07004149 recent_rotated[0] += rstat->recent_rotated[0];
4150 recent_rotated[1] += rstat->recent_rotated[1];
4151 recent_scanned[0] += rstat->recent_scanned[0];
4152 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004153 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004154 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4155 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4156 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4157 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004158 }
4159#endif
4160
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004161 return 0;
4162}
4163
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004164static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4165{
4166 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4167
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004168 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004169}
4170
4171static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4172 u64 val)
4173{
4174 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4175 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004176
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004177 if (val > 100)
4178 return -EINVAL;
4179
4180 if (cgrp->parent == NULL)
4181 return -EINVAL;
4182
4183 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004184
4185 cgroup_lock();
4186
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004187 /* If under hierarchy, only empty-root can set this value */
4188 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004189 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4190 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004191 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004192 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004193
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004194 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004195
Li Zefan068b38c2009-01-15 13:51:26 -08004196 cgroup_unlock();
4197
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004198 return 0;
4199}
4200
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004201static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4202{
4203 struct mem_cgroup_threshold_ary *t;
4204 u64 usage;
4205 int i;
4206
4207 rcu_read_lock();
4208 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004209 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004210 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004211 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004212
4213 if (!t)
4214 goto unlock;
4215
4216 usage = mem_cgroup_usage(memcg, swap);
4217
4218 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004219 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004220 * If it's not true, a threshold was crossed after last
4221 * call of __mem_cgroup_threshold().
4222 */
Phil Carmody5407a562010-05-26 14:42:42 -07004223 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004224
4225 /*
4226 * Iterate backward over array of thresholds starting from
4227 * current_threshold and check if a threshold is crossed.
4228 * If none of thresholds below usage is crossed, we read
4229 * only one element of the array here.
4230 */
4231 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4232 eventfd_signal(t->entries[i].eventfd, 1);
4233
4234 /* i = current_threshold + 1 */
4235 i++;
4236
4237 /*
4238 * Iterate forward over array of thresholds starting from
4239 * current_threshold+1 and check if a threshold is crossed.
4240 * If none of thresholds above usage is crossed, we read
4241 * only one element of the array here.
4242 */
4243 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4244 eventfd_signal(t->entries[i].eventfd, 1);
4245
4246 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004247 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004248unlock:
4249 rcu_read_unlock();
4250}
4251
4252static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4253{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004254 while (memcg) {
4255 __mem_cgroup_threshold(memcg, false);
4256 if (do_swap_account)
4257 __mem_cgroup_threshold(memcg, true);
4258
4259 memcg = parent_mem_cgroup(memcg);
4260 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004261}
4262
4263static int compare_thresholds(const void *a, const void *b)
4264{
4265 const struct mem_cgroup_threshold *_a = a;
4266 const struct mem_cgroup_threshold *_b = b;
4267
4268 return _a->threshold - _b->threshold;
4269}
4270
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004271static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004272{
4273 struct mem_cgroup_eventfd_list *ev;
4274
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004275 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004276 eventfd_signal(ev->eventfd, 1);
4277 return 0;
4278}
4279
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004280static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004281{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004282 struct mem_cgroup *iter;
4283
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004284 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004285 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004286}
4287
4288static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4289 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004290{
4291 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004292 struct mem_cgroup_thresholds *thresholds;
4293 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004294 int type = MEMFILE_TYPE(cft->private);
4295 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004296 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004297
4298 ret = res_counter_memparse_write_strategy(args, &threshold);
4299 if (ret)
4300 return ret;
4301
4302 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004303
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004304 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004305 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004306 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004307 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004308 else
4309 BUG();
4310
4311 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4312
4313 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004314 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004315 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4316
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004317 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004318
4319 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004320 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004321 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004322 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004323 ret = -ENOMEM;
4324 goto unlock;
4325 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004326 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004327
4328 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004329 if (thresholds->primary) {
4330 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004331 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004332 }
4333
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004334 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004335 new->entries[size - 1].eventfd = eventfd;
4336 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004337
4338 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004339 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004340 compare_thresholds, NULL);
4341
4342 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004343 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004344 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004345 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004346 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004347 * new->current_threshold will not be used until
4348 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004349 * it here.
4350 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004351 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004352 } else
4353 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004354 }
4355
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004356 /* Free old spare buffer and save old primary buffer as spare */
4357 kfree(thresholds->spare);
4358 thresholds->spare = thresholds->primary;
4359
4360 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004361
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004362 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004363 synchronize_rcu();
4364
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004365unlock:
4366 mutex_unlock(&memcg->thresholds_lock);
4367
4368 return ret;
4369}
4370
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004371static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004372 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004373{
4374 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004375 struct mem_cgroup_thresholds *thresholds;
4376 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004377 int type = MEMFILE_TYPE(cft->private);
4378 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004379 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004380
4381 mutex_lock(&memcg->thresholds_lock);
4382 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004383 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004384 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004385 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004386 else
4387 BUG();
4388
Anton Vorontsov371528c2012-02-24 05:14:46 +04004389 if (!thresholds->primary)
4390 goto unlock;
4391
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004392 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4393
4394 /* Check if a threshold crossed before removing */
4395 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4396
4397 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004398 size = 0;
4399 for (i = 0; i < thresholds->primary->size; i++) {
4400 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004401 size++;
4402 }
4403
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004404 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004405
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004406 /* Set thresholds array to NULL if we don't have thresholds */
4407 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004408 kfree(new);
4409 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004410 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004411 }
4412
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004413 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004414
4415 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004416 new->current_threshold = -1;
4417 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4418 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004419 continue;
4420
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004421 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004422 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004423 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004424 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004425 * until rcu_assign_pointer(), so it's safe to increment
4426 * it here.
4427 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004428 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004429 }
4430 j++;
4431 }
4432
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004433swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004434 /* Swap primary and spare array */
4435 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004436 /* If all events are unregistered, free the spare array */
4437 if (!new) {
4438 kfree(thresholds->spare);
4439 thresholds->spare = NULL;
4440 }
4441
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004442 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004443
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004444 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004445 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004446unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004447 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004448}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004449
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004450static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4451 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4452{
4453 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4454 struct mem_cgroup_eventfd_list *event;
4455 int type = MEMFILE_TYPE(cft->private);
4456
4457 BUG_ON(type != _OOM_TYPE);
4458 event = kmalloc(sizeof(*event), GFP_KERNEL);
4459 if (!event)
4460 return -ENOMEM;
4461
Michal Hocko1af8efe2011-07-26 16:08:24 -07004462 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004463
4464 event->eventfd = eventfd;
4465 list_add(&event->list, &memcg->oom_notify);
4466
4467 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004468 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004469 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004470 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004471
4472 return 0;
4473}
4474
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004475static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004476 struct cftype *cft, struct eventfd_ctx *eventfd)
4477{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004478 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004479 struct mem_cgroup_eventfd_list *ev, *tmp;
4480 int type = MEMFILE_TYPE(cft->private);
4481
4482 BUG_ON(type != _OOM_TYPE);
4483
Michal Hocko1af8efe2011-07-26 16:08:24 -07004484 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004485
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004486 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004487 if (ev->eventfd == eventfd) {
4488 list_del(&ev->list);
4489 kfree(ev);
4490 }
4491 }
4492
Michal Hocko1af8efe2011-07-26 16:08:24 -07004493 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004494}
4495
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004496static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4497 struct cftype *cft, struct cgroup_map_cb *cb)
4498{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004499 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004500
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004501 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004502
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004503 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004504 cb->fill(cb, "under_oom", 1);
4505 else
4506 cb->fill(cb, "under_oom", 0);
4507 return 0;
4508}
4509
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004510static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4511 struct cftype *cft, u64 val)
4512{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004513 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004514 struct mem_cgroup *parent;
4515
4516 /* cannot set to root cgroup and only 0 and 1 are allowed */
4517 if (!cgrp->parent || !((val == 0) || (val == 1)))
4518 return -EINVAL;
4519
4520 parent = mem_cgroup_from_cont(cgrp->parent);
4521
4522 cgroup_lock();
4523 /* oom-kill-disable is a flag for subhierarchy. */
4524 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004525 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004526 cgroup_unlock();
4527 return -EINVAL;
4528 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004529 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004530 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004531 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004532 cgroup_unlock();
4533 return 0;
4534}
4535
Glauber Costae5671df2011-12-11 21:47:01 +00004536#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03004537static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004538{
Glauber Costa1d62e432012-04-09 19:36:33 -03004539 return mem_cgroup_sockets_init(memcg, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004540};
4541
Glauber Costa1d62e432012-04-09 19:36:33 -03004542static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004543{
Glauber Costa1d62e432012-04-09 19:36:33 -03004544 mem_cgroup_sockets_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004545}
Glauber Costae5671df2011-12-11 21:47:01 +00004546#else
Glauber Costacbe128e32012-04-09 19:36:34 -03004547static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004548{
4549 return 0;
4550}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004551
Glauber Costa1d62e432012-04-09 19:36:33 -03004552static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004553{
4554}
Glauber Costae5671df2011-12-11 21:47:01 +00004555#endif
4556
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004557static struct cftype mem_cgroup_files[] = {
4558 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004559 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004560 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004561 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004562 .register_event = mem_cgroup_usage_register_event,
4563 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004564 },
4565 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004566 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004567 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004568 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004569 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004570 },
4571 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004572 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004573 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004574 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004575 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004576 },
4577 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004578 .name = "soft_limit_in_bytes",
4579 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4580 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004581 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07004582 },
4583 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004584 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004585 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004586 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004587 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004588 },
Balbir Singh8697d332008-02-07 00:13:59 -08004589 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004590 .name = "stat",
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004591 .read_seq_string = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004592 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004593 {
4594 .name = "force_empty",
4595 .trigger = mem_cgroup_force_empty_write,
4596 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004597 {
4598 .name = "use_hierarchy",
4599 .write_u64 = mem_cgroup_hierarchy_write,
4600 .read_u64 = mem_cgroup_hierarchy_read,
4601 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004602 {
4603 .name = "swappiness",
4604 .read_u64 = mem_cgroup_swappiness_read,
4605 .write_u64 = mem_cgroup_swappiness_write,
4606 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004607 {
4608 .name = "move_charge_at_immigrate",
4609 .read_u64 = mem_cgroup_move_charge_read,
4610 .write_u64 = mem_cgroup_move_charge_write,
4611 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004612 {
4613 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004614 .read_map = mem_cgroup_oom_control_read,
4615 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004616 .register_event = mem_cgroup_oom_register_event,
4617 .unregister_event = mem_cgroup_oom_unregister_event,
4618 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4619 },
Ying Han406eb0c2011-05-26 16:25:37 -07004620#ifdef CONFIG_NUMA
4621 {
4622 .name = "numa_stat",
Johannes Weinerfada52c2012-05-29 15:07:06 -07004623 .read_seq_string = mem_control_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004624 },
4625#endif
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004626#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004627 {
4628 .name = "memsw.usage_in_bytes",
4629 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07004630 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004631 .register_event = mem_cgroup_usage_register_event,
4632 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004633 },
4634 {
4635 .name = "memsw.max_usage_in_bytes",
4636 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4637 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004638 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004639 },
4640 {
4641 .name = "memsw.limit_in_bytes",
4642 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4643 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07004644 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004645 },
4646 {
4647 .name = "memsw.failcnt",
4648 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4649 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07004650 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004651 },
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004652#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004653 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004654};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004655
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004656static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004657{
4658 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004659 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004660 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004661 /*
4662 * This routine is called against possible nodes.
4663 * But it's BUG to call kmalloc() against offline node.
4664 *
4665 * TODO: this routine can waste much memory for nodes which will
4666 * never be onlined. It's better to use memory hotplug callback
4667 * function.
4668 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004669 if (!node_state(node, N_NORMAL_MEMORY))
4670 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004671 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004672 if (!pn)
4673 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004674
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004675 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4676 mz = &pn->zoneinfo[zone];
Konstantin Khlebnikov7f5e86c2012-05-29 15:06:58 -07004677 lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004678 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004679 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004680 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004681 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004682 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004683 return 0;
4684}
4685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004686static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004687{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004688 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004689}
4690
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004691static struct mem_cgroup *mem_cgroup_alloc(void)
4692{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004693 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004694 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004695
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004696 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004697 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004698 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004699 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004700 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004701
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004702 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004703 return NULL;
4704
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004705 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4706 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004707 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004708 spin_lock_init(&memcg->pcp_counter_lock);
4709 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004710
4711out_free:
4712 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004713 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004714 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004715 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004716 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004717}
4718
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004719/*
Glauber Costa3afe36b2012-05-29 15:07:10 -07004720 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
Hugh Dickins59927fb2012-03-15 15:17:07 -07004721 * but in process context. The work_freeing structure is overlaid
4722 * on the rcu_freeing structure, which itself is overlaid on memsw.
4723 */
Glauber Costa3afe36b2012-05-29 15:07:10 -07004724static void free_work(struct work_struct *work)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004725{
4726 struct mem_cgroup *memcg;
Glauber Costa3afe36b2012-05-29 15:07:10 -07004727 int size = sizeof(struct mem_cgroup);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004728
4729 memcg = container_of(work, struct mem_cgroup, work_freeing);
Glauber Costa3f134612012-05-29 15:07:11 -07004730 /*
4731 * We need to make sure that (at least for now), the jump label
4732 * destruction code runs outside of the cgroup lock. This is because
4733 * get_online_cpus(), which is called from the static_branch update,
4734 * can't be called inside the cgroup_lock. cpusets are the ones
4735 * enforcing this dependency, so if they ever change, we might as well.
4736 *
4737 * schedule_work() will guarantee this happens. Be careful if you need
4738 * to move this code around, and make sure it is outside
4739 * the cgroup_lock.
4740 */
4741 disarm_sock_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004742 if (size < PAGE_SIZE)
4743 kfree(memcg);
4744 else
4745 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004746}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004747
4748static void free_rcu(struct rcu_head *rcu_head)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004749{
4750 struct mem_cgroup *memcg;
4751
4752 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004753 INIT_WORK(&memcg->work_freeing, free_work);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004754 schedule_work(&memcg->work_freeing);
4755}
4756
4757/*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004758 * At destroying mem_cgroup, references from swap_cgroup can remain.
4759 * (scanning all at force_empty is too costly...)
4760 *
4761 * Instead of clearing all references at force_empty, we remember
4762 * the number of reference from swap_cgroup and free mem_cgroup when
4763 * it goes down to 0.
4764 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004765 * Removal of cgroup itself succeeds regardless of refs from swap.
4766 */
4767
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004768static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004769{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004770 int node;
4771
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004772 mem_cgroup_remove_from_trees(memcg);
4773 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004774
Bob Liu3ed28fa2012-01-12 17:19:04 -08004775 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004776 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004777
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004778 free_percpu(memcg->stat);
Glauber Costa3afe36b2012-05-29 15:07:10 -07004779 call_rcu(&memcg->rcu_freeing, free_rcu);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004780}
4781
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004782static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004783{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004784 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004785}
4786
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004787static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004788{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004789 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4790 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4791 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004792 if (parent)
4793 mem_cgroup_put(parent);
4794 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004795}
4796
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004797static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004798{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004799 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004800}
4801
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004802/*
4803 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4804 */
Glauber Costae1aab162011-12-11 21:47:03 +00004805struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004806{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004807 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004808 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004809 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004810}
Glauber Costae1aab162011-12-11 21:47:03 +00004811EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004812
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004813#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4814static void __init enable_swap_cgroup(void)
4815{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004816 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004817 do_swap_account = 1;
4818}
4819#else
4820static void __init enable_swap_cgroup(void)
4821{
4822}
4823#endif
4824
Balbir Singhf64c3f52009-09-23 15:56:37 -07004825static int mem_cgroup_soft_limit_tree_init(void)
4826{
4827 struct mem_cgroup_tree_per_node *rtpn;
4828 struct mem_cgroup_tree_per_zone *rtpz;
4829 int tmp, node, zone;
4830
Bob Liu3ed28fa2012-01-12 17:19:04 -08004831 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004832 tmp = node;
4833 if (!node_state(node, N_NORMAL_MEMORY))
4834 tmp = -1;
4835 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4836 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004837 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004838
4839 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4840
4841 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4842 rtpz = &rtpn->rb_tree_per_zone[zone];
4843 rtpz->rb_root = RB_ROOT;
4844 spin_lock_init(&rtpz->lock);
4845 }
4846 }
4847 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004848
4849err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004850 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004851 if (!soft_limit_tree.rb_tree_per_node[node])
4852 break;
4853 kfree(soft_limit_tree.rb_tree_per_node[node]);
4854 soft_limit_tree.rb_tree_per_node[node] = NULL;
4855 }
4856 return 1;
4857
Balbir Singhf64c3f52009-09-23 15:56:37 -07004858}
4859
Li Zefan0eb253e2009-01-15 13:51:25 -08004860static struct cgroup_subsys_state * __ref
Li Zefan761b3ef2012-01-31 13:47:36 +08004861mem_cgroup_create(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004862{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004863 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004864 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004865 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004866
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004867 memcg = mem_cgroup_alloc();
4868 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004869 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004870
Bob Liu3ed28fa2012-01-12 17:19:04 -08004871 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004872 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004873 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004874
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004875 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004876 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004877 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004878 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004879 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004880 if (mem_cgroup_soft_limit_tree_init())
4881 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004882 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004883 for_each_possible_cpu(cpu) {
4884 struct memcg_stock_pcp *stock =
4885 &per_cpu(memcg_stock, cpu);
4886 INIT_WORK(&stock->work, drain_local_stock);
4887 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004888 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004889 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004890 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004891 memcg->use_hierarchy = parent->use_hierarchy;
4892 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004893 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004894
Balbir Singh18f59ea2009-01-07 18:08:07 -08004895 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004896 res_counter_init(&memcg->res, &parent->res);
4897 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004898 /*
4899 * We increment refcnt of the parent to ensure that we can
4900 * safely access it on res_counter_charge/uncharge.
4901 * This refcnt will be decremented when freeing this
4902 * mem_cgroup(see mem_cgroup_put).
4903 */
4904 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004905 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004906 res_counter_init(&memcg->res, NULL);
4907 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004908 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004909 memcg->last_scanned_node = MAX_NUMNODES;
4910 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08004911
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004912 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004913 memcg->swappiness = mem_cgroup_swappiness(parent);
4914 atomic_set(&memcg->refcnt, 1);
4915 memcg->move_charge_at_immigrate = 0;
4916 mutex_init(&memcg->thresholds_lock);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07004917 spin_lock_init(&memcg->move_lock);
Glauber Costacbe128e32012-04-09 19:36:34 -03004918
4919 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
4920 if (error) {
4921 /*
4922 * We call put now because our (and parent's) refcnts
4923 * are already in place. mem_cgroup_put() will internally
4924 * call __mem_cgroup_free, so return directly
4925 */
4926 mem_cgroup_put(memcg);
4927 return ERR_PTR(error);
4928 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004929 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004930free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004931 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004932 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004933}
4934
Li Zefan761b3ef2012-01-31 13:47:36 +08004935static int mem_cgroup_pre_destroy(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004936{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004937 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004938
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004939 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004940}
4941
Li Zefan761b3ef2012-01-31 13:47:36 +08004942static void mem_cgroup_destroy(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004943{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004944 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004945
Glauber Costa1d62e432012-04-09 19:36:33 -03004946 kmem_cgroup_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004947
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004948 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004949}
4950
Daisuke Nishimura02491442010-03-10 15:22:17 -08004951#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004952/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004953#define PRECHARGE_COUNT_AT_ONCE 256
4954static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004955{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004956 int ret = 0;
4957 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004958 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004959
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004960 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004961 mc.precharge += count;
4962 /* we don't need css_get for root */
4963 return ret;
4964 }
4965 /* try to charge at once */
4966 if (count > 1) {
4967 struct res_counter *dummy;
4968 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004969 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004970 * by cgroup_lock_live_cgroup() that it is not removed and we
4971 * are still under the same cgroup_mutex. So we can postpone
4972 * css_get().
4973 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004974 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004975 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004976 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004977 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004978 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004979 goto one_by_one;
4980 }
4981 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004982 return ret;
4983 }
4984one_by_one:
4985 /* fall back to one by one charge */
4986 while (count--) {
4987 if (signal_pending(current)) {
4988 ret = -EINTR;
4989 break;
4990 }
4991 if (!batch_count--) {
4992 batch_count = PRECHARGE_COUNT_AT_ONCE;
4993 cond_resched();
4994 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004995 ret = __mem_cgroup_try_charge(NULL,
4996 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004997 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004998 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004999 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005000 mc.precharge++;
5001 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005002 return ret;
5003}
5004
5005/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005006 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005007 * @vma: the vma the pte to be checked belongs
5008 * @addr: the address corresponding to the pte to be checked
5009 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005010 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005011 *
5012 * Returns
5013 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5014 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5015 * move charge. if @target is not NULL, the page is stored in target->page
5016 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005017 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5018 * target for charge migration. if @target is not NULL, the entry is stored
5019 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005020 *
5021 * Called with pte lock held.
5022 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005023union mc_target {
5024 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005025 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005026};
5027
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005028enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005029 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005030 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005031 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005032};
5033
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005034static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5035 unsigned long addr, pte_t ptent)
5036{
5037 struct page *page = vm_normal_page(vma, addr, ptent);
5038
5039 if (!page || !page_mapped(page))
5040 return NULL;
5041 if (PageAnon(page)) {
5042 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005043 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005044 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005045 } else if (!move_file())
5046 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005047 return NULL;
5048 if (!get_page_unless_zero(page))
5049 return NULL;
5050
5051 return page;
5052}
5053
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005054#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005055static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5056 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5057{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005058 struct page *page = NULL;
5059 swp_entry_t ent = pte_to_swp_entry(ptent);
5060
5061 if (!move_anon() || non_swap_entry(ent))
5062 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005063 /*
5064 * Because lookup_swap_cache() updates some statistics counter,
5065 * we call find_get_page() with swapper_space directly.
5066 */
5067 page = find_get_page(&swapper_space, ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005068 if (do_swap_account)
5069 entry->val = ent.val;
5070
5071 return page;
5072}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005073#else
5074static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5075 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5076{
5077 return NULL;
5078}
5079#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005080
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005081static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5082 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5083{
5084 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005085 struct address_space *mapping;
5086 pgoff_t pgoff;
5087
5088 if (!vma->vm_file) /* anonymous vma */
5089 return NULL;
5090 if (!move_file())
5091 return NULL;
5092
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005093 mapping = vma->vm_file->f_mapping;
5094 if (pte_none(ptent))
5095 pgoff = linear_page_index(vma, addr);
5096 else /* pte_file(ptent) is true */
5097 pgoff = pte_to_pgoff(ptent);
5098
5099 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005100 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005101
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005102#ifdef CONFIG_SWAP
5103 /* shmem/tmpfs may report page out on swap: account for that too. */
5104 if (radix_tree_exceptional_entry(page)) {
5105 swp_entry_t swap = radix_to_swp_entry(page);
5106 if (do_swap_account)
5107 *entry = swap;
5108 page = find_get_page(&swapper_space, swap.val);
5109 }
5110#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005111 return page;
5112}
5113
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005114static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005115 unsigned long addr, pte_t ptent, union mc_target *target)
5116{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005117 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005118 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005119 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005120 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005121
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005122 if (pte_present(ptent))
5123 page = mc_handle_present_pte(vma, addr, ptent);
5124 else if (is_swap_pte(ptent))
5125 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005126 else if (pte_none(ptent) || pte_file(ptent))
5127 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005128
5129 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005130 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005131 if (page) {
5132 pc = lookup_page_cgroup(page);
5133 /*
5134 * Do only loose check w/o page_cgroup lock.
5135 * mem_cgroup_move_account() checks the pc is valid or not under
5136 * the lock.
5137 */
5138 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5139 ret = MC_TARGET_PAGE;
5140 if (target)
5141 target->page = page;
5142 }
5143 if (!ret || !target)
5144 put_page(page);
5145 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005146 /* There is a swap entry and a page doesn't exist or isn't charged */
5147 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005148 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005149 ret = MC_TARGET_SWAP;
5150 if (target)
5151 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005152 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005153 return ret;
5154}
5155
Naoya Horiguchi12724852012-03-21 16:34:28 -07005156#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5157/*
5158 * We don't consider swapping or file mapped pages because THP does not
5159 * support them for now.
5160 * Caller should make sure that pmd_trans_huge(pmd) is true.
5161 */
5162static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5163 unsigned long addr, pmd_t pmd, union mc_target *target)
5164{
5165 struct page *page = NULL;
5166 struct page_cgroup *pc;
5167 enum mc_target_type ret = MC_TARGET_NONE;
5168
5169 page = pmd_page(pmd);
5170 VM_BUG_ON(!page || !PageHead(page));
5171 if (!move_anon())
5172 return ret;
5173 pc = lookup_page_cgroup(page);
5174 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5175 ret = MC_TARGET_PAGE;
5176 if (target) {
5177 get_page(page);
5178 target->page = page;
5179 }
5180 }
5181 return ret;
5182}
5183#else
5184static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5185 unsigned long addr, pmd_t pmd, union mc_target *target)
5186{
5187 return MC_TARGET_NONE;
5188}
5189#endif
5190
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005191static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5192 unsigned long addr, unsigned long end,
5193 struct mm_walk *walk)
5194{
5195 struct vm_area_struct *vma = walk->private;
5196 pte_t *pte;
5197 spinlock_t *ptl;
5198
Naoya Horiguchi12724852012-03-21 16:34:28 -07005199 if (pmd_trans_huge_lock(pmd, vma) == 1) {
5200 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5201 mc.precharge += HPAGE_PMD_NR;
5202 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005203 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005204 }
Dave Hansen03319322011-03-22 16:32:56 -07005205
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005206 if (pmd_trans_unstable(pmd))
5207 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005208 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5209 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005210 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005211 mc.precharge++; /* increment precharge temporarily */
5212 pte_unmap_unlock(pte - 1, ptl);
5213 cond_resched();
5214
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005215 return 0;
5216}
5217
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005218static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5219{
5220 unsigned long precharge;
5221 struct vm_area_struct *vma;
5222
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005223 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005224 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5225 struct mm_walk mem_cgroup_count_precharge_walk = {
5226 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5227 .mm = mm,
5228 .private = vma,
5229 };
5230 if (is_vm_hugetlb_page(vma))
5231 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005232 walk_page_range(vma->vm_start, vma->vm_end,
5233 &mem_cgroup_count_precharge_walk);
5234 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005235 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005236
5237 precharge = mc.precharge;
5238 mc.precharge = 0;
5239
5240 return precharge;
5241}
5242
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005243static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5244{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005245 unsigned long precharge = mem_cgroup_count_precharge(mm);
5246
5247 VM_BUG_ON(mc.moving_task);
5248 mc.moving_task = current;
5249 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005250}
5251
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005252/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5253static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005254{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005255 struct mem_cgroup *from = mc.from;
5256 struct mem_cgroup *to = mc.to;
5257
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005258 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005259 if (mc.precharge) {
5260 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5261 mc.precharge = 0;
5262 }
5263 /*
5264 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5265 * we must uncharge here.
5266 */
5267 if (mc.moved_charge) {
5268 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5269 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005270 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005271 /* we must fixup refcnts and charges */
5272 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005273 /* uncharge swap account from the old cgroup */
5274 if (!mem_cgroup_is_root(mc.from))
5275 res_counter_uncharge(&mc.from->memsw,
5276 PAGE_SIZE * mc.moved_swap);
5277 __mem_cgroup_put(mc.from, mc.moved_swap);
5278
5279 if (!mem_cgroup_is_root(mc.to)) {
5280 /*
5281 * we charged both to->res and to->memsw, so we should
5282 * uncharge to->res.
5283 */
5284 res_counter_uncharge(&mc.to->res,
5285 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005286 }
5287 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005288 mc.moved_swap = 0;
5289 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005290 memcg_oom_recover(from);
5291 memcg_oom_recover(to);
5292 wake_up_all(&mc.waitq);
5293}
5294
5295static void mem_cgroup_clear_mc(void)
5296{
5297 struct mem_cgroup *from = mc.from;
5298
5299 /*
5300 * we must clear moving_task before waking up waiters at the end of
5301 * task migration.
5302 */
5303 mc.moving_task = NULL;
5304 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005305 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005306 mc.from = NULL;
5307 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005308 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005309 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005310}
5311
Li Zefan761b3ef2012-01-31 13:47:36 +08005312static int mem_cgroup_can_attach(struct cgroup *cgroup,
5313 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005314{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005315 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005316 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005317 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005318
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005319 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005320 struct mm_struct *mm;
5321 struct mem_cgroup *from = mem_cgroup_from_task(p);
5322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005323 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005324
5325 mm = get_task_mm(p);
5326 if (!mm)
5327 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005328 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005329 if (mm->owner == p) {
5330 VM_BUG_ON(mc.from);
5331 VM_BUG_ON(mc.to);
5332 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005333 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005334 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005335 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005336 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005337 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005338 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005339 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005340 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005341
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005342 ret = mem_cgroup_precharge_mc(mm);
5343 if (ret)
5344 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005345 }
5346 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005347 }
5348 return ret;
5349}
5350
Li Zefan761b3ef2012-01-31 13:47:36 +08005351static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5352 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005353{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005354 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005355}
5356
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005357static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5358 unsigned long addr, unsigned long end,
5359 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005360{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005361 int ret = 0;
5362 struct vm_area_struct *vma = walk->private;
5363 pte_t *pte;
5364 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005365 enum mc_target_type target_type;
5366 union mc_target target;
5367 struct page *page;
5368 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005369
Naoya Horiguchi12724852012-03-21 16:34:28 -07005370 /*
5371 * We don't take compound_lock() here but no race with splitting thp
5372 * happens because:
5373 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5374 * under splitting, which means there's no concurrent thp split,
5375 * - if another thread runs into split_huge_page() just after we
5376 * entered this if-block, the thread must wait for page table lock
5377 * to be unlocked in __split_huge_page_splitting(), where the main
5378 * part of thp split is not executed yet.
5379 */
5380 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005381 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005382 spin_unlock(&vma->vm_mm->page_table_lock);
5383 return 0;
5384 }
5385 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5386 if (target_type == MC_TARGET_PAGE) {
5387 page = target.page;
5388 if (!isolate_lru_page(page)) {
5389 pc = lookup_page_cgroup(page);
5390 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07005391 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005392 mc.precharge -= HPAGE_PMD_NR;
5393 mc.moved_charge += HPAGE_PMD_NR;
5394 }
5395 putback_lru_page(page);
5396 }
5397 put_page(page);
5398 }
5399 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005400 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005401 }
5402
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005403 if (pmd_trans_unstable(pmd))
5404 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005405retry:
5406 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5407 for (; addr != end; addr += PAGE_SIZE) {
5408 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005409 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005410
5411 if (!mc.precharge)
5412 break;
5413
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005414 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005415 case MC_TARGET_PAGE:
5416 page = target.page;
5417 if (isolate_lru_page(page))
5418 goto put;
5419 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005420 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07005421 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005422 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005423 /* we uncharge from mc.from later. */
5424 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005425 }
5426 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005427put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005428 put_page(page);
5429 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005430 case MC_TARGET_SWAP:
5431 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07005432 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005433 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005434 /* we fixup refcnts and charges later. */
5435 mc.moved_swap++;
5436 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005437 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005438 default:
5439 break;
5440 }
5441 }
5442 pte_unmap_unlock(pte - 1, ptl);
5443 cond_resched();
5444
5445 if (addr != end) {
5446 /*
5447 * We have consumed all precharges we got in can_attach().
5448 * We try charge one by one, but don't do any additional
5449 * charges to mc.to if we have failed in charge once in attach()
5450 * phase.
5451 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005452 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005453 if (!ret)
5454 goto retry;
5455 }
5456
5457 return ret;
5458}
5459
5460static void mem_cgroup_move_charge(struct mm_struct *mm)
5461{
5462 struct vm_area_struct *vma;
5463
5464 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005465retry:
5466 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5467 /*
5468 * Someone who are holding the mmap_sem might be waiting in
5469 * waitq. So we cancel all extra charges, wake up all waiters,
5470 * and retry. Because we cancel precharges, we might not be able
5471 * to move enough charges, but moving charge is a best-effort
5472 * feature anyway, so it wouldn't be a big problem.
5473 */
5474 __mem_cgroup_clear_mc();
5475 cond_resched();
5476 goto retry;
5477 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005478 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5479 int ret;
5480 struct mm_walk mem_cgroup_move_charge_walk = {
5481 .pmd_entry = mem_cgroup_move_charge_pte_range,
5482 .mm = mm,
5483 .private = vma,
5484 };
5485 if (is_vm_hugetlb_page(vma))
5486 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005487 ret = walk_page_range(vma->vm_start, vma->vm_end,
5488 &mem_cgroup_move_charge_walk);
5489 if (ret)
5490 /*
5491 * means we have consumed all precharges and failed in
5492 * doing additional charge. Just abandon here.
5493 */
5494 break;
5495 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005496 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005497}
5498
Li Zefan761b3ef2012-01-31 13:47:36 +08005499static void mem_cgroup_move_task(struct cgroup *cont,
5500 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005501{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005502 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005503 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005504
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005505 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005506 if (mc.to)
5507 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005508 mmput(mm);
5509 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005510 if (mc.to)
5511 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005512}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005513#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08005514static int mem_cgroup_can_attach(struct cgroup *cgroup,
5515 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005516{
5517 return 0;
5518}
Li Zefan761b3ef2012-01-31 13:47:36 +08005519static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5520 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005521{
5522}
Li Zefan761b3ef2012-01-31 13:47:36 +08005523static void mem_cgroup_move_task(struct cgroup *cont,
5524 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005525{
5526}
5527#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005528
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005529struct cgroup_subsys mem_cgroup_subsys = {
5530 .name = "memory",
5531 .subsys_id = mem_cgroup_subsys_id,
5532 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005533 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005534 .destroy = mem_cgroup_destroy,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005535 .can_attach = mem_cgroup_can_attach,
5536 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005537 .attach = mem_cgroup_move_task,
Tejun Heo6bc10342012-04-01 12:09:55 -07005538 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005539 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005540 .use_id = 1,
Tejun Heo48ddbe12012-04-01 12:09:56 -07005541 .__DEPRECATED_clear_css_refs = true,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005542};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005543
5544#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005545static int __init enable_swap_account(char *s)
5546{
5547 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005548 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005549 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005550 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005551 really_do_swap_account = 0;
5552 return 1;
5553}
Michal Hockoa2c89902011-05-24 17:12:50 -07005554__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005555
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005556#endif