blob: 7685d4a0b3ce811f187d9dd978282e0a99476a56 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080013 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 */
23
24#include <linux/res_counter.h>
25#include <linux/memcontrol.h>
26#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080027#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080028#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080029#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080030#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070035#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040036#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080037#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070038#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080041#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080043#include <linux/eventfd.h>
44#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080045#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080046#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070047#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070048#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080050#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070051#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080052#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000053#include <net/sock.h>
54#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080055
Balbir Singh8697d332008-02-07 00:13:59 -080056#include <asm/uaccess.h>
57
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070058#include <trace/events/vmscan.h>
59
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070060struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070061#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070062struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070065/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080066int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080067
68/* for remember boot option*/
69#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70static int really_do_swap_account __initdata = 1;
71#else
72static int really_do_swap_account __initdata = 0;
73#endif
74
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075#else
76#define do_swap_account (0)
77#endif
78
79
Balbir Singh8cdea7c2008-02-07 00:13:50 -080080/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080081 * Statistics for memory cgroup.
82 */
83enum mem_cgroup_stat_index {
84 /*
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 */
87 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070088 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080089 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070090 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -070091 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080092 MEM_CGROUP_STAT_NSTATS,
93};
94
Johannes Weinere9f89742011-03-23 16:42:37 -070095enum mem_cgroup_events_index {
96 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
97 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
98 MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
Ying Han456f9982011-05-26 16:25:38 -070099 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
100 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700101 MEM_CGROUP_EVENTS_NSTATS,
102};
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700103/*
104 * Per memcg event counter is incremented at every pagein/pageout. With THP,
105 * it will be incremated by the number of pages. This counter is used for
106 * for trigger some periodic events. This is straightforward and better
107 * than using jiffies etc. to handle periodic memcg event.
108 */
109enum mem_cgroup_events_target {
110 MEM_CGROUP_TARGET_THRESH,
111 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700112 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700113 MEM_CGROUP_NTARGETS,
114};
115#define THRESHOLDS_EVENTS_TARGET (128)
116#define SOFTLIMIT_EVENTS_TARGET (1024)
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700117#define NUMAINFO_EVENTS_TARGET (1024)
Johannes Weinere9f89742011-03-23 16:42:37 -0700118
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800119struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700120 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700121 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700122 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800123};
124
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800125struct mem_cgroup_reclaim_iter {
126 /* css_id of the last scanned hierarchy member */
127 int position;
128 /* scan generation, increased every round-trip */
129 unsigned int generation;
130};
131
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800132/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800133 * per-zone information in memory controller.
134 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800135struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800136 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700137 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800138
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800139 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
140
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800141 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700142 struct rb_node tree_node; /* RB tree node */
143 unsigned long long usage_in_excess;/* Set to the value by which */
144 /* the soft limit is exceeded*/
145 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700146 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700147 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800148};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800149
150struct mem_cgroup_per_node {
151 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
152};
153
154struct mem_cgroup_lru_info {
155 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
156};
157
158/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700159 * Cgroups above their limits are maintained in a RB-Tree, independent of
160 * their hierarchy representation
161 */
162
163struct mem_cgroup_tree_per_zone {
164 struct rb_root rb_root;
165 spinlock_t lock;
166};
167
168struct mem_cgroup_tree_per_node {
169 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
170};
171
172struct mem_cgroup_tree {
173 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
174};
175
176static struct mem_cgroup_tree soft_limit_tree __read_mostly;
177
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800178struct mem_cgroup_threshold {
179 struct eventfd_ctx *eventfd;
180 u64 threshold;
181};
182
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700183/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800184struct mem_cgroup_threshold_ary {
185 /* An array index points to threshold just below usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700186 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800187 /* Size of entries[] */
188 unsigned int size;
189 /* Array of thresholds */
190 struct mem_cgroup_threshold entries[0];
191};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700192
193struct mem_cgroup_thresholds {
194 /* Primary thresholds array */
195 struct mem_cgroup_threshold_ary *primary;
196 /*
197 * Spare threshold array.
198 * This is needed to make mem_cgroup_unregister_event() "never fail".
199 * It must be able to store at least primary->size - 1 entries.
200 */
201 struct mem_cgroup_threshold_ary *spare;
202};
203
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700204/* for OOM */
205struct mem_cgroup_eventfd_list {
206 struct list_head list;
207 struct eventfd_ctx *eventfd;
208};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700210static void mem_cgroup_threshold(struct mem_cgroup *memcg);
211static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800212
Balbir Singhf64c3f52009-09-23 15:56:37 -0700213/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800214 * The memory controller data structure. The memory controller controls both
215 * page cache and RSS per cgroup. We would eventually like to provide
216 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
217 * to help the administrator determine what knobs to tune.
218 *
219 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800220 * we hit the water mark. May be even add a low water mark, such that
221 * no reclaim occurs from a cgroup at it's low water mark, this is
222 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800223 */
224struct mem_cgroup {
225 struct cgroup_subsys_state css;
226 /*
227 * the counter to account for memory usage
228 */
229 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700230
231 union {
232 /*
233 * the counter to account for mem+swap usage.
234 */
235 struct res_counter memsw;
236
237 /*
238 * rcu_freeing is used only when freeing struct mem_cgroup,
239 * so put it into a union to avoid wasting more memory.
240 * It must be disjoint from the css field. It could be
241 * in a union with the res field, but res plays a much
242 * larger part in mem_cgroup life than memsw, and might
243 * be of interest, even at time of free, when debugging.
244 * So share rcu_head with the less interesting memsw.
245 */
246 struct rcu_head rcu_freeing;
247 /*
248 * But when using vfree(), that cannot be done at
249 * interrupt time, so we must then queue the work.
250 */
251 struct work_struct work_freeing;
252 };
253
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800254 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800255 * Per cgroup active and inactive list, similar to the
256 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800257 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800258 struct mem_cgroup_lru_info info;
Ying Han889976d2011-05-26 16:25:33 -0700259 int last_scanned_node;
260#if MAX_NUMNODES > 1
261 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700262 atomic_t numainfo_events;
263 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700264#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800265 /*
266 * Should the accounting and control be hierarchical, per subtree?
267 */
268 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700269
270 bool oom_lock;
271 atomic_t under_oom;
272
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800273 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800274
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700275 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700276 /* OOM-Killer disable */
277 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800278
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700279 /* set when res.limit == memsw.limit */
280 bool memsw_is_minimum;
281
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800282 /* protect arrays of thresholds */
283 struct mutex thresholds_lock;
284
285 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700286 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700287
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800288 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700289 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700290
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700291 /* For oom notifier event fd */
292 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700293
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800294 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800295 * Should we move charges of a task when a task is moved into this
296 * mem_cgroup ? And what type of charges should we move ?
297 */
298 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800299 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700300 * set > 0 if pages under this cgroup are moving to other cgroup.
301 */
302 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700303 /* taken only while moving_account > 0 */
304 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700305 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800306 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800307 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800308 struct mem_cgroup_stat_cpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700309 /*
310 * used when a cpu is offlined or other synchronizations
311 * See mem_cgroup_read_stat().
312 */
313 struct mem_cgroup_stat_cpu nocpu_base;
314 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000315
316#ifdef CONFIG_INET
317 struct tcp_memcontrol tcp_mem;
318#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800319};
320
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800321/* Stuffs for move charges at task migration. */
322/*
323 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
324 * left-shifted bitmap of these types.
325 */
326enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800327 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700328 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800329 NR_MOVE_TYPE,
330};
331
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800332/* "mc" and its members are protected by cgroup_mutex */
333static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800334 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800335 struct mem_cgroup *from;
336 struct mem_cgroup *to;
337 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800338 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800339 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800340 struct task_struct *moving_task; /* a task moving charges */
341 wait_queue_head_t waitq; /* a waitq for other context */
342} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700343 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800344 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
345};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800346
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700347static bool move_anon(void)
348{
349 return test_bit(MOVE_CHARGE_TYPE_ANON,
350 &mc.to->move_charge_at_immigrate);
351}
352
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700353static bool move_file(void)
354{
355 return test_bit(MOVE_CHARGE_TYPE_FILE,
356 &mc.to->move_charge_at_immigrate);
357}
358
Balbir Singh4e416952009-09-23 15:56:39 -0700359/*
360 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
361 * limit reclaim to prevent infinite loops, if they ever occur.
362 */
363#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
364#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
365
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800366enum charge_type {
367 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
368 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700369 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700370 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800371 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700372 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700373 NR_CHARGE_TYPE,
374};
375
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800376/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000377#define _MEM (0)
378#define _MEMSWAP (1)
379#define _OOM_TYPE (2)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800380#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
381#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
382#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700383/* Used for OOM nofiier */
384#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800385
Balbir Singh75822b42009-09-23 15:56:38 -0700386/*
387 * Reclaim flags for mem_cgroup_hierarchical_reclaim
388 */
389#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
390#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
391#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
392#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
393
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700394static void mem_cgroup_get(struct mem_cgroup *memcg);
395static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000396
397/* Writing them here to avoid exposing memcg's inner layout */
398#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae1aab162011-12-11 21:47:03 +0000399#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000400#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000401
402static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
403void sock_update_memcg(struct sock *sk)
404{
Glauber Costa376be5f2012-01-20 04:57:14 +0000405 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000406 struct mem_cgroup *memcg;
407
408 BUG_ON(!sk->sk_prot->proto_cgroup);
409
Glauber Costaf3f511e2012-01-05 20:16:39 +0000410 /* Socket cloning can throw us here with sk_cgrp already
411 * filled. It won't however, necessarily happen from
412 * process context. So the test for root memcg given
413 * the current task's memcg won't help us in this case.
414 *
415 * Respecting the original socket's memcg is a better
416 * decision in this case.
417 */
418 if (sk->sk_cgrp) {
419 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
420 mem_cgroup_get(sk->sk_cgrp->memcg);
421 return;
422 }
423
Glauber Costae1aab162011-12-11 21:47:03 +0000424 rcu_read_lock();
425 memcg = mem_cgroup_from_task(current);
426 if (!mem_cgroup_is_root(memcg)) {
427 mem_cgroup_get(memcg);
428 sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg);
429 }
430 rcu_read_unlock();
431 }
432}
433EXPORT_SYMBOL(sock_update_memcg);
434
435void sock_release_memcg(struct sock *sk)
436{
Glauber Costa376be5f2012-01-20 04:57:14 +0000437 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000438 struct mem_cgroup *memcg;
439 WARN_ON(!sk->sk_cgrp->memcg);
440 memcg = sk->sk_cgrp->memcg;
441 mem_cgroup_put(memcg);
442 }
443}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000444
Glauber Costa319d3b92012-01-15 22:04:39 +0000445#ifdef CONFIG_INET
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000446struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
447{
448 if (!memcg || mem_cgroup_is_root(memcg))
449 return NULL;
450
451 return &memcg->tcp_mem.cg_proto;
452}
453EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000454#endif /* CONFIG_INET */
455#endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
456
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700457static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800458
Balbir Singhf64c3f52009-09-23 15:56:37 -0700459static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700460mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700461{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700462 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700463}
464
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700465struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100466{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700467 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100468}
469
Balbir Singhf64c3f52009-09-23 15:56:37 -0700470static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700471page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700472{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700473 int nid = page_to_nid(page);
474 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700475
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700476 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700477}
478
479static struct mem_cgroup_tree_per_zone *
480soft_limit_tree_node_zone(int nid, int zid)
481{
482 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
483}
484
485static struct mem_cgroup_tree_per_zone *
486soft_limit_tree_from_page(struct page *page)
487{
488 int nid = page_to_nid(page);
489 int zid = page_zonenum(page);
490
491 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
492}
493
494static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700495__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700496 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700497 struct mem_cgroup_tree_per_zone *mctz,
498 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700499{
500 struct rb_node **p = &mctz->rb_root.rb_node;
501 struct rb_node *parent = NULL;
502 struct mem_cgroup_per_zone *mz_node;
503
504 if (mz->on_tree)
505 return;
506
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700507 mz->usage_in_excess = new_usage_in_excess;
508 if (!mz->usage_in_excess)
509 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700510 while (*p) {
511 parent = *p;
512 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
513 tree_node);
514 if (mz->usage_in_excess < mz_node->usage_in_excess)
515 p = &(*p)->rb_left;
516 /*
517 * We can't avoid mem cgroups that are over their soft
518 * limit by the same amount
519 */
520 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
521 p = &(*p)->rb_right;
522 }
523 rb_link_node(&mz->tree_node, parent, p);
524 rb_insert_color(&mz->tree_node, &mctz->rb_root);
525 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700526}
527
528static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700529__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700530 struct mem_cgroup_per_zone *mz,
531 struct mem_cgroup_tree_per_zone *mctz)
532{
533 if (!mz->on_tree)
534 return;
535 rb_erase(&mz->tree_node, &mctz->rb_root);
536 mz->on_tree = false;
537}
538
539static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700540mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700541 struct mem_cgroup_per_zone *mz,
542 struct mem_cgroup_tree_per_zone *mctz)
543{
544 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700545 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700546 spin_unlock(&mctz->lock);
547}
548
Balbir Singhf64c3f52009-09-23 15:56:37 -0700549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700550static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700551{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700552 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700553 struct mem_cgroup_per_zone *mz;
554 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700555 int nid = page_to_nid(page);
556 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700557 mctz = soft_limit_tree_from_page(page);
558
559 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700560 * Necessary to update all ancestors when hierarchy is used.
561 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700562 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700563 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
564 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
565 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700566 /*
567 * We have to update the tree if mz is on RB-tree or
568 * mem is over its softlimit.
569 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700570 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700571 spin_lock(&mctz->lock);
572 /* if on-tree, remove it */
573 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700574 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700575 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700576 * Insert again. mz->usage_in_excess will be updated.
577 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700578 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700579 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700580 spin_unlock(&mctz->lock);
581 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700582 }
583}
584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700585static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700586{
587 int node, zone;
588 struct mem_cgroup_per_zone *mz;
589 struct mem_cgroup_tree_per_zone *mctz;
590
Bob Liu3ed28fa2012-01-12 17:19:04 -0800591 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700592 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700594 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700595 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700596 }
597 }
598}
599
Balbir Singh4e416952009-09-23 15:56:39 -0700600static struct mem_cgroup_per_zone *
601__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
602{
603 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700604 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700605
606retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700607 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700608 rightmost = rb_last(&mctz->rb_root);
609 if (!rightmost)
610 goto done; /* Nothing to reclaim from */
611
612 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
613 /*
614 * Remove the node now but someone else can add it back,
615 * we will to add it back at the end of reclaim to its correct
616 * position in the tree.
617 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700618 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
619 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
620 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700621 goto retry;
622done:
623 return mz;
624}
625
626static struct mem_cgroup_per_zone *
627mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
628{
629 struct mem_cgroup_per_zone *mz;
630
631 spin_lock(&mctz->lock);
632 mz = __mem_cgroup_largest_soft_limit_node(mctz);
633 spin_unlock(&mctz->lock);
634 return mz;
635}
636
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700637/*
638 * Implementation Note: reading percpu statistics for memcg.
639 *
640 * Both of vmstat[] and percpu_counter has threshold and do periodic
641 * synchronization to implement "quick" read. There are trade-off between
642 * reading cost and precision of value. Then, we may have a chance to implement
643 * a periodic synchronizion of counter in memcg's counter.
644 *
645 * But this _read() function is used for user interface now. The user accounts
646 * memory usage by memory cgroup and he _always_ requires exact value because
647 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
648 * have to visit all online cpus and make sum. So, for now, unnecessary
649 * synchronization is not implemented. (just implemented for cpu hotplug)
650 *
651 * If there are kernel internal actions which can make use of some not-exact
652 * value, and reading all cpu value can be performance bottleneck in some
653 * common workload, threashold and synchonization as vmstat[] should be
654 * implemented.
655 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700656static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700657 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800658{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700659 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800660 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800661
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700662 get_online_cpus();
663 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700664 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700665#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700666 spin_lock(&memcg->pcp_counter_lock);
667 val += memcg->nocpu_base.count[idx];
668 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700669#endif
670 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800671 return val;
672}
673
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700674static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700675 bool charge)
676{
677 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700678 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700679}
680
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700681static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700682 enum mem_cgroup_events_index idx)
683{
684 unsigned long val = 0;
685 int cpu;
686
687 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700689#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700690 spin_lock(&memcg->pcp_counter_lock);
691 val += memcg->nocpu_base.events[idx];
692 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700693#endif
694 return val;
695}
696
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700697static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700698 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800699{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800700 preempt_disable();
701
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700702 /*
703 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
704 * counted as CACHE even if it's on ANON LRU.
705 */
706 if (anon)
707 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700708 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800709 else
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -0700710 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700711 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700712
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800713 /* pagein of a big page is an event. So, ignore page size */
714 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700715 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800716 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700717 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800718 nr_pages = -nr_pages; /* for event */
719 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800720
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700721 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800722
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800723 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800724}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800725
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700726unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700727mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700728 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700729{
730 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700731 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700732 unsigned long ret = 0;
733
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700734 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700735
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700736 for_each_lru(lru) {
737 if (BIT(lru) & lru_mask)
738 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700739 }
740 return ret;
741}
742
743static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700744mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700745 int nid, unsigned int lru_mask)
746{
Ying Han889976d2011-05-26 16:25:33 -0700747 u64 total = 0;
748 int zid;
749
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700750 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700751 total += mem_cgroup_zone_nr_lru_pages(memcg,
752 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700753
Ying Han889976d2011-05-26 16:25:33 -0700754 return total;
755}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700756
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700757static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700758 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800759{
Ying Han889976d2011-05-26 16:25:33 -0700760 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800761 u64 total = 0;
762
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700763 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800765 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800766}
767
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800768static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
769 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800770{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700771 unsigned long val, next;
772
Steven Rostedt47994012011-11-02 13:38:33 -0700773 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
774 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700775 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800776 if ((long)next - (long)val < 0) {
777 switch (target) {
778 case MEM_CGROUP_TARGET_THRESH:
779 next = val + THRESHOLDS_EVENTS_TARGET;
780 break;
781 case MEM_CGROUP_TARGET_SOFTLIMIT:
782 next = val + SOFTLIMIT_EVENTS_TARGET;
783 break;
784 case MEM_CGROUP_TARGET_NUMAINFO:
785 next = val + NUMAINFO_EVENTS_TARGET;
786 break;
787 default:
788 break;
789 }
790 __this_cpu_write(memcg->stat->targets[target], next);
791 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700792 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800793 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800794}
795
796/*
797 * Check events in order.
798 *
799 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700800static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800801{
Steven Rostedt47994012011-11-02 13:38:33 -0700802 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800803 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800804 if (unlikely(mem_cgroup_event_ratelimit(memcg,
805 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800806 bool do_softlimit;
807 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800808
809 do_softlimit = mem_cgroup_event_ratelimit(memcg,
810 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700811#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800812 do_numainfo = mem_cgroup_event_ratelimit(memcg,
813 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700814#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800815 preempt_enable();
816
817 mem_cgroup_threshold(memcg);
818 if (unlikely(do_softlimit))
819 mem_cgroup_update_tree(memcg, page);
820#if MAX_NUMNODES > 1
821 if (unlikely(do_numainfo))
822 atomic_inc(&memcg->numainfo_events);
823#endif
824 } else
825 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800826}
827
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000828struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800829{
830 return container_of(cgroup_subsys_state(cont,
831 mem_cgroup_subsys_id), struct mem_cgroup,
832 css);
833}
834
Balbir Singhcf475ad2008-04-29 01:00:16 -0700835struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800836{
Balbir Singh31a78f22008-09-28 23:09:31 +0100837 /*
838 * mm_update_next_owner() may clear mm->owner to NULL
839 * if it races with swapoff, page migration, etc.
840 * So this can be called with p == NULL.
841 */
842 if (unlikely(!p))
843 return NULL;
844
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800845 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
846 struct mem_cgroup, css);
847}
848
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700849struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800850{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700851 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700852
853 if (!mm)
854 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800855 /*
856 * Because we have no locks, mm->owner's may be being moved to other
857 * cgroup. We use css_tryget() here even if this looks
858 * pessimistic (rather than adding locks here).
859 */
860 rcu_read_lock();
861 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700862 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
863 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800864 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700865 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800866 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700867 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800868}
869
Johannes Weiner56600482012-01-12 17:17:59 -0800870/**
871 * mem_cgroup_iter - iterate over memory cgroup hierarchy
872 * @root: hierarchy root
873 * @prev: previously returned memcg, NULL on first invocation
874 * @reclaim: cookie for shared reclaim walks, NULL for full walks
875 *
876 * Returns references to children of the hierarchy below @root, or
877 * @root itself, or %NULL after a full round-trip.
878 *
879 * Caller must pass the return value in @prev on subsequent
880 * invocations for reference counting, or use mem_cgroup_iter_break()
881 * to cancel a hierarchy walk before the round-trip is complete.
882 *
883 * Reclaimers can specify a zone and a priority level in @reclaim to
884 * divide up the memcgs in the hierarchy among all concurrent
885 * reclaimers operating on the same zone and priority.
886 */
887struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
888 struct mem_cgroup *prev,
889 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700890{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800891 struct mem_cgroup *memcg = NULL;
892 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700893
Johannes Weiner56600482012-01-12 17:17:59 -0800894 if (mem_cgroup_disabled())
895 return NULL;
896
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700897 if (!root)
898 root = root_mem_cgroup;
899
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800900 if (prev && !reclaim)
901 id = css_id(&prev->css);
902
903 if (prev && prev != root)
904 css_put(&prev->css);
905
906 if (!root->use_hierarchy && root != root_mem_cgroup) {
907 if (prev)
908 return NULL;
909 return root;
910 }
911
912 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800913 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800914 struct cgroup_subsys_state *css;
915
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800916 if (reclaim) {
917 int nid = zone_to_nid(reclaim->zone);
918 int zid = zone_idx(reclaim->zone);
919 struct mem_cgroup_per_zone *mz;
920
921 mz = mem_cgroup_zoneinfo(root, nid, zid);
922 iter = &mz->reclaim_iter[reclaim->priority];
923 if (prev && reclaim->generation != iter->generation)
924 return NULL;
925 id = iter->position;
926 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800927
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700928 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800929 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
930 if (css) {
931 if (css == &root->css || css_tryget(css))
932 memcg = container_of(css,
933 struct mem_cgroup, css);
934 } else
935 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700936 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700937
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800938 if (reclaim) {
939 iter->position = id;
940 if (!css)
941 iter->generation++;
942 else if (!prev && memcg)
943 reclaim->generation = iter->generation;
944 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800945
946 if (prev && !css)
947 return NULL;
948 }
949 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700950}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800951
Johannes Weiner56600482012-01-12 17:17:59 -0800952/**
953 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
954 * @root: hierarchy root
955 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
956 */
957void mem_cgroup_iter_break(struct mem_cgroup *root,
958 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800959{
960 if (!root)
961 root = root_mem_cgroup;
962 if (prev && prev != root)
963 css_put(&prev->css);
964}
965
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700966/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800967 * Iteration constructs for visiting all cgroups (under a tree). If
968 * loops are exited prematurely (break), mem_cgroup_iter_break() must
969 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700970 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800971#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800972 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800973 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800974 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700975
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800976#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800977 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800978 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800979 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700980
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700981static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700982{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700983 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700984}
985
Ying Han456f9982011-05-26 16:25:38 -0700986void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
987{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700988 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700989
990 if (!mm)
991 return;
992
993 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700994 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
995 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700996 goto out;
997
998 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -0700999 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001000 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1001 break;
1002 case PGMAJFAULT:
1003 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001004 break;
1005 default:
1006 BUG();
1007 }
1008out:
1009 rcu_read_unlock();
1010}
1011EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1012
Johannes Weiner925b7672012-01-12 17:18:15 -08001013/**
1014 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1015 * @zone: zone of the wanted lruvec
1016 * @mem: memcg of the wanted lruvec
1017 *
1018 * Returns the lru list vector holding pages for the given @zone and
1019 * @mem. This can be the global zone lruvec, if the memory controller
1020 * is disabled.
1021 */
1022struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1023 struct mem_cgroup *memcg)
1024{
1025 struct mem_cgroup_per_zone *mz;
1026
1027 if (mem_cgroup_disabled())
1028 return &zone->lruvec;
1029
1030 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1031 return &mz->lruvec;
1032}
1033
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001034/*
1035 * Following LRU functions are allowed to be used without PCG_LOCK.
1036 * Operations are called by routine of global LRU independently from memcg.
1037 * What we have to take care of here is validness of pc->mem_cgroup.
1038 *
1039 * Changes to pc->mem_cgroup happens when
1040 * 1. charge
1041 * 2. moving account
1042 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1043 * It is added to LRU before charge.
1044 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1045 * When moving account, the page is not on LRU. It's isolated.
1046 */
1047
Johannes Weiner925b7672012-01-12 17:18:15 -08001048/**
1049 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1050 * @zone: zone of the page
1051 * @page: the page
1052 * @lru: current lru
1053 *
1054 * This function accounts for @page being added to @lru, and returns
1055 * the lruvec for the given @zone and the memcg @page is charged to.
1056 *
1057 * The callsite is then responsible for physically linking the page to
1058 * the returned lruvec->lists[@lru].
Minchan Kim3f58a822011-03-22 16:32:53 -07001059 */
Johannes Weiner925b7672012-01-12 17:18:15 -08001060struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page,
1061 enum lru_list lru)
Minchan Kim3f58a822011-03-22 16:32:53 -07001062{
1063 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001064 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001065 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001066
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001067 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001068 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001069
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001070 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001071 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001072
1073 /*
1074 * Surreptitiously switch any uncharged page to root:
1075 * an uncharged page off lru does nothing to secure
1076 * its former mem_cgroup from sudden removal.
1077 *
1078 * Our caller holds lru_lock, and PageCgroupUsed is updated
1079 * under page_cgroup lock: between them, they make all uses
1080 * of pc->mem_cgroup safe.
1081 */
1082 if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1083 pc->mem_cgroup = memcg = root_mem_cgroup;
1084
Johannes Weiner925b7672012-01-12 17:18:15 -08001085 mz = page_cgroup_zoneinfo(memcg, page);
1086 /* compound_order() is stabilized through lru_lock */
Hugh Dickins1eb49272012-03-21 16:34:19 -07001087 mz->lru_size[lru] += 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001088 return &mz->lruvec;
1089}
1090
1091/**
1092 * mem_cgroup_lru_del_list - account for removing an lru page
1093 * @page: the page
1094 * @lru: target lru
1095 *
1096 * This function accounts for @page being removed from @lru.
1097 *
1098 * The callsite is then responsible for physically unlinking
1099 * @page->lru.
1100 */
1101void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru)
1102{
1103 struct mem_cgroup_per_zone *mz;
1104 struct mem_cgroup *memcg;
1105 struct page_cgroup *pc;
1106
1107 if (mem_cgroup_disabled())
1108 return;
1109
1110 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001111 memcg = pc->mem_cgroup;
1112 VM_BUG_ON(!memcg);
Johannes Weiner925b7672012-01-12 17:18:15 -08001113 mz = page_cgroup_zoneinfo(memcg, page);
1114 /* huge page split is done under lru_lock. so, we have no races. */
Hugh Dickins1eb49272012-03-21 16:34:19 -07001115 VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page)));
1116 mz->lru_size[lru] -= 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001117}
1118
1119void mem_cgroup_lru_del(struct page *page)
1120{
1121 mem_cgroup_lru_del_list(page, page_lru(page));
1122}
1123
1124/**
1125 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1126 * @zone: zone of the page
1127 * @page: the page
1128 * @from: current lru
1129 * @to: target lru
1130 *
1131 * This function accounts for @page being moved between the lrus @from
1132 * and @to, and returns the lruvec for the given @zone and the memcg
1133 * @page is charged to.
1134 *
1135 * The callsite is then responsible for physically relinking
1136 * @page->lru to the returned lruvec->lists[@to].
1137 */
1138struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone,
1139 struct page *page,
1140 enum lru_list from,
1141 enum lru_list to)
1142{
1143 /* XXX: Optimize this, especially for @from == @to */
1144 mem_cgroup_lru_del_list(page, from);
1145 return mem_cgroup_lru_add_list(zone, page, to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001146}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001147
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001148/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001149 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001150 * hierarchy subtree
1151 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001152static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1153 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001154{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001155 if (root_memcg != memcg) {
1156 return (root_memcg->use_hierarchy &&
1157 css_is_ancestor(&memcg->css, &root_memcg->css));
Michal Hocko3e920412011-07-26 16:08:29 -07001158 }
1159
1160 return true;
1161}
1162
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001163int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001164{
1165 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001166 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001167 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001168
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001169 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001170 if (p) {
1171 curr = try_get_mem_cgroup_from_mm(p->mm);
1172 task_unlock(p);
1173 } else {
1174 /*
1175 * All threads may have already detached their mm's, but the oom
1176 * killer still needs to detect if they have already been oom
1177 * killed to prevent needlessly killing additional tasks.
1178 */
1179 task_lock(task);
1180 curr = mem_cgroup_from_task(task);
1181 if (curr)
1182 css_get(&curr->css);
1183 task_unlock(task);
1184 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001185 if (!curr)
1186 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001187 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001188 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001189 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001190 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1191 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001192 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001193 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001194 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001195 return ret;
1196}
1197
Johannes Weiner9b272972011-11-02 13:38:23 -07001198int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001199{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001200 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001201 int nid = zone_to_nid(zone);
1202 int zid = zone_idx(zone);
1203 unsigned long inactive;
1204 unsigned long active;
1205 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001206
Johannes Weiner9b272972011-11-02 13:38:23 -07001207 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1208 BIT(LRU_INACTIVE_ANON));
1209 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1210 BIT(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
Johannes Weiner9b272972011-11-02 13:38:23 -07001221int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001222{
1223 unsigned long active;
1224 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001225 int zid = zone_idx(zone);
1226 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001227
Johannes Weiner9b272972011-11-02 13:38:23 -07001228 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1229 BIT(LRU_INACTIVE_FILE));
1230 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1231 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001232
1233 return (active > inactive);
1234}
1235
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001236struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1237 struct zone *zone)
1238{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001239 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001240 int zid = zone_idx(zone);
1241 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1242
1243 return &mz->reclaim_stat;
1244}
1245
1246struct zone_reclaim_stat *
1247mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1248{
1249 struct page_cgroup *pc;
1250 struct mem_cgroup_per_zone *mz;
1251
1252 if (mem_cgroup_disabled())
1253 return NULL;
1254
1255 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001256 if (!PageCgroupUsed(pc))
1257 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001258 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1259 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001260 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001261 return &mz->reclaim_stat;
1262}
1263
Balbir Singh6d61ef42009-01-07 18:08:06 -08001264#define mem_cgroup_from_res_counter(counter, member) \
1265 container_of(counter, struct mem_cgroup, member)
1266
Johannes Weiner19942822011-02-01 15:52:43 -08001267/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001268 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1269 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001270 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001271 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001272 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001273 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001274static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001275{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001276 unsigned long long margin;
1277
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001278 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001279 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001280 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001281 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001282}
1283
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001284int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001285{
1286 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001287
1288 /* root ? */
1289 if (cgrp->parent == NULL)
1290 return vm_swappiness;
1291
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001292 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001293}
1294
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001295/*
1296 * memcg->moving_account is used for checking possibility that some thread is
1297 * calling move_account(). When a thread on CPU-A starts moving pages under
1298 * a memcg, other threads should check memcg->moving_account under
1299 * rcu_read_lock(), like this:
1300 *
1301 * CPU-A CPU-B
1302 * rcu_read_lock()
1303 * memcg->moving_account+1 if (memcg->mocing_account)
1304 * take heavy locks.
1305 * synchronize_rcu() update something.
1306 * rcu_read_unlock()
1307 * start move here.
1308 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001309
1310/* for quick checking without looking up memcg */
1311atomic_t memcg_moving __read_mostly;
1312
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001313static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001314{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001315 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001316 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001317 synchronize_rcu();
1318}
1319
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001320static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001321{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001322 /*
1323 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1324 * We check NULL in callee rather than caller.
1325 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001326 if (memcg) {
1327 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001328 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001329 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001330}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001331
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001332/*
1333 * 2 routines for checking "mem" is under move_account() or not.
1334 *
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001335 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1336 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001337 * pc->mem_cgroup may be overwritten.
1338 *
1339 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1340 * under hierarchy of moving cgroups. This is for
1341 * waiting at hith-memory prressure caused by "move".
1342 */
1343
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001344static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001345{
1346 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001347 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001348}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001349
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001350static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001351{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001352 struct mem_cgroup *from;
1353 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001354 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001355 /*
1356 * Unlike task_move routines, we access mc.to, mc.from not under
1357 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1358 */
1359 spin_lock(&mc.lock);
1360 from = mc.from;
1361 to = mc.to;
1362 if (!from)
1363 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001364
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001365 ret = mem_cgroup_same_or_subtree(memcg, from)
1366 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001367unlock:
1368 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001369 return ret;
1370}
1371
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001372static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001373{
1374 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001375 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001376 DEFINE_WAIT(wait);
1377 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1378 /* moving charge context might have finished. */
1379 if (mc.moving_task)
1380 schedule();
1381 finish_wait(&mc.waitq, &wait);
1382 return true;
1383 }
1384 }
1385 return false;
1386}
1387
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001388/*
1389 * Take this lock when
1390 * - a code tries to modify page's memcg while it's USED.
1391 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001392 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001393 */
1394static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1395 unsigned long *flags)
1396{
1397 spin_lock_irqsave(&memcg->move_lock, *flags);
1398}
1399
1400static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1401 unsigned long *flags)
1402{
1403 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1404}
1405
Balbir Singhe2224322009-04-02 16:57:39 -07001406/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001407 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001408 * @memcg: The memory cgroup that went over limit
1409 * @p: Task that is going to be killed
1410 *
1411 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1412 * enabled
1413 */
1414void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1415{
1416 struct cgroup *task_cgrp;
1417 struct cgroup *mem_cgrp;
1418 /*
1419 * Need a buffer in BSS, can't rely on allocations. The code relies
1420 * on the assumption that OOM is serialized for memory controller.
1421 * If this assumption is broken, revisit this code.
1422 */
1423 static char memcg_name[PATH_MAX];
1424 int ret;
1425
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001426 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001427 return;
1428
Balbir Singhe2224322009-04-02 16:57:39 -07001429 rcu_read_lock();
1430
1431 mem_cgrp = memcg->css.cgroup;
1432 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1433
1434 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1435 if (ret < 0) {
1436 /*
1437 * Unfortunately, we are unable to convert to a useful name
1438 * But we'll still print out the usage information
1439 */
1440 rcu_read_unlock();
1441 goto done;
1442 }
1443 rcu_read_unlock();
1444
1445 printk(KERN_INFO "Task in %s killed", memcg_name);
1446
1447 rcu_read_lock();
1448 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1449 if (ret < 0) {
1450 rcu_read_unlock();
1451 goto done;
1452 }
1453 rcu_read_unlock();
1454
1455 /*
1456 * Continues from above, so we don't need an KERN_ level
1457 */
1458 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1459done:
1460
1461 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1462 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1463 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1464 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1465 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1466 "failcnt %llu\n",
1467 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1468 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1469 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1470}
1471
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001472/*
1473 * This function returns the number of memcg under hierarchy tree. Returns
1474 * 1(self count) if no children.
1475 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001476static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001477{
1478 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001479 struct mem_cgroup *iter;
1480
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001481 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001482 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001483 return num;
1484}
1485
Balbir Singh6d61ef42009-01-07 18:08:06 -08001486/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001487 * Return the memory (and swap, if configured) limit for a memcg.
1488 */
1489u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1490{
1491 u64 limit;
1492 u64 memsw;
1493
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001494 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1495 limit += total_swap_pages << PAGE_SHIFT;
1496
David Rientjesa63d83f2010-08-09 17:19:46 -07001497 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1498 /*
1499 * If memsw is finite and limits the amount of swap space available
1500 * to this memcg, return that limit.
1501 */
1502 return min(limit, memsw);
1503}
1504
Johannes Weiner56600482012-01-12 17:17:59 -08001505static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1506 gfp_t gfp_mask,
1507 unsigned long flags)
1508{
1509 unsigned long total = 0;
1510 bool noswap = false;
1511 int loop;
1512
1513 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1514 noswap = true;
1515 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1516 noswap = true;
1517
1518 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1519 if (loop)
1520 drain_all_stock_async(memcg);
1521 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1522 /*
1523 * Allow limit shrinkers, which are triggered directly
1524 * by userspace, to catch signals and stop reclaim
1525 * after minimal progress, regardless of the margin.
1526 */
1527 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1528 break;
1529 if (mem_cgroup_margin(memcg))
1530 break;
1531 /*
1532 * If nothing was reclaimed after two attempts, there
1533 * may be no reclaimable pages in this hierarchy.
1534 */
1535 if (loop && !total)
1536 break;
1537 }
1538 return total;
1539}
1540
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001541/**
1542 * test_mem_cgroup_node_reclaimable
1543 * @mem: the target memcg
1544 * @nid: the node ID to be checked.
1545 * @noswap : specify true here if the user wants flle only information.
1546 *
1547 * This function returns whether the specified memcg contains any
1548 * reclaimable pages on a node. Returns true if there are any reclaimable
1549 * pages in the node.
1550 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001551static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001552 int nid, bool noswap)
1553{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001554 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001555 return true;
1556 if (noswap || !total_swap_pages)
1557 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001558 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001559 return true;
1560 return false;
1561
1562}
Ying Han889976d2011-05-26 16:25:33 -07001563#if MAX_NUMNODES > 1
1564
1565/*
1566 * Always updating the nodemask is not very good - even if we have an empty
1567 * list or the wrong list here, we can start from some node and traverse all
1568 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1569 *
1570 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001571static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001572{
1573 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001574 /*
1575 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1576 * pagein/pageout changes since the last update.
1577 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001578 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001579 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001580 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001581 return;
1582
Ying Han889976d2011-05-26 16:25:33 -07001583 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001584 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001585
1586 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1587
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001588 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1589 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001590 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001591
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001592 atomic_set(&memcg->numainfo_events, 0);
1593 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001594}
1595
1596/*
1597 * Selecting a node where we start reclaim from. Because what we need is just
1598 * reducing usage counter, start from anywhere is O,K. Considering
1599 * memory reclaim from current node, there are pros. and cons.
1600 *
1601 * Freeing memory from current node means freeing memory from a node which
1602 * we'll use or we've used. So, it may make LRU bad. And if several threads
1603 * hit limits, it will see a contention on a node. But freeing from remote
1604 * node means more costs for memory reclaim because of memory latency.
1605 *
1606 * Now, we use round-robin. Better algorithm is welcomed.
1607 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001608int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001609{
1610 int node;
1611
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001612 mem_cgroup_may_update_nodemask(memcg);
1613 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001614
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001615 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001616 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001617 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001618 /*
1619 * We call this when we hit limit, not when pages are added to LRU.
1620 * No LRU may hold pages because all pages are UNEVICTABLE or
1621 * memcg is too small and all pages are not on LRU. In that case,
1622 * we use curret node.
1623 */
1624 if (unlikely(node == MAX_NUMNODES))
1625 node = numa_node_id();
1626
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001628 return node;
1629}
1630
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001631/*
1632 * Check all nodes whether it contains reclaimable pages or not.
1633 * For quick scan, we make use of scan_nodes. This will allow us to skip
1634 * unused nodes. But scan_nodes is lazily updated and may not cotain
1635 * enough new information. We need to do double check.
1636 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001637bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001638{
1639 int nid;
1640
1641 /*
1642 * quick check...making use of scan_node.
1643 * We can skip unused nodes.
1644 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001645 if (!nodes_empty(memcg->scan_nodes)) {
1646 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001647 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001648 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001649
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001650 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001651 return true;
1652 }
1653 }
1654 /*
1655 * Check rest of nodes.
1656 */
1657 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001658 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001659 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001660 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001661 return true;
1662 }
1663 return false;
1664}
1665
Ying Han889976d2011-05-26 16:25:33 -07001666#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001667int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001668{
1669 return 0;
1670}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001671
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001672bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001673{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001674 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001675}
Ying Han889976d2011-05-26 16:25:33 -07001676#endif
1677
Johannes Weiner56600482012-01-12 17:17:59 -08001678static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1679 struct zone *zone,
1680 gfp_t gfp_mask,
1681 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001682{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001683 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001684 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001685 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001686 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001687 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001688 struct mem_cgroup_reclaim_cookie reclaim = {
1689 .zone = zone,
1690 .priority = 0,
1691 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001692
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001693 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001694
Balbir Singh4e416952009-09-23 15:56:39 -07001695 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001696 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001697 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001698 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001699 if (loop >= 2) {
1700 /*
1701 * If we have not been able to reclaim
1702 * anything, it might because there are
1703 * no reclaimable pages under this hierarchy
1704 */
Johannes Weiner56600482012-01-12 17:17:59 -08001705 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001706 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001707 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001708 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001709 * excess >> 2 is not to excessive so as to
1710 * reclaim too much, nor too less that we keep
1711 * coming back to reclaim from this cgroup
1712 */
1713 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001714 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001715 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001716 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001717 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001718 }
Johannes Weiner56600482012-01-12 17:17:59 -08001719 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001720 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001721 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1722 zone, &nr_scanned);
1723 *total_scanned += nr_scanned;
1724 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001725 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001726 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001727 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001728 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001729}
1730
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001731/*
1732 * Check OOM-Killer is already running under our hierarchy.
1733 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001734 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001735 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001736static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001737{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001738 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001739
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001740 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001741 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001742 /*
1743 * this subtree of our hierarchy is already locked
1744 * so we cannot give a lock.
1745 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001746 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001747 mem_cgroup_iter_break(memcg, iter);
1748 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001749 } else
1750 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001751 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001752
Michal Hocko79dfdac2011-07-26 16:08:23 -07001753 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001754 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001755
1756 /*
1757 * OK, we failed to lock the whole subtree so we have to clean up
1758 * what we set up to the failing subtree
1759 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001760 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001761 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001762 mem_cgroup_iter_break(memcg, iter);
1763 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001764 }
1765 iter->oom_lock = false;
1766 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001767 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001768}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001769
Michal Hocko79dfdac2011-07-26 16:08:23 -07001770/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001771 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001772 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001773static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001774{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001775 struct mem_cgroup *iter;
1776
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001777 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001778 iter->oom_lock = false;
1779 return 0;
1780}
1781
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001782static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001783{
1784 struct mem_cgroup *iter;
1785
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001786 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001787 atomic_inc(&iter->under_oom);
1788}
1789
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001790static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001791{
1792 struct mem_cgroup *iter;
1793
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001794 /*
1795 * When a new child is created while the hierarchy is under oom,
1796 * mem_cgroup_oom_lock() may not be called. We have to use
1797 * atomic_add_unless() here.
1798 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001799 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001800 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001801}
1802
Michal Hocko1af8efe2011-07-26 16:08:24 -07001803static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001804static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1805
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001806struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001807 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001808 wait_queue_t wait;
1809};
1810
1811static int memcg_oom_wake_function(wait_queue_t *wait,
1812 unsigned mode, int sync, void *arg)
1813{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001814 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1815 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001816 struct oom_wait_info *oom_wait_info;
1817
1818 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001819 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001820
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001821 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001822 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001823 * Then we can use css_is_ancestor without taking care of RCU.
1824 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001825 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1826 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001827 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001828 return autoremove_wake_function(wait, mode, sync, arg);
1829}
1830
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001831static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001832{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001833 /* for filtering, pass "memcg" as argument. */
1834 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001835}
1836
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001837static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001838{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001839 if (memcg && atomic_read(&memcg->under_oom))
1840 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001841}
1842
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001843/*
1844 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1845 */
David Rientjese845e192012-03-21 16:34:10 -07001846bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001847{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001848 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001849 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001850
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001851 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001852 owait.wait.flags = 0;
1853 owait.wait.func = memcg_oom_wake_function;
1854 owait.wait.private = current;
1855 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001856 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001857 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001858
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001859 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001860 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001861 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001862 /*
1863 * Even if signal_pending(), we can't quit charge() loop without
1864 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1865 * under OOM is always welcomed, use TASK_KILLABLE here.
1866 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001867 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001868 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001869 need_to_kill = false;
1870 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001871 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001872 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001873
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001874 if (need_to_kill) {
1875 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07001876 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001877 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001878 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001879 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001880 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001881 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001882 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001883 mem_cgroup_oom_unlock(memcg);
1884 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001885 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001886
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001887 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001888
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001889 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1890 return false;
1891 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001892 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001893 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001894}
1895
Balbir Singhd69b0422009-06-17 16:26:34 -07001896/*
1897 * Currently used to update mapped file statistics, but the routine can be
1898 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001899 *
1900 * Notes: Race condition
1901 *
1902 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1903 * it tends to be costly. But considering some conditions, we doesn't need
1904 * to do so _always_.
1905 *
1906 * Considering "charge", lock_page_cgroup() is not required because all
1907 * file-stat operations happen after a page is attached to radix-tree. There
1908 * are no race with "charge".
1909 *
1910 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1911 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1912 * if there are race with "uncharge". Statistics itself is properly handled
1913 * by flags.
1914 *
1915 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001916 * small, we check mm->moving_account and detect there are possibility of race
1917 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001918 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001919
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001920void __mem_cgroup_begin_update_page_stat(struct page *page,
1921 bool *locked, unsigned long *flags)
1922{
1923 struct mem_cgroup *memcg;
1924 struct page_cgroup *pc;
1925
1926 pc = lookup_page_cgroup(page);
1927again:
1928 memcg = pc->mem_cgroup;
1929 if (unlikely(!memcg || !PageCgroupUsed(pc)))
1930 return;
1931 /*
1932 * If this memory cgroup is not under account moving, we don't
1933 * need to take move_lock_page_cgroup(). Because we already hold
1934 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001935 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001936 */
Andrew Morton13fd1dd2012-03-21 16:34:26 -07001937 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001938 return;
1939
1940 move_lock_mem_cgroup(memcg, flags);
1941 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
1942 move_unlock_mem_cgroup(memcg, flags);
1943 goto again;
1944 }
1945 *locked = true;
1946}
1947
1948void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
1949{
1950 struct page_cgroup *pc = lookup_page_cgroup(page);
1951
1952 /*
1953 * It's guaranteed that pc->mem_cgroup never changes while
1954 * lock is held because a routine modifies pc->mem_cgroup
1955 * should take move_lock_page_cgroup().
1956 */
1957 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1958}
1959
Greg Thelen2a7106f2011-01-13 15:47:37 -08001960void mem_cgroup_update_page_stat(struct page *page,
1961 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001962{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001963 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001964 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001965 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001966
Johannes Weinercfa44942012-01-12 17:18:38 -08001967 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001968 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001969
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001970 memcg = pc->mem_cgroup;
1971 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001972 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001973
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001974 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001975 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08001976 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001977 break;
1978 default:
1979 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001980 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001981
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001982 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07001983}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001984
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001985/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001986 * size of first charge trial. "32" comes from vmscan.c's magic value.
1987 * TODO: maybe necessary to use big numbers in big irons.
1988 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001989#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001990struct memcg_stock_pcp {
1991 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001992 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001993 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001994 unsigned long flags;
1995#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001996};
1997static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001998static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001999
2000/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002001 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002002 * from local stock and true is returned. If the stock is 0 or charges from a
2003 * cgroup which is not current target, returns false. This stock will be
2004 * refilled.
2005 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002006static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002007{
2008 struct memcg_stock_pcp *stock;
2009 bool ret = true;
2010
2011 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002012 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002013 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002014 else /* need to call res_counter_charge */
2015 ret = false;
2016 put_cpu_var(memcg_stock);
2017 return ret;
2018}
2019
2020/*
2021 * Returns stocks cached in percpu to res_counter and reset cached information.
2022 */
2023static void drain_stock(struct memcg_stock_pcp *stock)
2024{
2025 struct mem_cgroup *old = stock->cached;
2026
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002027 if (stock->nr_pages) {
2028 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2029
2030 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002031 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002032 res_counter_uncharge(&old->memsw, bytes);
2033 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002034 }
2035 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002036}
2037
2038/*
2039 * This must be called under preempt disabled or must be called by
2040 * a thread which is pinned to local cpu.
2041 */
2042static void drain_local_stock(struct work_struct *dummy)
2043{
2044 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2045 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002046 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002047}
2048
2049/*
2050 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002051 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002052 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002053static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002054{
2055 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2056
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002057 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002058 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002059 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002060 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002061 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002062 put_cpu_var(memcg_stock);
2063}
2064
2065/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002066 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002067 * of the hierarchy under it. sync flag says whether we should block
2068 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002069 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002070static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002071{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002072 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002073
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002074 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002075 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002076 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002077 for_each_online_cpu(cpu) {
2078 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002079 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002080
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002081 memcg = stock->cached;
2082 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002083 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002084 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002085 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002086 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2087 if (cpu == curcpu)
2088 drain_local_stock(&stock->work);
2089 else
2090 schedule_work_on(cpu, &stock->work);
2091 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002092 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002093 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002094
2095 if (!sync)
2096 goto out;
2097
2098 for_each_online_cpu(cpu) {
2099 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002100 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002101 flush_work(&stock->work);
2102 }
2103out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002104 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002105}
2106
2107/*
2108 * Tries to drain stocked charges in other cpus. This function is asynchronous
2109 * and just put a work per cpu for draining localy on each cpu. Caller can
2110 * expects some charges will be back to res_counter later but cannot wait for
2111 * it.
2112 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002113static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002114{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002115 /*
2116 * If someone calls draining, avoid adding more kworker runs.
2117 */
2118 if (!mutex_trylock(&percpu_charge_mutex))
2119 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002120 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002121 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002122}
2123
2124/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002125static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002126{
2127 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002128 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002129 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002130 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002131}
2132
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002133/*
2134 * This function drains percpu counter value from DEAD cpu and
2135 * move it to local cpu. Note that this function can be preempted.
2136 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002137static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002138{
2139 int i;
2140
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002141 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002142 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002143 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002144
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002145 per_cpu(memcg->stat->count[i], cpu) = 0;
2146 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002147 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002148 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002149 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002150
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002151 per_cpu(memcg->stat->events[i], cpu) = 0;
2152 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002153 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002154 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002155}
2156
2157static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002158 unsigned long action,
2159 void *hcpu)
2160{
2161 int cpu = (unsigned long)hcpu;
2162 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002163 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002164
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002165 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002166 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002167
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002168 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002169 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002170
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002171 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002172 mem_cgroup_drain_pcp_counter(iter, cpu);
2173
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002174 stock = &per_cpu(memcg_stock, cpu);
2175 drain_stock(stock);
2176 return NOTIFY_OK;
2177}
2178
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002179
2180/* See __mem_cgroup_try_charge() for details */
2181enum {
2182 CHARGE_OK, /* success */
2183 CHARGE_RETRY, /* need to retry but retry is not bad */
2184 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2185 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2186 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2187};
2188
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002189static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002190 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002191{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002192 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002193 struct mem_cgroup *mem_over_limit;
2194 struct res_counter *fail_res;
2195 unsigned long flags = 0;
2196 int ret;
2197
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002198 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002199
2200 if (likely(!ret)) {
2201 if (!do_swap_account)
2202 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002203 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002204 if (likely(!ret))
2205 return CHARGE_OK;
2206
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002207 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002208 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2209 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2210 } else
2211 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002212 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002213 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2214 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002215 *
2216 * Never reclaim on behalf of optional batching, retry with a
2217 * single page instead.
2218 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002219 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002220 return CHARGE_RETRY;
2221
2222 if (!(gfp_mask & __GFP_WAIT))
2223 return CHARGE_WOULDBLOCK;
2224
Johannes Weiner56600482012-01-12 17:17:59 -08002225 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002226 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002227 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002228 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002229 * Even though the limit is exceeded at this point, reclaim
2230 * may have been able to free some pages. Retry the charge
2231 * before killing the task.
2232 *
2233 * Only for regular pages, though: huge pages are rather
2234 * unlikely to succeed so close to the limit, and we fall back
2235 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002236 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002237 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002238 return CHARGE_RETRY;
2239
2240 /*
2241 * At task move, charge accounts can be doubly counted. So, it's
2242 * better to wait until the end of task_move if something is going on.
2243 */
2244 if (mem_cgroup_wait_acct_move(mem_over_limit))
2245 return CHARGE_RETRY;
2246
2247 /* If we don't need to call oom-killer at el, return immediately */
2248 if (!oom_check)
2249 return CHARGE_NOMEM;
2250 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002251 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002252 return CHARGE_OOM_DIE;
2253
2254 return CHARGE_RETRY;
2255}
2256
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002257/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002258 * __mem_cgroup_try_charge() does
2259 * 1. detect memcg to be charged against from passed *mm and *ptr,
2260 * 2. update res_counter
2261 * 3. call memory reclaim if necessary.
2262 *
2263 * In some special case, if the task is fatal, fatal_signal_pending() or
2264 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2265 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2266 * as possible without any hazards. 2: all pages should have a valid
2267 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2268 * pointer, that is treated as a charge to root_mem_cgroup.
2269 *
2270 * So __mem_cgroup_try_charge() will return
2271 * 0 ... on success, filling *ptr with a valid memcg pointer.
2272 * -ENOMEM ... charge failure because of resource limits.
2273 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2274 *
2275 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2276 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002277 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002278static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002279 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002280 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002281 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002282 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002283{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002284 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002285 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002286 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002287 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002288
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002289 /*
2290 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2291 * in system level. So, allow to go ahead dying process in addition to
2292 * MEMDIE process.
2293 */
2294 if (unlikely(test_thread_flag(TIF_MEMDIE)
2295 || fatal_signal_pending(current)))
2296 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002297
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002298 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002299 * We always charge the cgroup the mm_struct belongs to.
2300 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002301 * thread group leader migrates. It's possible that mm is not
2302 * set, if so charge the init_mm (happens for pagecache usage).
2303 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002304 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002305 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002306again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002307 if (*ptr) { /* css should be a valid one */
2308 memcg = *ptr;
2309 VM_BUG_ON(css_is_removed(&memcg->css));
2310 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002311 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002312 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002313 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002314 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002315 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002316 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002317
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002318 rcu_read_lock();
2319 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002320 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002321 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002322 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002323 * race with swapoff. Then, we have small risk of mis-accouning.
2324 * But such kind of mis-account by race always happens because
2325 * we don't have cgroup_mutex(). It's overkill and we allo that
2326 * small race, here.
2327 * (*) swapoff at el will charge against mm-struct not against
2328 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002329 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002330 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002331 if (!memcg)
2332 memcg = root_mem_cgroup;
2333 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002334 rcu_read_unlock();
2335 goto done;
2336 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002337 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002338 /*
2339 * It seems dagerous to access memcg without css_get().
2340 * But considering how consume_stok works, it's not
2341 * necessary. If consume_stock success, some charges
2342 * from this memcg are cached on this cpu. So, we
2343 * don't need to call css_get()/css_tryget() before
2344 * calling consume_stock().
2345 */
2346 rcu_read_unlock();
2347 goto done;
2348 }
2349 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002350 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002351 rcu_read_unlock();
2352 goto again;
2353 }
2354 rcu_read_unlock();
2355 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002356
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002357 do {
2358 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002359
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002360 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002361 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002362 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002363 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002364 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002365
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002366 oom_check = false;
2367 if (oom && !nr_oom_retries) {
2368 oom_check = true;
2369 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2370 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002371
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002372 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002373 switch (ret) {
2374 case CHARGE_OK:
2375 break;
2376 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002377 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002378 css_put(&memcg->css);
2379 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002380 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002381 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002382 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002383 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002384 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002385 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002386 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002387 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002388 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002389 /* If oom, we never return -ENOMEM */
2390 nr_oom_retries--;
2391 break;
2392 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002393 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002394 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002395 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002396 } while (ret != CHARGE_OK);
2397
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002398 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002399 refill_stock(memcg, batch - nr_pages);
2400 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002401done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002402 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002403 return 0;
2404nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002405 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002406 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002407bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002408 *ptr = root_mem_cgroup;
2409 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002410}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002411
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002412/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002413 * Somemtimes we have to undo a charge we got by try_charge().
2414 * This function is for that and do uncharge, put css's refcnt.
2415 * gotten by try_charge().
2416 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002417static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b82011-03-23 16:42:33 -07002418 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002419{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002420 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b82011-03-23 16:42:33 -07002421 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002422
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002423 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002424 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002425 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002426 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002427}
2428
2429/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002430 * A helper function to get mem_cgroup from ID. must be called under
2431 * rcu_read_lock(). The caller must check css_is_removed() or some if
2432 * it's concern. (dropping refcnt from swap can be called against removed
2433 * memcg.)
2434 */
2435static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2436{
2437 struct cgroup_subsys_state *css;
2438
2439 /* ID 0 is unused ID */
2440 if (!id)
2441 return NULL;
2442 css = css_lookup(&mem_cgroup_subsys, id);
2443 if (!css)
2444 return NULL;
2445 return container_of(css, struct mem_cgroup, css);
2446}
2447
Wu Fengguange42d9d52009-12-16 12:19:59 +01002448struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002449{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002450 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002451 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002452 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002453 swp_entry_t ent;
2454
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002455 VM_BUG_ON(!PageLocked(page));
2456
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002457 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002458 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002459 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002460 memcg = pc->mem_cgroup;
2461 if (memcg && !css_tryget(&memcg->css))
2462 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002463 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002464 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002465 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002466 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002467 memcg = mem_cgroup_lookup(id);
2468 if (memcg && !css_tryget(&memcg->css))
2469 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002470 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002471 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002472 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002473 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002474}
2475
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002476static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002477 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002478 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002479 enum charge_type ctype,
2480 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002481{
Johannes Weinerce587e62012-04-24 20:22:33 +02002482 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002483 struct zone *uninitialized_var(zone);
2484 bool was_on_lru = false;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002485 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002486
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002487 lock_page_cgroup(pc);
2488 if (unlikely(PageCgroupUsed(pc))) {
2489 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002490 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002491 return;
2492 }
2493 /*
2494 * we don't need page_cgroup_lock about tail pages, becase they are not
2495 * accessed by any other context at this point.
2496 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002497
2498 /*
2499 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2500 * may already be on some other mem_cgroup's LRU. Take care of it.
2501 */
2502 if (lrucare) {
2503 zone = page_zone(page);
2504 spin_lock_irq(&zone->lru_lock);
2505 if (PageLRU(page)) {
2506 ClearPageLRU(page);
2507 del_page_from_lru_list(zone, page, page_lru(page));
2508 was_on_lru = true;
2509 }
2510 }
2511
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002512 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002513 /*
2514 * We access a page_cgroup asynchronously without lock_page_cgroup().
2515 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2516 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2517 * before USED bit, we need memory barrier here.
2518 * See mem_cgroup_add_lru_list(), etc.
2519 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002520 smp_wmb();
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002521 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002522
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002523 if (lrucare) {
2524 if (was_on_lru) {
2525 VM_BUG_ON(PageLRU(page));
2526 SetPageLRU(page);
2527 add_page_to_lru_list(zone, page, page_lru(page));
2528 }
2529 spin_unlock_irq(&zone->lru_lock);
2530 }
2531
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002532 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2533 anon = true;
2534 else
2535 anon = false;
2536
2537 mem_cgroup_charge_statistics(memcg, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002538 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002539
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002540 /*
2541 * "charge_statistics" updated event counter. Then, check it.
2542 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2543 * if they exceeds softlimit.
2544 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002545 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002546}
2547
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002548#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2549
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002550#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION))
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002551/*
2552 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002553 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2554 * charge/uncharge will be never happen and move_account() is done under
2555 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002556 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002557void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002558{
2559 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002560 struct page_cgroup *pc;
2561 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002562
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002563 if (mem_cgroup_disabled())
2564 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002565 for (i = 1; i < HPAGE_PMD_NR; i++) {
2566 pc = head_pc + i;
2567 pc->mem_cgroup = head_pc->mem_cgroup;
2568 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002569 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2570 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002571}
Hugh Dickins12d27102012-01-12 17:19:52 -08002572#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002573
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002574/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002575 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002576 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002577 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002578 * @pc: page_cgroup of the page.
2579 * @from: mem_cgroup which the page is moved from.
2580 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002581 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002582 *
2583 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002584 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002585 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002586 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002587 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002588 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002589 * true, this function does "uncharge" from old cgroup, but it doesn't if
2590 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002591 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002592static int mem_cgroup_move_account(struct page *page,
2593 unsigned int nr_pages,
2594 struct page_cgroup *pc,
2595 struct mem_cgroup *from,
2596 struct mem_cgroup *to,
2597 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002598{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002599 unsigned long flags;
2600 int ret;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002601 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002602
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002603 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002604 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002605 /*
2606 * The page is isolated from LRU. So, collapse function
2607 * will not handle this page. But page splitting can happen.
2608 * Do this check under compound_page_lock(). The caller should
2609 * hold it.
2610 */
2611 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002612 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002613 goto out;
2614
2615 lock_page_cgroup(pc);
2616
2617 ret = -EINVAL;
2618 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2619 goto unlock;
2620
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002621 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002622
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002623 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002624 /* Update mapped_file data for mem_cgroup */
2625 preempt_disable();
2626 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2627 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2628 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002629 }
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002630 mem_cgroup_charge_statistics(from, anon, -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002631 if (uncharge)
2632 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b82011-03-23 16:42:33 -07002633 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002634
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002635 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002636 pc->mem_cgroup = to;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002637 mem_cgroup_charge_statistics(to, anon, nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002638 /*
2639 * We charges against "to" which may not have any tasks. Then, "to"
2640 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002641 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002642 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002643 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002644 */
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07002645 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002646 ret = 0;
2647unlock:
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002648 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002649 /*
2650 * check events
2651 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002652 memcg_check_events(to, page);
2653 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002654out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002655 return ret;
2656}
2657
2658/*
2659 * move charges to its parent.
2660 */
2661
Johannes Weiner5564e882011-03-23 16:42:29 -07002662static int mem_cgroup_move_parent(struct page *page,
2663 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002664 struct mem_cgroup *child,
2665 gfp_t gfp_mask)
2666{
2667 struct cgroup *cg = child->css.cgroup;
2668 struct cgroup *pcg = cg->parent;
2669 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002670 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002671 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002672 int ret;
2673
2674 /* Is ROOT ? */
2675 if (!pcg)
2676 return -EINVAL;
2677
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002678 ret = -EBUSY;
2679 if (!get_page_unless_zero(page))
2680 goto out;
2681 if (isolate_lru_page(page))
2682 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002683
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002684 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002685
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002686 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002687 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002688 if (ret)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002689 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002690
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002691 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002692 flags = compound_lock_irqsave(page);
2693
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002694 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002695 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002696 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002697
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002698 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002699 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002700put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002701 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002702put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002703 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002704out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002705 return ret;
2706}
2707
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002708/*
2709 * Charge the memory controller for page usage.
2710 * Return
2711 * 0 if the charge was successful
2712 * < 0 if the cgroup is over its limit
2713 */
2714static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002715 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002716{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002717 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002718 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002719 bool oom = true;
2720 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002721
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002722 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002723 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002724 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002725 /*
2726 * Never OOM-kill a process for a huge page. The
2727 * fault handler will fall back to regular pages.
2728 */
2729 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002730 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002731
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002732 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002733 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002734 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02002735 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002736 return 0;
2737}
2738
2739int mem_cgroup_newpage_charge(struct page *page,
2740 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002741{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002742 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002743 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002744 VM_BUG_ON(page_mapped(page));
2745 VM_BUG_ON(page->mapping && !PageAnon(page));
2746 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002747 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002748 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002749}
2750
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002751static void
2752__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2753 enum charge_type ctype);
2754
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002755int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2756 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002757{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002758 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002759 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002760 int ret;
2761
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002762 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002763 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002764 if (PageCompound(page))
2765 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002766
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002767 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002768 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002769 if (!page_is_file_cache(page))
2770 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002771
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002772 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002773 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002774 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002775 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002776 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002777 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2778 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002779 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002780}
2781
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002782/*
2783 * While swap-in, try_charge -> commit or cancel, the page is locked.
2784 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002785 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002786 * "commit()" or removed by "cancel()"
2787 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002788int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2789 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002790 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002791{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002792 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002793 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002794
Johannes Weiner72835c82012-01-12 17:18:32 -08002795 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002796
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002797 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002798 return 0;
2799
2800 if (!do_swap_account)
2801 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002802 /*
2803 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002804 * the pte, and even removed page from swap cache: in those cases
2805 * do_swap_page()'s pte_same() test will fail; but there's also a
2806 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002807 */
2808 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002809 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002810 memcg = try_get_mem_cgroup_from_page(page);
2811 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002812 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002813 *memcgp = memcg;
2814 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002815 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002816 if (ret == -EINTR)
2817 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002818 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002819charge_cur_mm:
2820 if (unlikely(!mm))
2821 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002822 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2823 if (ret == -EINTR)
2824 ret = 0;
2825 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002826}
2827
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002828static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002829__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002830 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002831{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002832 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002833 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002834 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002835 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002836 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002837
Johannes Weinerce587e62012-04-24 20:22:33 +02002838 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002839 /*
2840 * Now swap is on-memory. This means this page may be
2841 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002842 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2843 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2844 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002845 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002846 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002847 swp_entry_t ent = {.val = page_private(page)};
Johannes Weiner72835c82012-01-12 17:18:32 -08002848 struct mem_cgroup *swap_memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002849 unsigned short id;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002850
2851 id = swap_cgroup_record(ent, 0);
2852 rcu_read_lock();
Johannes Weiner72835c82012-01-12 17:18:32 -08002853 swap_memcg = mem_cgroup_lookup(id);
2854 if (swap_memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002855 /*
2856 * This recorded memcg can be obsolete one. So, avoid
2857 * calling css_tryget
2858 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002859 if (!mem_cgroup_is_root(swap_memcg))
2860 res_counter_uncharge(&swap_memcg->memsw,
2861 PAGE_SIZE);
2862 mem_cgroup_swap_statistics(swap_memcg, false);
2863 mem_cgroup_put(swap_memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002864 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002865 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002866 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002867 /*
2868 * At swapin, we may charge account against cgroup which has no tasks.
2869 * So, rmdir()->pre_destroy() can be called while we do this charge.
2870 * In that case, we need to call pre_destroy() again. check it here.
2871 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002872 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002873}
2874
Johannes Weiner72835c82012-01-12 17:18:32 -08002875void mem_cgroup_commit_charge_swapin(struct page *page,
2876 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002877{
Johannes Weiner72835c82012-01-12 17:18:32 -08002878 __mem_cgroup_commit_charge_swapin(page, memcg,
2879 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002880}
2881
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002882void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002883{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002884 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002885 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002886 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002887 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002888 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002889}
2890
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002891static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002892 unsigned int nr_pages,
2893 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002894{
2895 struct memcg_batch_info *batch = NULL;
2896 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002897
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002898 /* If swapout, usage of swap doesn't decrease */
2899 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2900 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002901
2902 batch = &current->memcg_batch;
2903 /*
2904 * In usual, we do css_get() when we remember memcg pointer.
2905 * But in this case, we keep res->usage until end of a series of
2906 * uncharges. Then, it's ok to ignore memcg's refcnt.
2907 */
2908 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002909 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002910 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002911 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002912 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002913 * the same cgroup and we have chance to coalesce uncharges.
2914 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2915 * because we want to do uncharge as soon as possible.
2916 */
2917
2918 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2919 goto direct_uncharge;
2920
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002921 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002922 goto direct_uncharge;
2923
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002924 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002925 * In typical case, batch->memcg == mem. This means we can
2926 * merge a series of uncharges to an uncharge of res_counter.
2927 * If not, we uncharge res_counter ony by one.
2928 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002929 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002930 goto direct_uncharge;
2931 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002932 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002933 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002934 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002935 return;
2936direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002937 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002938 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002939 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2940 if (unlikely(batch->memcg != memcg))
2941 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002942}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002943
Balbir Singh8697d332008-02-07 00:13:59 -08002944/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002945 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002946 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002947static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002948__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002949{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002950 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002951 unsigned int nr_pages = 1;
2952 struct page_cgroup *pc;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002953 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002954
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002955 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002956 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002957
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002958 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002959 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002960
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002961 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002962 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002963 VM_BUG_ON(!PageTransHuge(page));
2964 }
Balbir Singh8697d332008-02-07 00:13:59 -08002965 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002966 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002967 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002968 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002969 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002970 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002971
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002972 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002973
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002974 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002975
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002976 if (!PageCgroupUsed(pc))
2977 goto unlock_out;
2978
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002979 anon = PageAnon(page);
2980
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002981 switch (ctype) {
2982 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07002983 /*
2984 * Generally PageAnon tells if it's the anon statistics to be
2985 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2986 * used before page reached the stage of being marked PageAnon.
2987 */
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07002988 anon = true;
2989 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002990 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07002991 /* See mem_cgroup_prepare_migration() */
2992 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002993 goto unlock_out;
2994 break;
2995 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2996 if (!PageAnon(page)) { /* Shared memory */
2997 if (page->mapping && !page_is_file_cache(page))
2998 goto unlock_out;
2999 } else if (page_mapped(page)) /* Anon */
3000 goto unlock_out;
3001 break;
3002 default:
3003 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003004 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003005
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003006 mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003007
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003008 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003009 /*
3010 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3011 * freed from LRU. This is safe because uncharged page is expected not
3012 * to be reused (freed soon). Exception is SwapCache, it's handled by
3013 * special functions.
3014 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003015
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003016 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003017 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003018 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003019 * will never be freed.
3020 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003021 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003022 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003023 mem_cgroup_swap_statistics(memcg, true);
3024 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003025 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003026 if (!mem_cgroup_is_root(memcg))
3027 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003028
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003029 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003030
3031unlock_out:
3032 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003033 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003034}
3035
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003036void mem_cgroup_uncharge_page(struct page *page)
3037{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003038 /* early check. */
3039 if (page_mapped(page))
3040 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08003041 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003042 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3043}
3044
3045void mem_cgroup_uncharge_cache_page(struct page *page)
3046{
3047 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003048 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003049 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3050}
3051
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003052/*
3053 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3054 * In that cases, pages are freed continuously and we can expect pages
3055 * are in the same memcg. All these calls itself limits the number of
3056 * pages freed at once, then uncharge_start/end() is called properly.
3057 * This may be called prural(2) times in a context,
3058 */
3059
3060void mem_cgroup_uncharge_start(void)
3061{
3062 current->memcg_batch.do_batch++;
3063 /* We can do nest. */
3064 if (current->memcg_batch.do_batch == 1) {
3065 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003066 current->memcg_batch.nr_pages = 0;
3067 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003068 }
3069}
3070
3071void mem_cgroup_uncharge_end(void)
3072{
3073 struct memcg_batch_info *batch = &current->memcg_batch;
3074
3075 if (!batch->do_batch)
3076 return;
3077
3078 batch->do_batch--;
3079 if (batch->do_batch) /* If stacked, do nothing. */
3080 return;
3081
3082 if (!batch->memcg)
3083 return;
3084 /*
3085 * This "batch->memcg" is valid without any css_get/put etc...
3086 * bacause we hide charges behind us.
3087 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003088 if (batch->nr_pages)
3089 res_counter_uncharge(&batch->memcg->res,
3090 batch->nr_pages * PAGE_SIZE);
3091 if (batch->memsw_nr_pages)
3092 res_counter_uncharge(&batch->memcg->memsw,
3093 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003094 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003095 /* forget this pointer (for sanity check) */
3096 batch->memcg = NULL;
3097}
3098
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003099#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003100/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003101 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003102 * memcg information is recorded to swap_cgroup of "ent"
3103 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003104void
3105mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003106{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003107 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003108 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003109
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003110 if (!swapout) /* this was a swap cache but the swap is unused ! */
3111 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3112
3113 memcg = __mem_cgroup_uncharge_common(page, ctype);
3114
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003115 /*
3116 * record memcg information, if swapout && memcg != NULL,
3117 * mem_cgroup_get() was called in uncharge().
3118 */
3119 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003120 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003121}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003122#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003123
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003124#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3125/*
3126 * called from swap_entry_free(). remove record in swap_cgroup and
3127 * uncharge "memsw" account.
3128 */
3129void mem_cgroup_uncharge_swap(swp_entry_t ent)
3130{
3131 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003132 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003133
3134 if (!do_swap_account)
3135 return;
3136
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003137 id = swap_cgroup_record(ent, 0);
3138 rcu_read_lock();
3139 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003140 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003141 /*
3142 * We uncharge this because swap is freed.
3143 * This memcg can be obsolete one. We avoid calling css_tryget
3144 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003145 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003146 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003147 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003148 mem_cgroup_put(memcg);
3149 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003150 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003151}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003152
3153/**
3154 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3155 * @entry: swap entry to be moved
3156 * @from: mem_cgroup which the entry is moved from
3157 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003158 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003159 *
3160 * It succeeds only when the swap_cgroup's record for this entry is the same
3161 * as the mem_cgroup's id of @from.
3162 *
3163 * Returns 0 on success, -EINVAL on failure.
3164 *
3165 * The caller must have charged to @to, IOW, called res_counter_charge() about
3166 * both res and memsw, and called css_get().
3167 */
3168static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003169 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003170{
3171 unsigned short old_id, new_id;
3172
3173 old_id = css_id(&from->css);
3174 new_id = css_id(&to->css);
3175
3176 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003177 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003178 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003179 /*
3180 * This function is only called from task migration context now.
3181 * It postpones res_counter and refcount handling till the end
3182 * of task migration(mem_cgroup_clear_mc()) for performance
3183 * improvement. But we cannot postpone mem_cgroup_get(to)
3184 * because if the process that has been moved to @to does
3185 * swap-in, the refcount of @to might be decreased to 0.
3186 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003187 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003188 if (need_fixup) {
3189 if (!mem_cgroup_is_root(from))
3190 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3191 mem_cgroup_put(from);
3192 /*
3193 * we charged both to->res and to->memsw, so we should
3194 * uncharge to->res.
3195 */
3196 if (!mem_cgroup_is_root(to))
3197 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003198 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003199 return 0;
3200 }
3201 return -EINVAL;
3202}
3203#else
3204static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003205 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003206{
3207 return -EINVAL;
3208}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003209#endif
3210
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003211/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003212 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3213 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003214 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003215int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003216 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003217{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003218 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003219 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003220 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003221 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003222
Johannes Weiner72835c82012-01-12 17:18:32 -08003223 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003224
Andrea Arcangeliec168512011-01-13 15:46:56 -08003225 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003226 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003227 return 0;
3228
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003229 pc = lookup_page_cgroup(page);
3230 lock_page_cgroup(pc);
3231 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003232 memcg = pc->mem_cgroup;
3233 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003234 /*
3235 * At migrating an anonymous page, its mapcount goes down
3236 * to 0 and uncharge() will be called. But, even if it's fully
3237 * unmapped, migration may fail and this page has to be
3238 * charged again. We set MIGRATION flag here and delay uncharge
3239 * until end_migration() is called
3240 *
3241 * Corner Case Thinking
3242 * A)
3243 * When the old page was mapped as Anon and it's unmap-and-freed
3244 * while migration was ongoing.
3245 * If unmap finds the old page, uncharge() of it will be delayed
3246 * until end_migration(). If unmap finds a new page, it's
3247 * uncharged when it make mapcount to be 1->0. If unmap code
3248 * finds swap_migration_entry, the new page will not be mapped
3249 * and end_migration() will find it(mapcount==0).
3250 *
3251 * B)
3252 * When the old page was mapped but migraion fails, the kernel
3253 * remaps it. A charge for it is kept by MIGRATION flag even
3254 * if mapcount goes down to 0. We can do remap successfully
3255 * without charging it again.
3256 *
3257 * C)
3258 * The "old" page is under lock_page() until the end of
3259 * migration, so, the old page itself will not be swapped-out.
3260 * If the new page is swapped out before end_migraton, our
3261 * hook to usual swap-out path will catch the event.
3262 */
3263 if (PageAnon(page))
3264 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003265 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003266 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003267 /*
3268 * If the page is not charged at this point,
3269 * we return here.
3270 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003271 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003272 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003273
Johannes Weiner72835c82012-01-12 17:18:32 -08003274 *memcgp = memcg;
3275 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003276 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003277 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003278 if (PageAnon(page)) {
3279 lock_page_cgroup(pc);
3280 ClearPageCgroupMigration(pc);
3281 unlock_page_cgroup(pc);
3282 /*
3283 * The old page may be fully unmapped while we kept it.
3284 */
3285 mem_cgroup_uncharge_page(page);
3286 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003287 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003288 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003289 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003290 /*
3291 * We charge new page before it's used/mapped. So, even if unlock_page()
3292 * is called before end_migration, we can catch all events on this new
3293 * page. In the case new page is migrated but not remapped, new page's
3294 * mapcount will be finally 0 and we call uncharge in end_migration().
3295 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003296 if (PageAnon(page))
3297 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3298 else if (page_is_file_cache(page))
3299 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3300 else
3301 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Johannes Weinerce587e62012-04-24 20:22:33 +02003302 __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003303 return ret;
3304}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003305
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003306/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003307void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003308 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003309{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003310 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003311 struct page_cgroup *pc;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003312 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003313
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003314 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003315 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003316 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003317 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003318 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003319 used = oldpage;
3320 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003321 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003322 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003323 unused = oldpage;
3324 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003325 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003326 * We disallowed uncharge of pages under migration because mapcount
3327 * of the page goes down to zero, temporarly.
3328 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003329 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003330 pc = lookup_page_cgroup(oldpage);
3331 lock_page_cgroup(pc);
3332 ClearPageCgroupMigration(pc);
3333 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003334 anon = PageAnon(used);
3335 __mem_cgroup_uncharge_common(unused,
3336 anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED
3337 : MEM_CGROUP_CHARGE_TYPE_CACHE);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003338
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003339 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003340 * If a page is a file cache, radix-tree replacement is very atomic
3341 * and we can skip this check. When it was an Anon page, its mapcount
3342 * goes down to 0. But because we added MIGRATION flage, it's not
3343 * uncharged yet. There are several case but page->mapcount check
3344 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3345 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003346 */
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003347 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003348 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003349 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003350 * At migration, we may charge account against cgroup which has no
3351 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003352 * So, rmdir()->pre_destroy() can be called while we do this charge.
3353 * In that case, we need to call pre_destroy() again. check it here.
3354 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003355 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003356}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003357
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003358/*
3359 * At replace page cache, newpage is not under any memcg but it's on
3360 * LRU. So, this function doesn't touch res_counter but handles LRU
3361 * in correct way. Both pages are locked so we cannot race with uncharge.
3362 */
3363void mem_cgroup_replace_page_cache(struct page *oldpage,
3364 struct page *newpage)
3365{
3366 struct mem_cgroup *memcg;
3367 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003368 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003369
3370 if (mem_cgroup_disabled())
3371 return;
3372
3373 pc = lookup_page_cgroup(oldpage);
3374 /* fix accounting on old pages */
3375 lock_page_cgroup(pc);
3376 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyukib2402852012-03-21 16:34:22 -07003377 mem_cgroup_charge_statistics(memcg, false, -1);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003378 ClearPageCgroupUsed(pc);
3379 unlock_page_cgroup(pc);
3380
3381 if (PageSwapBacked(oldpage))
3382 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3383
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003384 /*
3385 * Even if newpage->mapping was NULL before starting replacement,
3386 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3387 * LRU while we overwrite pc->mem_cgroup.
3388 */
Johannes Weinerce587e62012-04-24 20:22:33 +02003389 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003390}
3391
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003392#ifdef CONFIG_DEBUG_VM
3393static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3394{
3395 struct page_cgroup *pc;
3396
3397 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003398 /*
3399 * Can be NULL while feeding pages into the page allocator for
3400 * the first time, i.e. during boot or memory hotplug;
3401 * or when mem_cgroup_disabled().
3402 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003403 if (likely(pc) && PageCgroupUsed(pc))
3404 return pc;
3405 return NULL;
3406}
3407
3408bool mem_cgroup_bad_page_check(struct page *page)
3409{
3410 if (mem_cgroup_disabled())
3411 return false;
3412
3413 return lookup_page_cgroup_used(page) != NULL;
3414}
3415
3416void mem_cgroup_print_bad_page(struct page *page)
3417{
3418 struct page_cgroup *pc;
3419
3420 pc = lookup_page_cgroup_used(page);
3421 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003422 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003423 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003424 }
3425}
3426#endif
3427
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003428static DEFINE_MUTEX(set_limit_mutex);
3429
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003430static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003431 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003432{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003433 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003434 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003435 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003436 int children = mem_cgroup_count_children(memcg);
3437 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003438 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003439
3440 /*
3441 * For keeping hierarchical_reclaim simple, how long we should retry
3442 * is depends on callers. We set our retry-count to be function
3443 * of # of children which we should visit in this loop.
3444 */
3445 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3446
3447 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003448
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003449 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003450 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003451 if (signal_pending(current)) {
3452 ret = -EINTR;
3453 break;
3454 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003455 /*
3456 * Rather than hide all in some function, I do this in
3457 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003458 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003459 */
3460 mutex_lock(&set_limit_mutex);
3461 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3462 if (memswlimit < val) {
3463 ret = -EINVAL;
3464 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003465 break;
3466 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003467
3468 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3469 if (memlimit < val)
3470 enlarge = 1;
3471
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003472 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003473 if (!ret) {
3474 if (memswlimit == val)
3475 memcg->memsw_is_minimum = true;
3476 else
3477 memcg->memsw_is_minimum = false;
3478 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003479 mutex_unlock(&set_limit_mutex);
3480
3481 if (!ret)
3482 break;
3483
Johannes Weiner56600482012-01-12 17:17:59 -08003484 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3485 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003486 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3487 /* Usage is reduced ? */
3488 if (curusage >= oldusage)
3489 retry_count--;
3490 else
3491 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003492 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003493 if (!ret && enlarge)
3494 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003495
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003496 return ret;
3497}
3498
Li Zefan338c8432009-06-17 16:27:15 -07003499static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3500 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003501{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003502 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003503 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003504 int children = mem_cgroup_count_children(memcg);
3505 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003506 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003507
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003508 /* see mem_cgroup_resize_res_limit */
3509 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3510 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003511 while (retry_count) {
3512 if (signal_pending(current)) {
3513 ret = -EINTR;
3514 break;
3515 }
3516 /*
3517 * Rather than hide all in some function, I do this in
3518 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003519 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003520 */
3521 mutex_lock(&set_limit_mutex);
3522 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3523 if (memlimit > val) {
3524 ret = -EINVAL;
3525 mutex_unlock(&set_limit_mutex);
3526 break;
3527 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003528 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3529 if (memswlimit < val)
3530 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003531 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003532 if (!ret) {
3533 if (memlimit == val)
3534 memcg->memsw_is_minimum = true;
3535 else
3536 memcg->memsw_is_minimum = false;
3537 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003538 mutex_unlock(&set_limit_mutex);
3539
3540 if (!ret)
3541 break;
3542
Johannes Weiner56600482012-01-12 17:17:59 -08003543 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3544 MEM_CGROUP_RECLAIM_NOSWAP |
3545 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003546 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003547 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003548 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003549 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003550 else
3551 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003552 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003553 if (!ret && enlarge)
3554 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003555 return ret;
3556}
3557
Balbir Singh4e416952009-09-23 15:56:39 -07003558unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003559 gfp_t gfp_mask,
3560 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003561{
3562 unsigned long nr_reclaimed = 0;
3563 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3564 unsigned long reclaimed;
3565 int loop = 0;
3566 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003567 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003568 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003569
3570 if (order > 0)
3571 return 0;
3572
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003573 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003574 /*
3575 * This loop can run a while, specially if mem_cgroup's continuously
3576 * keep exceeding their soft limit and putting the system under
3577 * pressure
3578 */
3579 do {
3580 if (next_mz)
3581 mz = next_mz;
3582 else
3583 mz = mem_cgroup_largest_soft_limit_node(mctz);
3584 if (!mz)
3585 break;
3586
Ying Han0ae5e892011-05-26 16:25:25 -07003587 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003588 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08003589 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003590 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003591 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003592 spin_lock(&mctz->lock);
3593
3594 /*
3595 * If we failed to reclaim anything from this memory cgroup
3596 * it is time to move on to the next cgroup
3597 */
3598 next_mz = NULL;
3599 if (!reclaimed) {
3600 do {
3601 /*
3602 * Loop until we find yet another one.
3603 *
3604 * By the time we get the soft_limit lock
3605 * again, someone might have aded the
3606 * group back on the RB tree. Iterate to
3607 * make sure we get a different mem.
3608 * mem_cgroup_largest_soft_limit_node returns
3609 * NULL if no other cgroup is present on
3610 * the tree
3611 */
3612 next_mz =
3613 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003614 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003615 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003616 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003617 break;
3618 } while (1);
3619 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003620 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3621 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003622 /*
3623 * One school of thought says that we should not add
3624 * back the node to the tree if reclaim returns 0.
3625 * But our reclaim could return 0, simply because due
3626 * to priority we are exposing a smaller subset of
3627 * memory to reclaim from. Consider this as a longer
3628 * term TODO.
3629 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003630 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003631 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003632 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003633 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003634 loop++;
3635 /*
3636 * Could not reclaim anything and there are no more
3637 * mem cgroups to try or we seem to be looping without
3638 * reclaiming anything.
3639 */
3640 if (!nr_reclaimed &&
3641 (next_mz == NULL ||
3642 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3643 break;
3644 } while (!nr_reclaimed);
3645 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07003646 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07003647 return nr_reclaimed;
3648}
3649
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003650/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003651 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003652 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3653 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003654static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003655 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003656{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003657 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003658 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003659 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003660 struct page *busy;
3661 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003662 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003663
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003664 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003665 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003666 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003667
Hugh Dickins1eb49272012-03-21 16:34:19 -07003668 loop = mz->lru_size[lru];
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003669 /* give some margin against EBUSY etc...*/
3670 loop += 256;
3671 busy = NULL;
3672 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003673 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003674 struct page *page;
3675
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003676 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003677 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003678 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003679 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003680 break;
3681 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003682 page = list_entry(list->prev, struct page, lru);
3683 if (busy == page) {
3684 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003685 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003686 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003687 continue;
3688 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003689 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003690
Johannes Weiner925b7672012-01-12 17:18:15 -08003691 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003692
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003693 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003694 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003695 break;
3696
3697 if (ret == -EBUSY || ret == -EINVAL) {
3698 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003699 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003700 cond_resched();
3701 } else
3702 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003703 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003704
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003705 if (!ret && !list_empty(list))
3706 return -EBUSY;
3707 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003708}
3709
3710/*
3711 * make mem_cgroup's charge to be 0 if there is no task.
3712 * This enables deleting this mem_cgroup.
3713 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003714static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003715{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003716 int ret;
3717 int node, zid, shrink;
3718 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003719 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003720
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003721 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003722
3723 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003724 /* should free all ? */
3725 if (free_all)
3726 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003727move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003728 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003729 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003730 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003731 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003732 ret = -EINTR;
3733 if (signal_pending(current))
3734 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003735 /* This is for making all *used* pages to be on LRU. */
3736 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003737 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003738 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003739 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003740 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003741 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003742 enum lru_list lru;
3743 for_each_lru(lru) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003744 ret = mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003745 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003746 if (ret)
3747 break;
3748 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003749 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003750 if (ret)
3751 break;
3752 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003753 mem_cgroup_end_move(memcg);
3754 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003755 /* it seems parent cgroup doesn't have enough mem */
3756 if (ret == -ENOMEM)
3757 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003758 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003759 /* "ret" should also be checked to ensure all lists are empty. */
Glauber Costa569530f2012-04-12 12:49:13 -07003760 } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003761out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003762 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003763 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003764
3765try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003766 /* returns EBUSY if there is a task or if we come here twice. */
3767 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003768 ret = -EBUSY;
3769 goto out;
3770 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003771 /* we call try-to-free pages for make this cgroup empty */
3772 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003773 /* try to free all pages in this cgroup */
3774 shrink = 1;
Glauber Costa569530f2012-04-12 12:49:13 -07003775 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003776 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003777
3778 if (signal_pending(current)) {
3779 ret = -EINTR;
3780 goto out;
3781 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003782 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003783 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003784 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003785 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003786 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003787 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003788 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003789
3790 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003791 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003792 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003793 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003794}
3795
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003796int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3797{
3798 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3799}
3800
3801
Balbir Singh18f59ea2009-01-07 18:08:07 -08003802static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3803{
3804 return mem_cgroup_from_cont(cont)->use_hierarchy;
3805}
3806
3807static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3808 u64 val)
3809{
3810 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003811 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003812 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003813 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003814
3815 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003816 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003817
3818 cgroup_lock();
3819 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003820 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003821 * in the child subtrees. If it is unset, then the change can
3822 * occur, provided the current cgroup has no children.
3823 *
3824 * For the root cgroup, parent_mem is NULL, we allow value to be
3825 * set if there are no children.
3826 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003827 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003828 (val == 1 || val == 0)) {
3829 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003830 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003831 else
3832 retval = -EBUSY;
3833 } else
3834 retval = -EINVAL;
3835 cgroup_unlock();
3836
3837 return retval;
3838}
3839
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003840
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003841static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003842 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003843{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003844 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003845 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003846
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003847 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003848 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003849 val += mem_cgroup_read_stat(iter, idx);
3850
3851 if (val < 0) /* race ? */
3852 val = 0;
3853 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003854}
3855
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003856static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003857{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003858 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003859
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003860 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003861 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003862 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003863 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003864 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003865 }
3866
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003867 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3868 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003869
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003870 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003871 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003872
3873 return val << PAGE_SHIFT;
3874}
3875
Paul Menage2c3daa72008-04-29 00:59:58 -07003876static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003877{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003878 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003879 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003880 int type, name;
3881
3882 type = MEMFILE_TYPE(cft->private);
3883 name = MEMFILE_ATTR(cft->private);
3884 switch (type) {
3885 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003886 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003887 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003888 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003889 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003890 break;
3891 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003892 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003893 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003894 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003895 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003896 break;
3897 default:
3898 BUG();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003899 }
3900 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003901}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003902/*
3903 * The user of this function is...
3904 * RES_LIMIT.
3905 */
Paul Menage856c13a2008-07-25 01:47:04 -07003906static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3907 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003908{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003909 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003910 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003911 unsigned long long val;
3912 int ret;
3913
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003914 type = MEMFILE_TYPE(cft->private);
3915 name = MEMFILE_ATTR(cft->private);
3916 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003917 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003918 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3919 ret = -EINVAL;
3920 break;
3921 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003922 /* This function does all necessary parse...reuse it */
3923 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003924 if (ret)
3925 break;
3926 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003927 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003928 else
3929 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003930 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003931 case RES_SOFT_LIMIT:
3932 ret = res_counter_memparse_write_strategy(buffer, &val);
3933 if (ret)
3934 break;
3935 /*
3936 * For memsw, soft limits are hard to implement in terms
3937 * of semantics, for now, we support soft limits for
3938 * control without swap
3939 */
3940 if (type == _MEM)
3941 ret = res_counter_set_soft_limit(&memcg->res, val);
3942 else
3943 ret = -EINVAL;
3944 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003945 default:
3946 ret = -EINVAL; /* should be BUG() ? */
3947 break;
3948 }
3949 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003950}
3951
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003952static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3953 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3954{
3955 struct cgroup *cgroup;
3956 unsigned long long min_limit, min_memsw_limit, tmp;
3957
3958 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3959 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3960 cgroup = memcg->css.cgroup;
3961 if (!memcg->use_hierarchy)
3962 goto out;
3963
3964 while (cgroup->parent) {
3965 cgroup = cgroup->parent;
3966 memcg = mem_cgroup_from_cont(cgroup);
3967 if (!memcg->use_hierarchy)
3968 break;
3969 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3970 min_limit = min(min_limit, tmp);
3971 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3972 min_memsw_limit = min(min_memsw_limit, tmp);
3973 }
3974out:
3975 *mem_limit = min_limit;
3976 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003977}
3978
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003979static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003980{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003981 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003982 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003983
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003984 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003985 type = MEMFILE_TYPE(event);
3986 name = MEMFILE_ATTR(event);
3987 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003988 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003989 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003990 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003991 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003992 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003993 break;
3994 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003995 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003996 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003997 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003998 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003999 break;
4000 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004001
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07004002 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004003}
4004
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004005static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4006 struct cftype *cft)
4007{
4008 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4009}
4010
Daisuke Nishimura02491442010-03-10 15:22:17 -08004011#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004012static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4013 struct cftype *cft, u64 val)
4014{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004015 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004016
4017 if (val >= (1 << NR_MOVE_TYPE))
4018 return -EINVAL;
4019 /*
4020 * We check this value several times in both in can_attach() and
4021 * attach(), so we need cgroup lock to prevent this value from being
4022 * inconsistent.
4023 */
4024 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004025 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004026 cgroup_unlock();
4027
4028 return 0;
4029}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004030#else
4031static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4032 struct cftype *cft, u64 val)
4033{
4034 return -ENOSYS;
4035}
4036#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004037
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004038
4039/* For read statistics */
4040enum {
4041 MCS_CACHE,
4042 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004043 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004044 MCS_PGPGIN,
4045 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004046 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07004047 MCS_PGFAULT,
4048 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004049 MCS_INACTIVE_ANON,
4050 MCS_ACTIVE_ANON,
4051 MCS_INACTIVE_FILE,
4052 MCS_ACTIVE_FILE,
4053 MCS_UNEVICTABLE,
4054 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004055};
4056
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004057struct mcs_total_stat {
4058 s64 stat[NR_MCS_STAT];
4059};
4060
4061struct {
4062 char *local_name;
4063 char *total_name;
4064} memcg_stat_strings[NR_MCS_STAT] = {
4065 {"cache", "total_cache"},
4066 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004067 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004068 {"pgpgin", "total_pgpgin"},
4069 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004070 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004071 {"pgfault", "total_pgfault"},
4072 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004073 {"inactive_anon", "total_inactive_anon"},
4074 {"active_anon", "total_active_anon"},
4075 {"inactive_file", "total_inactive_file"},
4076 {"active_file", "total_active_file"},
4077 {"unevictable", "total_unevictable"}
4078};
4079
4080
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004081static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004082mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004083{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004084 s64 val;
4085
4086 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004087 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004088 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004089 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004090 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004091 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004092 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004093 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004094 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004095 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004096 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004097 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004098 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004099 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4100 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004101 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004102 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004103 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004104 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004105
4106 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004107 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004108 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004109 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004110 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004111 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004112 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004113 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004114 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004115 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004116 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004117}
4118
4119static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004120mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004121{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004122 struct mem_cgroup *iter;
4123
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004124 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004125 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004126}
4127
Ying Han406eb0c2011-05-26 16:25:37 -07004128#ifdef CONFIG_NUMA
4129static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4130{
4131 int nid;
4132 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4133 unsigned long node_nr;
4134 struct cgroup *cont = m->private;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004135 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07004136
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004137 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004138 seq_printf(m, "total=%lu", total_nr);
4139 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004140 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004141 seq_printf(m, " N%d=%lu", nid, node_nr);
4142 }
4143 seq_putc(m, '\n');
4144
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004145 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004146 seq_printf(m, "file=%lu", file_nr);
4147 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004148 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004149 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004150 seq_printf(m, " N%d=%lu", nid, node_nr);
4151 }
4152 seq_putc(m, '\n');
4153
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004154 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004155 seq_printf(m, "anon=%lu", anon_nr);
4156 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004157 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004158 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004159 seq_printf(m, " N%d=%lu", nid, node_nr);
4160 }
4161 seq_putc(m, '\n');
4162
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004163 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004164 seq_printf(m, "unevictable=%lu", unevictable_nr);
4165 for_each_node_state(nid, N_HIGH_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004166 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004167 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004168 seq_printf(m, " N%d=%lu", nid, node_nr);
4169 }
4170 seq_putc(m, '\n');
4171 return 0;
4172}
4173#endif /* CONFIG_NUMA */
4174
Paul Menagec64745c2008-04-29 01:00:02 -07004175static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4176 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004177{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004178 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004179 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004180 int i;
4181
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004182 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004183 mem_cgroup_get_local_stat(memcg, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004184
Ying Han406eb0c2011-05-26 16:25:37 -07004185
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004186 for (i = 0; i < NR_MCS_STAT; i++) {
4187 if (i == MCS_SWAP && !do_swap_account)
4188 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004189 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004190 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004191
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004192 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004193 {
4194 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004195 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004196 cb->fill(cb, "hierarchical_memory_limit", limit);
4197 if (do_swap_account)
4198 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4199 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004200
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004201 memset(&mystat, 0, sizeof(mystat));
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004202 mem_cgroup_get_total_stat(memcg, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004203 for (i = 0; i < NR_MCS_STAT; i++) {
4204 if (i == MCS_SWAP && !do_swap_account)
4205 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004206 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004207 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004208
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004209#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004210 {
4211 int nid, zid;
4212 struct mem_cgroup_per_zone *mz;
4213 unsigned long recent_rotated[2] = {0, 0};
4214 unsigned long recent_scanned[2] = {0, 0};
4215
4216 for_each_online_node(nid)
4217 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004218 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004219
4220 recent_rotated[0] +=
4221 mz->reclaim_stat.recent_rotated[0];
4222 recent_rotated[1] +=
4223 mz->reclaim_stat.recent_rotated[1];
4224 recent_scanned[0] +=
4225 mz->reclaim_stat.recent_scanned[0];
4226 recent_scanned[1] +=
4227 mz->reclaim_stat.recent_scanned[1];
4228 }
4229 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4230 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4231 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4232 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4233 }
4234#endif
4235
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004236 return 0;
4237}
4238
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004239static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4240{
4241 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4242
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004243 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004244}
4245
4246static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4247 u64 val)
4248{
4249 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4250 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004251
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004252 if (val > 100)
4253 return -EINVAL;
4254
4255 if (cgrp->parent == NULL)
4256 return -EINVAL;
4257
4258 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004259
4260 cgroup_lock();
4261
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004262 /* If under hierarchy, only empty-root can set this value */
4263 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004264 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4265 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004266 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004267 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004268
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004269 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004270
Li Zefan068b38c2009-01-15 13:51:26 -08004271 cgroup_unlock();
4272
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004273 return 0;
4274}
4275
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004276static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4277{
4278 struct mem_cgroup_threshold_ary *t;
4279 u64 usage;
4280 int i;
4281
4282 rcu_read_lock();
4283 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004284 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004285 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004286 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004287
4288 if (!t)
4289 goto unlock;
4290
4291 usage = mem_cgroup_usage(memcg, swap);
4292
4293 /*
4294 * current_threshold points to threshold just below usage.
4295 * If it's not true, a threshold was crossed after last
4296 * call of __mem_cgroup_threshold().
4297 */
Phil Carmody5407a562010-05-26 14:42:42 -07004298 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004299
4300 /*
4301 * Iterate backward over array of thresholds starting from
4302 * current_threshold and check if a threshold is crossed.
4303 * If none of thresholds below usage is crossed, we read
4304 * only one element of the array here.
4305 */
4306 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4307 eventfd_signal(t->entries[i].eventfd, 1);
4308
4309 /* i = current_threshold + 1 */
4310 i++;
4311
4312 /*
4313 * Iterate forward over array of thresholds starting from
4314 * current_threshold+1 and check if a threshold is crossed.
4315 * If none of thresholds above usage is crossed, we read
4316 * only one element of the array here.
4317 */
4318 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4319 eventfd_signal(t->entries[i].eventfd, 1);
4320
4321 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004322 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004323unlock:
4324 rcu_read_unlock();
4325}
4326
4327static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4328{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004329 while (memcg) {
4330 __mem_cgroup_threshold(memcg, false);
4331 if (do_swap_account)
4332 __mem_cgroup_threshold(memcg, true);
4333
4334 memcg = parent_mem_cgroup(memcg);
4335 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004336}
4337
4338static int compare_thresholds(const void *a, const void *b)
4339{
4340 const struct mem_cgroup_threshold *_a = a;
4341 const struct mem_cgroup_threshold *_b = b;
4342
4343 return _a->threshold - _b->threshold;
4344}
4345
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004346static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004347{
4348 struct mem_cgroup_eventfd_list *ev;
4349
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004350 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004351 eventfd_signal(ev->eventfd, 1);
4352 return 0;
4353}
4354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004355static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004356{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004357 struct mem_cgroup *iter;
4358
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004359 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004360 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004361}
4362
4363static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4364 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004365{
4366 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004367 struct mem_cgroup_thresholds *thresholds;
4368 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004369 int type = MEMFILE_TYPE(cft->private);
4370 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004371 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004372
4373 ret = res_counter_memparse_write_strategy(args, &threshold);
4374 if (ret)
4375 return ret;
4376
4377 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004378
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004379 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004380 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004381 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004382 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004383 else
4384 BUG();
4385
4386 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4387
4388 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004389 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004390 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4391
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004392 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004393
4394 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004395 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004396 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398 ret = -ENOMEM;
4399 goto unlock;
4400 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004401 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004402
4403 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004404 if (thresholds->primary) {
4405 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004406 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004407 }
4408
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004409 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004410 new->entries[size - 1].eventfd = eventfd;
4411 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004412
4413 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004414 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004415 compare_thresholds, NULL);
4416
4417 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004418 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004419 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004420 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004421 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004422 * new->current_threshold will not be used until
4423 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004424 * it here.
4425 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004426 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004427 }
4428 }
4429
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004430 /* Free old spare buffer and save old primary buffer as spare */
4431 kfree(thresholds->spare);
4432 thresholds->spare = thresholds->primary;
4433
4434 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004435
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004436 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004437 synchronize_rcu();
4438
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004439unlock:
4440 mutex_unlock(&memcg->thresholds_lock);
4441
4442 return ret;
4443}
4444
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004445static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004446 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004447{
4448 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004449 struct mem_cgroup_thresholds *thresholds;
4450 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004451 int type = MEMFILE_TYPE(cft->private);
4452 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004453 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004454
4455 mutex_lock(&memcg->thresholds_lock);
4456 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004457 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004458 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004459 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004460 else
4461 BUG();
4462
Anton Vorontsov371528c2012-02-24 05:14:46 +04004463 if (!thresholds->primary)
4464 goto unlock;
4465
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004466 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4467
4468 /* Check if a threshold crossed before removing */
4469 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4470
4471 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004472 size = 0;
4473 for (i = 0; i < thresholds->primary->size; i++) {
4474 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004475 size++;
4476 }
4477
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004478 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004479
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004480 /* Set thresholds array to NULL if we don't have thresholds */
4481 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004482 kfree(new);
4483 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004484 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004485 }
4486
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004487 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004488
4489 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004490 new->current_threshold = -1;
4491 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4492 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004493 continue;
4494
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004495 new->entries[j] = thresholds->primary->entries[i];
4496 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004497 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004498 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004499 * until rcu_assign_pointer(), so it's safe to increment
4500 * it here.
4501 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004502 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004503 }
4504 j++;
4505 }
4506
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004507swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004508 /* Swap primary and spare array */
4509 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004510 /* If all events are unregistered, free the spare array */
4511 if (!new) {
4512 kfree(thresholds->spare);
4513 thresholds->spare = NULL;
4514 }
4515
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004516 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004517
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004518 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004519 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004520unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004521 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004522}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004523
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004524static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4525 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4526{
4527 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4528 struct mem_cgroup_eventfd_list *event;
4529 int type = MEMFILE_TYPE(cft->private);
4530
4531 BUG_ON(type != _OOM_TYPE);
4532 event = kmalloc(sizeof(*event), GFP_KERNEL);
4533 if (!event)
4534 return -ENOMEM;
4535
Michal Hocko1af8efe2011-07-26 16:08:24 -07004536 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004537
4538 event->eventfd = eventfd;
4539 list_add(&event->list, &memcg->oom_notify);
4540
4541 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004542 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004543 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004544 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004545
4546 return 0;
4547}
4548
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004549static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004550 struct cftype *cft, struct eventfd_ctx *eventfd)
4551{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004552 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004553 struct mem_cgroup_eventfd_list *ev, *tmp;
4554 int type = MEMFILE_TYPE(cft->private);
4555
4556 BUG_ON(type != _OOM_TYPE);
4557
Michal Hocko1af8efe2011-07-26 16:08:24 -07004558 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004559
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004560 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004561 if (ev->eventfd == eventfd) {
4562 list_del(&ev->list);
4563 kfree(ev);
4564 }
4565 }
4566
Michal Hocko1af8efe2011-07-26 16:08:24 -07004567 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004568}
4569
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004570static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4571 struct cftype *cft, struct cgroup_map_cb *cb)
4572{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004573 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004574
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004575 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004576
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004577 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004578 cb->fill(cb, "under_oom", 1);
4579 else
4580 cb->fill(cb, "under_oom", 0);
4581 return 0;
4582}
4583
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004584static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4585 struct cftype *cft, u64 val)
4586{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004587 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004588 struct mem_cgroup *parent;
4589
4590 /* cannot set to root cgroup and only 0 and 1 are allowed */
4591 if (!cgrp->parent || !((val == 0) || (val == 1)))
4592 return -EINVAL;
4593
4594 parent = mem_cgroup_from_cont(cgrp->parent);
4595
4596 cgroup_lock();
4597 /* oom-kill-disable is a flag for subhierarchy. */
4598 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004599 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004600 cgroup_unlock();
4601 return -EINVAL;
4602 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004603 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004604 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004605 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004606 cgroup_unlock();
4607 return 0;
4608}
4609
Ying Han406eb0c2011-05-26 16:25:37 -07004610#ifdef CONFIG_NUMA
4611static const struct file_operations mem_control_numa_stat_file_operations = {
4612 .read = seq_read,
4613 .llseek = seq_lseek,
4614 .release = single_release,
4615};
4616
4617static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4618{
4619 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4620
4621 file->f_op = &mem_control_numa_stat_file_operations;
4622 return single_open(file, mem_control_numa_stat_show, cont);
4623}
4624#endif /* CONFIG_NUMA */
4625
Glauber Costae5671df2011-12-11 21:47:01 +00004626#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae5671df2011-12-11 21:47:01 +00004627static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4628{
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004629 /*
4630 * Part of this would be better living in a separate allocation
4631 * function, leaving us with just the cgroup tree population work.
4632 * We, however, depend on state such as network's proto_list that
4633 * is only initialized after cgroup creation. I found the less
4634 * cumbersome way to deal with it to defer it all to populate time
4635 */
Glauber Costa65c64ce2011-12-22 01:02:27 +00004636 return mem_cgroup_sockets_init(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004637};
4638
Li Zefan761b3ef2012-01-31 13:47:36 +08004639static void kmem_cgroup_destroy(struct cgroup *cont)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004640{
Li Zefan761b3ef2012-01-31 13:47:36 +08004641 mem_cgroup_sockets_destroy(cont);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004642}
Glauber Costae5671df2011-12-11 21:47:01 +00004643#else
4644static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4645{
4646 return 0;
4647}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004648
Li Zefan761b3ef2012-01-31 13:47:36 +08004649static void kmem_cgroup_destroy(struct cgroup *cont)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004650{
4651}
Glauber Costae5671df2011-12-11 21:47:01 +00004652#endif
4653
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004654static struct cftype mem_cgroup_files[] = {
4655 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004656 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004657 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004658 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004659 .register_event = mem_cgroup_usage_register_event,
4660 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004661 },
4662 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004663 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004664 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004665 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004666 .read_u64 = mem_cgroup_read,
4667 },
4668 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004669 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004670 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004671 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004672 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004673 },
4674 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004675 .name = "soft_limit_in_bytes",
4676 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4677 .write_string = mem_cgroup_write,
4678 .read_u64 = mem_cgroup_read,
4679 },
4680 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004681 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004682 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004683 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004684 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004685 },
Balbir Singh8697d332008-02-07 00:13:59 -08004686 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004687 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004688 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004689 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004690 {
4691 .name = "force_empty",
4692 .trigger = mem_cgroup_force_empty_write,
4693 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004694 {
4695 .name = "use_hierarchy",
4696 .write_u64 = mem_cgroup_hierarchy_write,
4697 .read_u64 = mem_cgroup_hierarchy_read,
4698 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004699 {
4700 .name = "swappiness",
4701 .read_u64 = mem_cgroup_swappiness_read,
4702 .write_u64 = mem_cgroup_swappiness_write,
4703 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004704 {
4705 .name = "move_charge_at_immigrate",
4706 .read_u64 = mem_cgroup_move_charge_read,
4707 .write_u64 = mem_cgroup_move_charge_write,
4708 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004709 {
4710 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004711 .read_map = mem_cgroup_oom_control_read,
4712 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004713 .register_event = mem_cgroup_oom_register_event,
4714 .unregister_event = mem_cgroup_oom_unregister_event,
4715 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4716 },
Ying Han406eb0c2011-05-26 16:25:37 -07004717#ifdef CONFIG_NUMA
4718 {
4719 .name = "numa_stat",
4720 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004721 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004722 },
4723#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004724};
4725
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004726#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4727static struct cftype memsw_cgroup_files[] = {
4728 {
4729 .name = "memsw.usage_in_bytes",
4730 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4731 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004732 .register_event = mem_cgroup_usage_register_event,
4733 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004734 },
4735 {
4736 .name = "memsw.max_usage_in_bytes",
4737 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4738 .trigger = mem_cgroup_reset,
4739 .read_u64 = mem_cgroup_read,
4740 },
4741 {
4742 .name = "memsw.limit_in_bytes",
4743 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4744 .write_string = mem_cgroup_write,
4745 .read_u64 = mem_cgroup_read,
4746 },
4747 {
4748 .name = "memsw.failcnt",
4749 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4750 .trigger = mem_cgroup_reset,
4751 .read_u64 = mem_cgroup_read,
4752 },
4753};
4754
4755static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4756{
4757 if (!do_swap_account)
4758 return 0;
4759 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4760 ARRAY_SIZE(memsw_cgroup_files));
4761};
4762#else
4763static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4764{
4765 return 0;
4766}
4767#endif
4768
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004769static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004770{
4771 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004772 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004773 enum lru_list lru;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004774 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004775 /*
4776 * This routine is called against possible nodes.
4777 * But it's BUG to call kmalloc() against offline node.
4778 *
4779 * TODO: this routine can waste much memory for nodes which will
4780 * never be onlined. It's better to use memory hotplug callback
4781 * function.
4782 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004783 if (!node_state(node, N_NORMAL_MEMORY))
4784 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004785 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004786 if (!pn)
4787 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004788
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004789 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4790 mz = &pn->zoneinfo[zone];
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004791 for_each_lru(lru)
4792 INIT_LIST_HEAD(&mz->lruvec.lists[lru]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004793 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004794 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004795 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004796 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004797 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004798 return 0;
4799}
4800
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004801static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004802{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004803 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004804}
4805
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004806static struct mem_cgroup *mem_cgroup_alloc(void)
4807{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004808 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004809 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004810
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004811 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004812 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004813 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004814 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004815 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004816
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004817 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004818 return NULL;
4819
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004820 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4821 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004822 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004823 spin_lock_init(&memcg->pcp_counter_lock);
4824 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004825
4826out_free:
4827 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004828 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004829 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004830 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004831 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004832}
4833
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004834/*
Hugh Dickins59927fb2012-03-15 15:17:07 -07004835 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4836 * but in process context. The work_freeing structure is overlaid
4837 * on the rcu_freeing structure, which itself is overlaid on memsw.
4838 */
4839static void vfree_work(struct work_struct *work)
4840{
4841 struct mem_cgroup *memcg;
4842
4843 memcg = container_of(work, struct mem_cgroup, work_freeing);
4844 vfree(memcg);
4845}
4846static void vfree_rcu(struct rcu_head *rcu_head)
4847{
4848 struct mem_cgroup *memcg;
4849
4850 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4851 INIT_WORK(&memcg->work_freeing, vfree_work);
4852 schedule_work(&memcg->work_freeing);
4853}
4854
4855/*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004856 * At destroying mem_cgroup, references from swap_cgroup can remain.
4857 * (scanning all at force_empty is too costly...)
4858 *
4859 * Instead of clearing all references at force_empty, we remember
4860 * the number of reference from swap_cgroup and free mem_cgroup when
4861 * it goes down to 0.
4862 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004863 * Removal of cgroup itself succeeds regardless of refs from swap.
4864 */
4865
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004866static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004867{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004868 int node;
4869
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004870 mem_cgroup_remove_from_trees(memcg);
4871 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004872
Bob Liu3ed28fa2012-01-12 17:19:04 -08004873 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004874 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004875
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004876 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004877 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004878 kfree_rcu(memcg, rcu_freeing);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004879 else
Hugh Dickins59927fb2012-03-15 15:17:07 -07004880 call_rcu(&memcg->rcu_freeing, vfree_rcu);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004881}
4882
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004883static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004884{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004885 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004886}
4887
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004888static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004890 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4891 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4892 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004893 if (parent)
4894 mem_cgroup_put(parent);
4895 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004896}
4897
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004898static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004899{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004900 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004901}
4902
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004903/*
4904 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4905 */
Glauber Costae1aab162011-12-11 21:47:03 +00004906struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004907{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004908 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004909 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004910 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004911}
Glauber Costae1aab162011-12-11 21:47:03 +00004912EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004913
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004914#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4915static void __init enable_swap_cgroup(void)
4916{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08004917 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004918 do_swap_account = 1;
4919}
4920#else
4921static void __init enable_swap_cgroup(void)
4922{
4923}
4924#endif
4925
Balbir Singhf64c3f52009-09-23 15:56:37 -07004926static int mem_cgroup_soft_limit_tree_init(void)
4927{
4928 struct mem_cgroup_tree_per_node *rtpn;
4929 struct mem_cgroup_tree_per_zone *rtpz;
4930 int tmp, node, zone;
4931
Bob Liu3ed28fa2012-01-12 17:19:04 -08004932 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004933 tmp = node;
4934 if (!node_state(node, N_NORMAL_MEMORY))
4935 tmp = -1;
4936 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4937 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004938 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004939
4940 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4941
4942 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4943 rtpz = &rtpn->rb_tree_per_zone[zone];
4944 rtpz->rb_root = RB_ROOT;
4945 spin_lock_init(&rtpz->lock);
4946 }
4947 }
4948 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004949
4950err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004951 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004952 if (!soft_limit_tree.rb_tree_per_node[node])
4953 break;
4954 kfree(soft_limit_tree.rb_tree_per_node[node]);
4955 soft_limit_tree.rb_tree_per_node[node] = NULL;
4956 }
4957 return 1;
4958
Balbir Singhf64c3f52009-09-23 15:56:37 -07004959}
4960
Li Zefan0eb253e2009-01-15 13:51:25 -08004961static struct cgroup_subsys_state * __ref
Li Zefan761b3ef2012-01-31 13:47:36 +08004962mem_cgroup_create(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004963{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004964 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004965 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004966 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004967
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004968 memcg = mem_cgroup_alloc();
4969 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004970 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004971
Bob Liu3ed28fa2012-01-12 17:19:04 -08004972 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004973 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004974 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004975
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004976 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004977 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004978 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004979 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004980 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004981 if (mem_cgroup_soft_limit_tree_init())
4982 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004983 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004984 for_each_possible_cpu(cpu) {
4985 struct memcg_stock_pcp *stock =
4986 &per_cpu(memcg_stock, cpu);
4987 INIT_WORK(&stock->work, drain_local_stock);
4988 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004989 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004990 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004991 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004992 memcg->use_hierarchy = parent->use_hierarchy;
4993 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004994 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004995
Balbir Singh18f59ea2009-01-07 18:08:07 -08004996 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004997 res_counter_init(&memcg->res, &parent->res);
4998 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004999 /*
5000 * We increment refcnt of the parent to ensure that we can
5001 * safely access it on res_counter_charge/uncharge.
5002 * This refcnt will be decremented when freeing this
5003 * mem_cgroup(see mem_cgroup_put).
5004 */
5005 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005006 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005007 res_counter_init(&memcg->res, NULL);
5008 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005009 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005010 memcg->last_scanned_node = MAX_NUMNODES;
5011 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08005012
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005013 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005014 memcg->swappiness = mem_cgroup_swappiness(parent);
5015 atomic_set(&memcg->refcnt, 1);
5016 memcg->move_charge_at_immigrate = 0;
5017 mutex_init(&memcg->thresholds_lock);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07005018 spin_lock_init(&memcg->move_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005019 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005020free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005021 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005022 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005023}
5024
Li Zefan761b3ef2012-01-31 13:47:36 +08005025static int mem_cgroup_pre_destroy(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005026{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005027 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005028
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005029 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005030}
5031
Li Zefan761b3ef2012-01-31 13:47:36 +08005032static void mem_cgroup_destroy(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005033{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005034 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005035
Li Zefan761b3ef2012-01-31 13:47:36 +08005036 kmem_cgroup_destroy(cont);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005037
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005038 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005039}
5040
5041static int mem_cgroup_populate(struct cgroup_subsys *ss,
5042 struct cgroup *cont)
5043{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005044 int ret;
5045
5046 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
5047 ARRAY_SIZE(mem_cgroup_files));
5048
5049 if (!ret)
5050 ret = register_memsw_files(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00005051
5052 if (!ret)
5053 ret = register_kmem_files(cont, ss);
5054
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005055 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005056}
5057
Daisuke Nishimura02491442010-03-10 15:22:17 -08005058#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005059/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005060#define PRECHARGE_COUNT_AT_ONCE 256
5061static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005062{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005063 int ret = 0;
5064 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005065 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005066
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005067 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005068 mc.precharge += count;
5069 /* we don't need css_get for root */
5070 return ret;
5071 }
5072 /* try to charge at once */
5073 if (count > 1) {
5074 struct res_counter *dummy;
5075 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005076 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005077 * by cgroup_lock_live_cgroup() that it is not removed and we
5078 * are still under the same cgroup_mutex. So we can postpone
5079 * css_get().
5080 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005081 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005082 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005083 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005084 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005085 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005086 goto one_by_one;
5087 }
5088 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005089 return ret;
5090 }
5091one_by_one:
5092 /* fall back to one by one charge */
5093 while (count--) {
5094 if (signal_pending(current)) {
5095 ret = -EINTR;
5096 break;
5097 }
5098 if (!batch_count--) {
5099 batch_count = PRECHARGE_COUNT_AT_ONCE;
5100 cond_resched();
5101 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005102 ret = __mem_cgroup_try_charge(NULL,
5103 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005104 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005105 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005106 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005107 mc.precharge++;
5108 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005109 return ret;
5110}
5111
5112/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005113 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005114 * @vma: the vma the pte to be checked belongs
5115 * @addr: the address corresponding to the pte to be checked
5116 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005117 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005118 *
5119 * Returns
5120 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5121 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5122 * move charge. if @target is not NULL, the page is stored in target->page
5123 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005124 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5125 * target for charge migration. if @target is not NULL, the entry is stored
5126 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005127 *
5128 * Called with pte lock held.
5129 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005130union mc_target {
5131 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005132 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005133};
5134
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005135enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005136 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005137 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005138 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005139};
5140
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005141static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5142 unsigned long addr, pte_t ptent)
5143{
5144 struct page *page = vm_normal_page(vma, addr, ptent);
5145
5146 if (!page || !page_mapped(page))
5147 return NULL;
5148 if (PageAnon(page)) {
5149 /* we don't move shared anon */
Hugh Dickinsbe22aec2012-03-09 13:37:32 -08005150 if (!move_anon() || page_mapcount(page) > 2)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005151 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005152 } else if (!move_file())
5153 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005154 return NULL;
5155 if (!get_page_unless_zero(page))
5156 return NULL;
5157
5158 return page;
5159}
5160
5161static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5162 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5163{
5164 int usage_count;
5165 struct page *page = NULL;
5166 swp_entry_t ent = pte_to_swp_entry(ptent);
5167
5168 if (!move_anon() || non_swap_entry(ent))
5169 return NULL;
5170 usage_count = mem_cgroup_count_swap_user(ent, &page);
5171 if (usage_count > 1) { /* we don't move shared anon */
5172 if (page)
5173 put_page(page);
5174 return NULL;
5175 }
5176 if (do_swap_account)
5177 entry->val = ent.val;
5178
5179 return page;
5180}
5181
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005182static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5183 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5184{
5185 struct page *page = NULL;
5186 struct inode *inode;
5187 struct address_space *mapping;
5188 pgoff_t pgoff;
5189
5190 if (!vma->vm_file) /* anonymous vma */
5191 return NULL;
5192 if (!move_file())
5193 return NULL;
5194
5195 inode = vma->vm_file->f_path.dentry->d_inode;
5196 mapping = vma->vm_file->f_mapping;
5197 if (pte_none(ptent))
5198 pgoff = linear_page_index(vma, addr);
5199 else /* pte_file(ptent) is true */
5200 pgoff = pte_to_pgoff(ptent);
5201
5202 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005203 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005204
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005205#ifdef CONFIG_SWAP
5206 /* shmem/tmpfs may report page out on swap: account for that too. */
5207 if (radix_tree_exceptional_entry(page)) {
5208 swp_entry_t swap = radix_to_swp_entry(page);
5209 if (do_swap_account)
5210 *entry = swap;
5211 page = find_get_page(&swapper_space, swap.val);
5212 }
5213#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005214 return page;
5215}
5216
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005217static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005218 unsigned long addr, pte_t ptent, union mc_target *target)
5219{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005220 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005221 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005222 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005223 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005224
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005225 if (pte_present(ptent))
5226 page = mc_handle_present_pte(vma, addr, ptent);
5227 else if (is_swap_pte(ptent))
5228 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005229 else if (pte_none(ptent) || pte_file(ptent))
5230 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005231
5232 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005233 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005234 if (page) {
5235 pc = lookup_page_cgroup(page);
5236 /*
5237 * Do only loose check w/o page_cgroup lock.
5238 * mem_cgroup_move_account() checks the pc is valid or not under
5239 * the lock.
5240 */
5241 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5242 ret = MC_TARGET_PAGE;
5243 if (target)
5244 target->page = page;
5245 }
5246 if (!ret || !target)
5247 put_page(page);
5248 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005249 /* There is a swap entry and a page doesn't exist or isn't charged */
5250 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005251 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005252 ret = MC_TARGET_SWAP;
5253 if (target)
5254 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005255 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005256 return ret;
5257}
5258
Naoya Horiguchi12724852012-03-21 16:34:28 -07005259#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5260/*
5261 * We don't consider swapping or file mapped pages because THP does not
5262 * support them for now.
5263 * Caller should make sure that pmd_trans_huge(pmd) is true.
5264 */
5265static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5266 unsigned long addr, pmd_t pmd, union mc_target *target)
5267{
5268 struct page *page = NULL;
5269 struct page_cgroup *pc;
5270 enum mc_target_type ret = MC_TARGET_NONE;
5271
5272 page = pmd_page(pmd);
5273 VM_BUG_ON(!page || !PageHead(page));
5274 if (!move_anon())
5275 return ret;
5276 pc = lookup_page_cgroup(page);
5277 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5278 ret = MC_TARGET_PAGE;
5279 if (target) {
5280 get_page(page);
5281 target->page = page;
5282 }
5283 }
5284 return ret;
5285}
5286#else
5287static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5288 unsigned long addr, pmd_t pmd, union mc_target *target)
5289{
5290 return MC_TARGET_NONE;
5291}
5292#endif
5293
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005294static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5295 unsigned long addr, unsigned long end,
5296 struct mm_walk *walk)
5297{
5298 struct vm_area_struct *vma = walk->private;
5299 pte_t *pte;
5300 spinlock_t *ptl;
5301
Naoya Horiguchi12724852012-03-21 16:34:28 -07005302 if (pmd_trans_huge_lock(pmd, vma) == 1) {
5303 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5304 mc.precharge += HPAGE_PMD_NR;
5305 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005306 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005307 }
Dave Hansen03319322011-03-22 16:32:56 -07005308
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005309 if (pmd_trans_unstable(pmd))
5310 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005311 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5312 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005313 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005314 mc.precharge++; /* increment precharge temporarily */
5315 pte_unmap_unlock(pte - 1, ptl);
5316 cond_resched();
5317
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005318 return 0;
5319}
5320
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005321static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5322{
5323 unsigned long precharge;
5324 struct vm_area_struct *vma;
5325
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005326 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005327 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5328 struct mm_walk mem_cgroup_count_precharge_walk = {
5329 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5330 .mm = mm,
5331 .private = vma,
5332 };
5333 if (is_vm_hugetlb_page(vma))
5334 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005335 walk_page_range(vma->vm_start, vma->vm_end,
5336 &mem_cgroup_count_precharge_walk);
5337 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005338 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005339
5340 precharge = mc.precharge;
5341 mc.precharge = 0;
5342
5343 return precharge;
5344}
5345
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005346static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5347{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005348 unsigned long precharge = mem_cgroup_count_precharge(mm);
5349
5350 VM_BUG_ON(mc.moving_task);
5351 mc.moving_task = current;
5352 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005353}
5354
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005355/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5356static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005357{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005358 struct mem_cgroup *from = mc.from;
5359 struct mem_cgroup *to = mc.to;
5360
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005361 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005362 if (mc.precharge) {
5363 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5364 mc.precharge = 0;
5365 }
5366 /*
5367 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5368 * we must uncharge here.
5369 */
5370 if (mc.moved_charge) {
5371 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5372 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005373 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005374 /* we must fixup refcnts and charges */
5375 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005376 /* uncharge swap account from the old cgroup */
5377 if (!mem_cgroup_is_root(mc.from))
5378 res_counter_uncharge(&mc.from->memsw,
5379 PAGE_SIZE * mc.moved_swap);
5380 __mem_cgroup_put(mc.from, mc.moved_swap);
5381
5382 if (!mem_cgroup_is_root(mc.to)) {
5383 /*
5384 * we charged both to->res and to->memsw, so we should
5385 * uncharge to->res.
5386 */
5387 res_counter_uncharge(&mc.to->res,
5388 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005389 }
5390 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005391 mc.moved_swap = 0;
5392 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005393 memcg_oom_recover(from);
5394 memcg_oom_recover(to);
5395 wake_up_all(&mc.waitq);
5396}
5397
5398static void mem_cgroup_clear_mc(void)
5399{
5400 struct mem_cgroup *from = mc.from;
5401
5402 /*
5403 * we must clear moving_task before waking up waiters at the end of
5404 * task migration.
5405 */
5406 mc.moving_task = NULL;
5407 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005408 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005409 mc.from = NULL;
5410 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005411 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005412 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005413}
5414
Li Zefan761b3ef2012-01-31 13:47:36 +08005415static int mem_cgroup_can_attach(struct cgroup *cgroup,
5416 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005417{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005418 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005419 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005420 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005421
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005422 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005423 struct mm_struct *mm;
5424 struct mem_cgroup *from = mem_cgroup_from_task(p);
5425
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005426 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005427
5428 mm = get_task_mm(p);
5429 if (!mm)
5430 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005431 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005432 if (mm->owner == p) {
5433 VM_BUG_ON(mc.from);
5434 VM_BUG_ON(mc.to);
5435 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005436 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005437 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005438 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005439 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005440 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005441 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005442 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005443 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005444
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005445 ret = mem_cgroup_precharge_mc(mm);
5446 if (ret)
5447 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005448 }
5449 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005450 }
5451 return ret;
5452}
5453
Li Zefan761b3ef2012-01-31 13:47:36 +08005454static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5455 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005456{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005457 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005458}
5459
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005460static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5461 unsigned long addr, unsigned long end,
5462 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005463{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005464 int ret = 0;
5465 struct vm_area_struct *vma = walk->private;
5466 pte_t *pte;
5467 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005468 enum mc_target_type target_type;
5469 union mc_target target;
5470 struct page *page;
5471 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005472
Naoya Horiguchi12724852012-03-21 16:34:28 -07005473 /*
5474 * We don't take compound_lock() here but no race with splitting thp
5475 * happens because:
5476 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5477 * under splitting, which means there's no concurrent thp split,
5478 * - if another thread runs into split_huge_page() just after we
5479 * entered this if-block, the thread must wait for page table lock
5480 * to be unlocked in __split_huge_page_splitting(), where the main
5481 * part of thp split is not executed yet.
5482 */
5483 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005484 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005485 spin_unlock(&vma->vm_mm->page_table_lock);
5486 return 0;
5487 }
5488 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5489 if (target_type == MC_TARGET_PAGE) {
5490 page = target.page;
5491 if (!isolate_lru_page(page)) {
5492 pc = lookup_page_cgroup(page);
5493 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5494 pc, mc.from, mc.to,
5495 false)) {
5496 mc.precharge -= HPAGE_PMD_NR;
5497 mc.moved_charge += HPAGE_PMD_NR;
5498 }
5499 putback_lru_page(page);
5500 }
5501 put_page(page);
5502 }
5503 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005504 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005505 }
5506
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005507 if (pmd_trans_unstable(pmd))
5508 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005509retry:
5510 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5511 for (; addr != end; addr += PAGE_SIZE) {
5512 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08005513 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005514
5515 if (!mc.precharge)
5516 break;
5517
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005518 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005519 case MC_TARGET_PAGE:
5520 page = target.page;
5521 if (isolate_lru_page(page))
5522 goto put;
5523 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005524 if (!mem_cgroup_move_account(page, 1, pc,
5525 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005526 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005527 /* we uncharge from mc.from later. */
5528 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005529 }
5530 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005531put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005532 put_page(page);
5533 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005534 case MC_TARGET_SWAP:
5535 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005536 if (!mem_cgroup_move_swap_account(ent,
5537 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005538 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005539 /* we fixup refcnts and charges later. */
5540 mc.moved_swap++;
5541 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005542 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005543 default:
5544 break;
5545 }
5546 }
5547 pte_unmap_unlock(pte - 1, ptl);
5548 cond_resched();
5549
5550 if (addr != end) {
5551 /*
5552 * We have consumed all precharges we got in can_attach().
5553 * We try charge one by one, but don't do any additional
5554 * charges to mc.to if we have failed in charge once in attach()
5555 * phase.
5556 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005557 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005558 if (!ret)
5559 goto retry;
5560 }
5561
5562 return ret;
5563}
5564
5565static void mem_cgroup_move_charge(struct mm_struct *mm)
5566{
5567 struct vm_area_struct *vma;
5568
5569 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005570retry:
5571 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5572 /*
5573 * Someone who are holding the mmap_sem might be waiting in
5574 * waitq. So we cancel all extra charges, wake up all waiters,
5575 * and retry. Because we cancel precharges, we might not be able
5576 * to move enough charges, but moving charge is a best-effort
5577 * feature anyway, so it wouldn't be a big problem.
5578 */
5579 __mem_cgroup_clear_mc();
5580 cond_resched();
5581 goto retry;
5582 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005583 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5584 int ret;
5585 struct mm_walk mem_cgroup_move_charge_walk = {
5586 .pmd_entry = mem_cgroup_move_charge_pte_range,
5587 .mm = mm,
5588 .private = vma,
5589 };
5590 if (is_vm_hugetlb_page(vma))
5591 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005592 ret = walk_page_range(vma->vm_start, vma->vm_end,
5593 &mem_cgroup_move_charge_walk);
5594 if (ret)
5595 /*
5596 * means we have consumed all precharges and failed in
5597 * doing additional charge. Just abandon here.
5598 */
5599 break;
5600 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005601 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005602}
5603
Li Zefan761b3ef2012-01-31 13:47:36 +08005604static void mem_cgroup_move_task(struct cgroup *cont,
5605 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005606{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005607 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005608 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005609
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005610 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005611 if (mc.to)
5612 mem_cgroup_move_charge(mm);
5613 put_swap_token(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005614 mmput(mm);
5615 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005616 if (mc.to)
5617 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005618}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005619#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08005620static int mem_cgroup_can_attach(struct cgroup *cgroup,
5621 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005622{
5623 return 0;
5624}
Li Zefan761b3ef2012-01-31 13:47:36 +08005625static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5626 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005627{
5628}
Li Zefan761b3ef2012-01-31 13:47:36 +08005629static void mem_cgroup_move_task(struct cgroup *cont,
5630 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005631{
5632}
5633#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005634
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005635struct cgroup_subsys mem_cgroup_subsys = {
5636 .name = "memory",
5637 .subsys_id = mem_cgroup_subsys_id,
5638 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005639 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005640 .destroy = mem_cgroup_destroy,
5641 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005642 .can_attach = mem_cgroup_can_attach,
5643 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005644 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005645 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005646 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005647};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005648
5649#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005650static int __init enable_swap_account(char *s)
5651{
5652 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005653 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005654 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005655 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005656 really_do_swap_account = 0;
5657 return 1;
5658}
Michal Hockoa2c89902011-05-24 17:12:50 -07005659__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005660
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005661#endif