blob: 010d6c14129ae320ab9655a573d49af33e034659 [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 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080017 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
21 *
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
26 */
27
28#include <linux/res_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080031#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080032#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080033#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
48#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080049#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080050#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070051#include <linux/vmalloc.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070054#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080057#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000058#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070059#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000060#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080061
Balbir Singh8697d332008-02-07 00:13:59 -080062#include <asm/uaccess.h>
63
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070064#include <trace/events/vmscan.h>
65
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070066struct cgroup_subsys mem_cgroup_subsys __read_mostly;
David Rientjes68ae5642012-12-12 13:51:57 -080067EXPORT_SYMBOL(mem_cgroup_subsys);
68
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070069#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070070static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080071
Andrew Mortonc255a452012-07-31 16:43:02 -070072#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070073/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080074int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080075
76/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070077#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080078static int really_do_swap_account __initdata = 1;
79#else
80static int really_do_swap_account __initdata = 0;
81#endif
82
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080083#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070084#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#endif
86
87
Balbir Singh8cdea7c2008-02-07 00:13:50 -080088/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080089 * Statistics for memory cgroup.
90 */
91enum mem_cgroup_stat_index {
92 /*
93 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
94 */
David Rientjesb070e652013-05-07 16:18:09 -070095 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
96 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
97 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
98 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
99 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800100 MEM_CGROUP_STAT_NSTATS,
101};
102
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700103static const char * const mem_cgroup_stat_names[] = {
104 "cache",
105 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700106 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107 "mapped_file",
108 "swap",
109};
110
Johannes Weinere9f89742011-03-23 16:42:37 -0700111enum mem_cgroup_events_index {
112 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
113 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700114 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
115 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700116 MEM_CGROUP_EVENTS_NSTATS,
117};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700118
119static const char * const mem_cgroup_events_names[] = {
120 "pgpgin",
121 "pgpgout",
122 "pgfault",
123 "pgmajfault",
124};
125
Sha Zhengju58cf1882013-02-22 16:32:05 -0800126static const char * const mem_cgroup_lru_names[] = {
127 "inactive_anon",
128 "active_anon",
129 "inactive_file",
130 "active_file",
131 "unevictable",
132};
133
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700134/*
135 * Per memcg event counter is incremented at every pagein/pageout. With THP,
136 * it will be incremated by the number of pages. This counter is used for
137 * for trigger some periodic events. This is straightforward and better
138 * than using jiffies etc. to handle periodic memcg event.
139 */
140enum mem_cgroup_events_target {
141 MEM_CGROUP_TARGET_THRESH,
142 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700143 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700144 MEM_CGROUP_NTARGETS,
145};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700146#define THRESHOLDS_EVENTS_TARGET 128
147#define SOFTLIMIT_EVENTS_TARGET 1024
148#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700149
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800150struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700151 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700152 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700153 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700154 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800155};
156
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800157struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700158 /*
159 * last scanned hierarchy member. Valid only if last_dead_count
160 * matches memcg->dead_count of the hierarchy root group.
161 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700162 struct mem_cgroup *last_visited;
Michal Hocko5f578162013-04-29 15:07:17 -0700163 unsigned long last_dead_count;
164
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800165 /* scan generation, increased every round-trip */
166 unsigned int generation;
167};
168
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800169/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800170 * per-zone information in memory controller.
171 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800172struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800173 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700174 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800175
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800176 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
177
Balbir Singhf64c3f52009-09-23 15:56:37 -0700178 struct rb_node tree_node; /* RB tree node */
179 unsigned long long usage_in_excess;/* Set to the value by which */
180 /* the soft limit is exceeded*/
181 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700182 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700183 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800184};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800185
186struct mem_cgroup_per_node {
187 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
188};
189
190struct mem_cgroup_lru_info {
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800191 struct mem_cgroup_per_node *nodeinfo[0];
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800192};
193
194/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700195 * Cgroups above their limits are maintained in a RB-Tree, independent of
196 * their hierarchy representation
197 */
198
199struct mem_cgroup_tree_per_zone {
200 struct rb_root rb_root;
201 spinlock_t lock;
202};
203
204struct mem_cgroup_tree_per_node {
205 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
206};
207
208struct mem_cgroup_tree {
209 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
210};
211
212static struct mem_cgroup_tree soft_limit_tree __read_mostly;
213
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800214struct mem_cgroup_threshold {
215 struct eventfd_ctx *eventfd;
216 u64 threshold;
217};
218
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700219/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800220struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700221 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700222 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800223 /* Size of entries[] */
224 unsigned int size;
225 /* Array of thresholds */
226 struct mem_cgroup_threshold entries[0];
227};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700228
229struct mem_cgroup_thresholds {
230 /* Primary thresholds array */
231 struct mem_cgroup_threshold_ary *primary;
232 /*
233 * Spare threshold array.
234 * This is needed to make mem_cgroup_unregister_event() "never fail".
235 * It must be able to store at least primary->size - 1 entries.
236 */
237 struct mem_cgroup_threshold_ary *spare;
238};
239
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700240/* for OOM */
241struct mem_cgroup_eventfd_list {
242 struct list_head list;
243 struct eventfd_ctx *eventfd;
244};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800245
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700246static void mem_cgroup_threshold(struct mem_cgroup *memcg);
247static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800248
Balbir Singhf64c3f52009-09-23 15:56:37 -0700249/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800250 * The memory controller data structure. The memory controller controls both
251 * page cache and RSS per cgroup. We would eventually like to provide
252 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
253 * to help the administrator determine what knobs to tune.
254 *
255 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800256 * we hit the water mark. May be even add a low water mark, such that
257 * no reclaim occurs from a cgroup at it's low water mark, this is
258 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800259 */
260struct mem_cgroup {
261 struct cgroup_subsys_state css;
262 /*
263 * the counter to account for memory usage
264 */
265 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700266
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700267 /* vmpressure notifications */
268 struct vmpressure vmpressure;
269
Hugh Dickins59927fb2012-03-15 15:17:07 -0700270 union {
271 /*
272 * the counter to account for mem+swap usage.
273 */
274 struct res_counter memsw;
275
276 /*
277 * rcu_freeing is used only when freeing struct mem_cgroup,
278 * so put it into a union to avoid wasting more memory.
279 * It must be disjoint from the css field. It could be
280 * in a union with the res field, but res plays a much
281 * larger part in mem_cgroup life than memsw, and might
282 * be of interest, even at time of free, when debugging.
283 * So share rcu_head with the less interesting memsw.
284 */
285 struct rcu_head rcu_freeing;
286 /*
Glauber Costa3afe36b2012-05-29 15:07:10 -0700287 * We also need some space for a worker in deferred freeing.
288 * By the time we call it, rcu_freeing is no longer in use.
Hugh Dickins59927fb2012-03-15 15:17:07 -0700289 */
290 struct work_struct work_freeing;
291 };
292
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800293 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800294 * the counter to account for kernel memory usage.
295 */
296 struct res_counter kmem;
297 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800298 * Should the accounting and control be hierarchical, per subtree?
299 */
300 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800301 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700302
303 bool oom_lock;
304 atomic_t under_oom;
305
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800306 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800307
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700308 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700309 /* OOM-Killer disable */
310 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800311
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700312 /* set when res.limit == memsw.limit */
313 bool memsw_is_minimum;
314
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800315 /* protect arrays of thresholds */
316 struct mutex thresholds_lock;
317
318 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700319 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700320
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800321 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700322 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700323
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700324 /* For oom notifier event fd */
325 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700326
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800327 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800328 * Should we move charges of a task when a task is moved into this
329 * mem_cgroup ? And what type of charges should we move ?
330 */
331 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800332 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700333 * set > 0 if pages under this cgroup are moving to other cgroup.
334 */
335 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700336 /* taken only while moving_account > 0 */
337 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700338 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800339 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800340 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700341 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700342 /*
343 * used when a cpu is offlined or other synchronizations
344 * See mem_cgroup_read_stat().
345 */
346 struct mem_cgroup_stat_cpu nocpu_base;
347 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000348
Michal Hocko5f578162013-04-29 15:07:17 -0700349 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700350#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000351 struct tcp_memcontrol tcp_mem;
352#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800353#if defined(CONFIG_MEMCG_KMEM)
354 /* analogous to slab_common's slab_caches list. per-memcg */
355 struct list_head memcg_slab_caches;
356 /* Not a spinlock, we can take a lot of time walking the list */
357 struct mutex slab_caches_mutex;
358 /* Index in the kmem_cache->memcg_params->memcg_caches array */
359 int kmemcg_id;
360#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800361
362 int last_scanned_node;
363#if MAX_NUMNODES > 1
364 nodemask_t scan_nodes;
365 atomic_t numainfo_events;
366 atomic_t numainfo_updating;
367#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700368
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800369 /*
370 * Per cgroup active and inactive list, similar to the
371 * per zone LRU lists.
372 *
373 * WARNING: This has to be the last element of the struct. Don't
374 * add new fields after this point.
375 */
376 struct mem_cgroup_lru_info info;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800377};
378
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800379static size_t memcg_size(void)
380{
381 return sizeof(struct mem_cgroup) +
382 nr_node_ids * sizeof(struct mem_cgroup_per_node);
383}
384
Glauber Costa510fc4e2012-12-18 14:21:47 -0800385/* internal only representation about the status of kmem accounting. */
386enum {
387 KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */
Glauber Costaa8964b92012-12-18 14:22:09 -0800388 KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */
Glauber Costa7de37682012-12-18 14:22:07 -0800389 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800390};
391
Glauber Costaa8964b92012-12-18 14:22:09 -0800392/* We account when limit is on, but only after call sites are patched */
393#define KMEM_ACCOUNTED_MASK \
394 ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED))
Glauber Costa510fc4e2012-12-18 14:21:47 -0800395
396#ifdef CONFIG_MEMCG_KMEM
397static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
398{
399 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
400}
Glauber Costa7de37682012-12-18 14:22:07 -0800401
402static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
403{
404 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
405}
406
Glauber Costaa8964b92012-12-18 14:22:09 -0800407static void memcg_kmem_set_activated(struct mem_cgroup *memcg)
408{
409 set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
410}
411
Glauber Costa55007d82012-12-18 14:22:38 -0800412static void memcg_kmem_clear_activated(struct mem_cgroup *memcg)
413{
414 clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
415}
416
Glauber Costa7de37682012-12-18 14:22:07 -0800417static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
418{
419 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
420 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
421}
422
423static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
424{
425 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
426 &memcg->kmem_account_flags);
427}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800428#endif
429
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800430/* Stuffs for move charges at task migration. */
431/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800432 * Types of charges to be moved. "move_charge_at_immitgrate" and
433 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800434 */
435enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800436 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700437 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800438 NR_MOVE_TYPE,
439};
440
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800441/* "mc" and its members are protected by cgroup_mutex */
442static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800443 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800444 struct mem_cgroup *from;
445 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800446 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800447 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800448 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800449 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800450 struct task_struct *moving_task; /* a task moving charges */
451 wait_queue_head_t waitq; /* a waitq for other context */
452} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700453 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800454 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
455};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800456
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700457static bool move_anon(void)
458{
Glauber Costaee5e8472013-02-22 16:34:50 -0800459 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700460}
461
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700462static bool move_file(void)
463{
Glauber Costaee5e8472013-02-22 16:34:50 -0800464 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700465}
466
Balbir Singh4e416952009-09-23 15:56:39 -0700467/*
468 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
469 * limit reclaim to prevent infinite loops, if they ever occur.
470 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700471#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
472#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700473
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800474enum charge_type {
475 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700476 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800477 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700478 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700479 NR_CHARGE_TYPE,
480};
481
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800482/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800483enum res_type {
484 _MEM,
485 _MEMSWAP,
486 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800487 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800488};
489
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700490#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
491#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800492#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700493/* Used for OOM nofiier */
494#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800495
Balbir Singh75822b42009-09-23 15:56:38 -0700496/*
497 * Reclaim flags for mem_cgroup_hierarchical_reclaim
498 */
499#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
500#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
501#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
502#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
503
Glauber Costa09998212013-02-22 16:34:55 -0800504/*
505 * The memcg_create_mutex will be held whenever a new cgroup is created.
506 * As a consequence, any change that needs to protect against new child cgroups
507 * appearing has to hold it as well.
508 */
509static DEFINE_MUTEX(memcg_create_mutex);
510
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700511static void mem_cgroup_get(struct mem_cgroup *memcg);
512static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000513
Wanpeng Lib2145142012-07-31 16:46:01 -0700514static inline
515struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
516{
517 return container_of(s, struct mem_cgroup, css);
518}
519
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700520/* Some nice accessors for the vmpressure. */
521struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
522{
523 if (!memcg)
524 memcg = root_mem_cgroup;
525 return &memcg->vmpressure;
526}
527
528struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
529{
530 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
531}
532
533struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
534{
535 return &mem_cgroup_from_css(css)->vmpressure;
536}
537
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700538static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
539{
540 return (memcg == root_mem_cgroup);
541}
542
Glauber Costae1aab162011-12-11 21:47:03 +0000543/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700544#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000545
Glauber Costae1aab162011-12-11 21:47:03 +0000546void sock_update_memcg(struct sock *sk)
547{
Glauber Costa376be5f2012-01-20 04:57:14 +0000548 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000549 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700550 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000551
552 BUG_ON(!sk->sk_prot->proto_cgroup);
553
Glauber Costaf3f511e2012-01-05 20:16:39 +0000554 /* Socket cloning can throw us here with sk_cgrp already
555 * filled. It won't however, necessarily happen from
556 * process context. So the test for root memcg given
557 * the current task's memcg won't help us in this case.
558 *
559 * Respecting the original socket's memcg is a better
560 * decision in this case.
561 */
562 if (sk->sk_cgrp) {
563 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
564 mem_cgroup_get(sk->sk_cgrp->memcg);
565 return;
566 }
567
Glauber Costae1aab162011-12-11 21:47:03 +0000568 rcu_read_lock();
569 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700570 cg_proto = sk->sk_prot->proto_cgroup(memcg);
571 if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
Glauber Costae1aab162011-12-11 21:47:03 +0000572 mem_cgroup_get(memcg);
Glauber Costa3f134612012-05-29 15:07:11 -0700573 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000574 }
575 rcu_read_unlock();
576 }
577}
578EXPORT_SYMBOL(sock_update_memcg);
579
580void sock_release_memcg(struct sock *sk)
581{
Glauber Costa376be5f2012-01-20 04:57:14 +0000582 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000583 struct mem_cgroup *memcg;
584 WARN_ON(!sk->sk_cgrp->memcg);
585 memcg = sk->sk_cgrp->memcg;
586 mem_cgroup_put(memcg);
587 }
588}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000589
590struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
591{
592 if (!memcg || mem_cgroup_is_root(memcg))
593 return NULL;
594
595 return &memcg->tcp_mem.cg_proto;
596}
597EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000598
Glauber Costa3f134612012-05-29 15:07:11 -0700599static void disarm_sock_keys(struct mem_cgroup *memcg)
600{
601 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
602 return;
603 static_key_slow_dec(&memcg_socket_limit_enabled);
604}
605#else
606static void disarm_sock_keys(struct mem_cgroup *memcg)
607{
608}
609#endif
610
Glauber Costaa8964b92012-12-18 14:22:09 -0800611#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800612/*
613 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
614 * There are two main reasons for not using the css_id for this:
615 * 1) this works better in sparse environments, where we have a lot of memcgs,
616 * but only a few kmem-limited. Or also, if we have, for instance, 200
617 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
618 * 200 entry array for that.
619 *
620 * 2) In order not to violate the cgroup API, we would like to do all memory
621 * allocation in ->create(). At that point, we haven't yet allocated the
622 * css_id. Having a separate index prevents us from messing with the cgroup
623 * core for this
624 *
625 * The current size of the caches array is stored in
626 * memcg_limited_groups_array_size. It will double each time we have to
627 * increase it.
628 */
629static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800630int memcg_limited_groups_array_size;
631
Glauber Costa55007d82012-12-18 14:22:38 -0800632/*
633 * MIN_SIZE is different than 1, because we would like to avoid going through
634 * the alloc/free process all the time. In a small machine, 4 kmem-limited
635 * cgroups is a reasonable guess. In the future, it could be a parameter or
636 * tunable, but that is strictly not necessary.
637 *
638 * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get
639 * this constant directly from cgroup, but it is understandable that this is
640 * better kept as an internal representation in cgroup.c. In any case, the
641 * css_id space is not getting any smaller, and we don't have to necessarily
642 * increase ours as well if it increases.
643 */
644#define MEMCG_CACHES_MIN_SIZE 4
645#define MEMCG_CACHES_MAX_SIZE 65535
646
Glauber Costad7f25f82012-12-18 14:22:40 -0800647/*
648 * A lot of the calls to the cache allocation functions are expected to be
649 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
650 * conditional to this static branch, we'll have to allow modules that does
651 * kmem_cache_alloc and the such to see this symbol as well
652 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800653struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800654EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800655
656static void disarm_kmem_keys(struct mem_cgroup *memcg)
657{
Glauber Costa55007d82012-12-18 14:22:38 -0800658 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800659 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800660 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
661 }
Glauber Costabea207c2012-12-18 14:22:11 -0800662 /*
663 * This check can't live in kmem destruction function,
664 * since the charges will outlive the cgroup
665 */
666 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800667}
668#else
669static void disarm_kmem_keys(struct mem_cgroup *memcg)
670{
671}
672#endif /* CONFIG_MEMCG_KMEM */
673
674static void disarm_static_keys(struct mem_cgroup *memcg)
675{
676 disarm_sock_keys(memcg);
677 disarm_kmem_keys(memcg);
678}
679
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800681
Balbir Singhf64c3f52009-09-23 15:56:37 -0700682static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700683mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700684{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800685 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700687}
688
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700689struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100690{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700691 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100692}
693
Balbir Singhf64c3f52009-09-23 15:56:37 -0700694static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700695page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700696{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700697 int nid = page_to_nid(page);
698 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700699
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700700 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700701}
702
703static struct mem_cgroup_tree_per_zone *
704soft_limit_tree_node_zone(int nid, int zid)
705{
706 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
707}
708
709static struct mem_cgroup_tree_per_zone *
710soft_limit_tree_from_page(struct page *page)
711{
712 int nid = page_to_nid(page);
713 int zid = page_zonenum(page);
714
715 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
716}
717
718static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700719__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700720 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700721 struct mem_cgroup_tree_per_zone *mctz,
722 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700723{
724 struct rb_node **p = &mctz->rb_root.rb_node;
725 struct rb_node *parent = NULL;
726 struct mem_cgroup_per_zone *mz_node;
727
728 if (mz->on_tree)
729 return;
730
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700731 mz->usage_in_excess = new_usage_in_excess;
732 if (!mz->usage_in_excess)
733 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700734 while (*p) {
735 parent = *p;
736 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
737 tree_node);
738 if (mz->usage_in_excess < mz_node->usage_in_excess)
739 p = &(*p)->rb_left;
740 /*
741 * We can't avoid mem cgroups that are over their soft
742 * limit by the same amount
743 */
744 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
745 p = &(*p)->rb_right;
746 }
747 rb_link_node(&mz->tree_node, parent, p);
748 rb_insert_color(&mz->tree_node, &mctz->rb_root);
749 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700750}
751
752static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700753__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700754 struct mem_cgroup_per_zone *mz,
755 struct mem_cgroup_tree_per_zone *mctz)
756{
757 if (!mz->on_tree)
758 return;
759 rb_erase(&mz->tree_node, &mctz->rb_root);
760 mz->on_tree = false;
761}
762
763static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700765 struct mem_cgroup_per_zone *mz,
766 struct mem_cgroup_tree_per_zone *mctz)
767{
768 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700769 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700770 spin_unlock(&mctz->lock);
771}
772
Balbir Singhf64c3f52009-09-23 15:56:37 -0700773
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700774static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700775{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700776 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700777 struct mem_cgroup_per_zone *mz;
778 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700779 int nid = page_to_nid(page);
780 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700781 mctz = soft_limit_tree_from_page(page);
782
783 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700784 * Necessary to update all ancestors when hierarchy is used.
785 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700786 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700787 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
788 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
789 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700790 /*
791 * We have to update the tree if mz is on RB-tree or
792 * mem is over its softlimit.
793 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700794 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700795 spin_lock(&mctz->lock);
796 /* if on-tree, remove it */
797 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700798 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700799 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700800 * Insert again. mz->usage_in_excess will be updated.
801 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700802 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700803 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700804 spin_unlock(&mctz->lock);
805 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700806 }
807}
808
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700809static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700810{
811 int node, zone;
812 struct mem_cgroup_per_zone *mz;
813 struct mem_cgroup_tree_per_zone *mctz;
814
Bob Liu3ed28fa2012-01-12 17:19:04 -0800815 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700816 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700817 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700818 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700819 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700820 }
821 }
822}
823
Balbir Singh4e416952009-09-23 15:56:39 -0700824static struct mem_cgroup_per_zone *
825__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
826{
827 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700828 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700829
830retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700831 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700832 rightmost = rb_last(&mctz->rb_root);
833 if (!rightmost)
834 goto done; /* Nothing to reclaim from */
835
836 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
837 /*
838 * Remove the node now but someone else can add it back,
839 * we will to add it back at the end of reclaim to its correct
840 * position in the tree.
841 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700842 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
843 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
844 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700845 goto retry;
846done:
847 return mz;
848}
849
850static struct mem_cgroup_per_zone *
851mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
852{
853 struct mem_cgroup_per_zone *mz;
854
855 spin_lock(&mctz->lock);
856 mz = __mem_cgroup_largest_soft_limit_node(mctz);
857 spin_unlock(&mctz->lock);
858 return mz;
859}
860
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700861/*
862 * Implementation Note: reading percpu statistics for memcg.
863 *
864 * Both of vmstat[] and percpu_counter has threshold and do periodic
865 * synchronization to implement "quick" read. There are trade-off between
866 * reading cost and precision of value. Then, we may have a chance to implement
867 * a periodic synchronizion of counter in memcg's counter.
868 *
869 * But this _read() function is used for user interface now. The user accounts
870 * memory usage by memory cgroup and he _always_ requires exact value because
871 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
872 * have to visit all online cpus and make sum. So, for now, unnecessary
873 * synchronization is not implemented. (just implemented for cpu hotplug)
874 *
875 * If there are kernel internal actions which can make use of some not-exact
876 * value, and reading all cpu value can be performance bottleneck in some
877 * common workload, threashold and synchonization as vmstat[] should be
878 * implemented.
879 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700880static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700881 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800882{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700883 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800884 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800885
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700886 get_online_cpus();
887 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700888 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700889#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700890 spin_lock(&memcg->pcp_counter_lock);
891 val += memcg->nocpu_base.count[idx];
892 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700893#endif
894 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800895 return val;
896}
897
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700898static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700899 bool charge)
900{
901 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700902 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700903}
904
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700905static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700906 enum mem_cgroup_events_index idx)
907{
908 unsigned long val = 0;
909 int cpu;
910
911 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700912 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700913#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700914 spin_lock(&memcg->pcp_counter_lock);
915 val += memcg->nocpu_base.events[idx];
916 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700917#endif
918 return val;
919}
920
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700921static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700922 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700923 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800924{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800925 preempt_disable();
926
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700927 /*
928 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
929 * counted as CACHE even if it's on ANON LRU.
930 */
931 if (anon)
932 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700933 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800934 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700935 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700936 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700937
David Rientjesb070e652013-05-07 16:18:09 -0700938 if (PageTransHuge(page))
939 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
940 nr_pages);
941
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800942 /* pagein of a big page is an event. So, ignore page size */
943 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700944 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800945 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700946 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800947 nr_pages = -nr_pages; /* for event */
948 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800949
Johannes Weiner13114712012-05-29 15:07:07 -0700950 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800951
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800952 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800953}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800954
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700955unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700956mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700957{
958 struct mem_cgroup_per_zone *mz;
959
960 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
961 return mz->lru_size[lru];
962}
963
964static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700965mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700966 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700967{
968 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700969 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700970 unsigned long ret = 0;
971
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700972 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700973
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700974 for_each_lru(lru) {
975 if (BIT(lru) & lru_mask)
976 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700977 }
978 return ret;
979}
980
981static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700982mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700983 int nid, unsigned int lru_mask)
984{
Ying Han889976d2011-05-26 16:25:33 -0700985 u64 total = 0;
986 int zid;
987
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700988 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700989 total += mem_cgroup_zone_nr_lru_pages(memcg,
990 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700991
Ying Han889976d2011-05-26 16:25:33 -0700992 return total;
993}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700994
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700995static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700996 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800997{
Ying Han889976d2011-05-26 16:25:33 -0700998 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800999 u64 total = 0;
1000
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001001 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001002 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001003 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -08001004}
1005
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001006static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
1007 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001008{
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001009 unsigned long val, next;
1010
Johannes Weiner13114712012-05-29 15:07:07 -07001011 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -07001012 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001013 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001014 if ((long)next - (long)val < 0) {
1015 switch (target) {
1016 case MEM_CGROUP_TARGET_THRESH:
1017 next = val + THRESHOLDS_EVENTS_TARGET;
1018 break;
1019 case MEM_CGROUP_TARGET_SOFTLIMIT:
1020 next = val + SOFTLIMIT_EVENTS_TARGET;
1021 break;
1022 case MEM_CGROUP_TARGET_NUMAINFO:
1023 next = val + NUMAINFO_EVENTS_TARGET;
1024 break;
1025 default:
1026 break;
1027 }
1028 __this_cpu_write(memcg->stat->targets[target], next);
1029 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001030 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001031 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001032}
1033
1034/*
1035 * Check events in order.
1036 *
1037 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001038static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001039{
Steven Rostedt47994012011-11-02 13:38:33 -07001040 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001041 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001042 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1043 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001044 bool do_softlimit;
1045 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001046
1047 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1048 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001049#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001050 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1051 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001052#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001053 preempt_enable();
1054
1055 mem_cgroup_threshold(memcg);
1056 if (unlikely(do_softlimit))
1057 mem_cgroup_update_tree(memcg, page);
1058#if MAX_NUMNODES > 1
1059 if (unlikely(do_numainfo))
1060 atomic_inc(&memcg->numainfo_events);
1061#endif
1062 } else
1063 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001064}
1065
Glauber Costad1a4c0b2011-12-11 21:47:04 +00001066struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001067{
Wanpeng Lib2145142012-07-31 16:46:01 -07001068 return mem_cgroup_from_css(
1069 cgroup_subsys_state(cont, mem_cgroup_subsys_id));
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001070}
1071
Balbir Singhcf475ad2008-04-29 01:00:16 -07001072struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001073{
Balbir Singh31a78f22008-09-28 23:09:31 +01001074 /*
1075 * mm_update_next_owner() may clear mm->owner to NULL
1076 * if it races with swapoff, page migration, etc.
1077 * So this can be called with p == NULL.
1078 */
1079 if (unlikely(!p))
1080 return NULL;
1081
Wanpeng Lib2145142012-07-31 16:46:01 -07001082 return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001083}
1084
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001085struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001086{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001087 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001088
1089 if (!mm)
1090 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001091 /*
1092 * Because we have no locks, mm->owner's may be being moved to other
1093 * cgroup. We use css_tryget() here even if this looks
1094 * pessimistic (rather than adding locks here).
1095 */
1096 rcu_read_lock();
1097 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001098 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1099 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001100 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001101 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001102 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001103 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001104}
1105
Michal Hocko16248d82013-04-29 15:07:19 -07001106/*
1107 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1108 * ref. count) or NULL if the whole root's subtree has been visited.
1109 *
1110 * helper function to be used by mem_cgroup_iter
1111 */
1112static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
1113 struct mem_cgroup *last_visited)
1114{
1115 struct cgroup *prev_cgroup, *next_cgroup;
1116
1117 /*
1118 * Root is not visited by cgroup iterators so it needs an
1119 * explicit visit.
1120 */
1121 if (!last_visited)
1122 return root;
1123
1124 prev_cgroup = (last_visited == root) ? NULL
1125 : last_visited->css.cgroup;
1126skip_node:
1127 next_cgroup = cgroup_next_descendant_pre(
1128 prev_cgroup, root->css.cgroup);
1129
1130 /*
1131 * Even if we found a group we have to make sure it is
1132 * alive. css && !memcg means that the groups should be
1133 * skipped and we should continue the tree walk.
1134 * last_visited css is safe to use because it is
1135 * protected by css_get and the tree walk is rcu safe.
1136 */
1137 if (next_cgroup) {
1138 struct mem_cgroup *mem = mem_cgroup_from_cont(
1139 next_cgroup);
1140 if (css_tryget(&mem->css))
1141 return mem;
1142 else {
1143 prev_cgroup = next_cgroup;
1144 goto skip_node;
1145 }
1146 }
1147
1148 return NULL;
1149}
1150
Johannes Weiner56600482012-01-12 17:17:59 -08001151/**
1152 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1153 * @root: hierarchy root
1154 * @prev: previously returned memcg, NULL on first invocation
1155 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1156 *
1157 * Returns references to children of the hierarchy below @root, or
1158 * @root itself, or %NULL after a full round-trip.
1159 *
1160 * Caller must pass the return value in @prev on subsequent
1161 * invocations for reference counting, or use mem_cgroup_iter_break()
1162 * to cancel a hierarchy walk before the round-trip is complete.
1163 *
1164 * Reclaimers can specify a zone and a priority level in @reclaim to
1165 * divide up the memcgs in the hierarchy among all concurrent
1166 * reclaimers operating on the same zone and priority.
1167 */
1168struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
1169 struct mem_cgroup *prev,
1170 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001171{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001172 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001173 struct mem_cgroup *last_visited = NULL;
Michal Hocko5f578162013-04-29 15:07:17 -07001174 unsigned long uninitialized_var(dead_count);
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001175
Johannes Weiner56600482012-01-12 17:17:59 -08001176 if (mem_cgroup_disabled())
1177 return NULL;
1178
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001179 if (!root)
1180 root = root_mem_cgroup;
1181
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001182 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001183 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001184
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001185 if (!root->use_hierarchy && root != root_mem_cgroup) {
1186 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001187 goto out_css_put;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001188 return root;
1189 }
1190
Michal Hocko542f85f2013-04-29 15:07:15 -07001191 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001192 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001193 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001194
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001195 if (reclaim) {
1196 int nid = zone_to_nid(reclaim->zone);
1197 int zid = zone_idx(reclaim->zone);
1198 struct mem_cgroup_per_zone *mz;
1199
1200 mz = mem_cgroup_zoneinfo(root, nid, zid);
1201 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001202 last_visited = iter->last_visited;
1203 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001204 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001205 goto out_unlock;
1206 }
Michal Hocko5f578162013-04-29 15:07:17 -07001207
1208 /*
1209 * If the dead_count mismatches, a destruction
1210 * has happened or is happening concurrently.
1211 * If the dead_count matches, a destruction
1212 * might still happen concurrently, but since
1213 * we checked under RCU, that destruction
1214 * won't free the object until we release the
1215 * RCU reader lock. Thus, the dead_count
1216 * check verifies the pointer is still valid,
1217 * css_tryget() verifies the cgroup pointed to
1218 * is alive.
1219 */
1220 dead_count = atomic_read(&root->dead_count);
1221 smp_rmb();
1222 last_visited = iter->last_visited;
1223 if (last_visited) {
1224 if ((dead_count != iter->last_dead_count) ||
1225 !css_tryget(&last_visited->css)) {
1226 last_visited = NULL;
1227 }
1228 }
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001229 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001230
Michal Hocko16248d82013-04-29 15:07:19 -07001231 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001232
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001233 if (reclaim) {
Michal Hocko542f85f2013-04-29 15:07:15 -07001234 if (last_visited)
1235 css_put(&last_visited->css);
1236
Michal Hocko19f39402013-04-29 15:07:18 -07001237 iter->last_visited = memcg;
Michal Hocko5f578162013-04-29 15:07:17 -07001238 smp_wmb();
1239 iter->last_dead_count = dead_count;
Michal Hocko542f85f2013-04-29 15:07:15 -07001240
Michal Hocko19f39402013-04-29 15:07:18 -07001241 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001242 iter->generation++;
1243 else if (!prev && memcg)
1244 reclaim->generation = iter->generation;
1245 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001246
Michal Hocko19f39402013-04-29 15:07:18 -07001247 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001248 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001249 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001250out_unlock:
1251 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001252out_css_put:
1253 if (prev && prev != root)
1254 css_put(&prev->css);
1255
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001256 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001257}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001258
Johannes Weiner56600482012-01-12 17:17:59 -08001259/**
1260 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1261 * @root: hierarchy root
1262 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1263 */
1264void mem_cgroup_iter_break(struct mem_cgroup *root,
1265 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001266{
1267 if (!root)
1268 root = root_mem_cgroup;
1269 if (prev && prev != root)
1270 css_put(&prev->css);
1271}
1272
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001273/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001274 * Iteration constructs for visiting all cgroups (under a tree). If
1275 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1276 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001277 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001278#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001279 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001280 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001281 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001282
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001283#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001284 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001285 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001286 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001287
David Rientjes68ae5642012-12-12 13:51:57 -08001288void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001289{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001290 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001291
Ying Han456f9982011-05-26 16:25:38 -07001292 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001293 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1294 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001295 goto out;
1296
1297 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001298 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001299 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1300 break;
1301 case PGMAJFAULT:
1302 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001303 break;
1304 default:
1305 BUG();
1306 }
1307out:
1308 rcu_read_unlock();
1309}
David Rientjes68ae5642012-12-12 13:51:57 -08001310EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001311
Johannes Weiner925b7672012-01-12 17:18:15 -08001312/**
1313 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1314 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001315 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001316 *
1317 * Returns the lru list vector holding pages for the given @zone and
1318 * @mem. This can be the global zone lruvec, if the memory controller
1319 * is disabled.
1320 */
1321struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1322 struct mem_cgroup *memcg)
1323{
1324 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001325 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001326
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001327 if (mem_cgroup_disabled()) {
1328 lruvec = &zone->lruvec;
1329 goto out;
1330 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001331
1332 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001333 lruvec = &mz->lruvec;
1334out:
1335 /*
1336 * Since a node can be onlined after the mem_cgroup was created,
1337 * we have to be prepared to initialize lruvec->zone here;
1338 * and if offlined then reonlined, we need to reinitialize it.
1339 */
1340 if (unlikely(lruvec->zone != zone))
1341 lruvec->zone = zone;
1342 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001343}
1344
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001345/*
1346 * Following LRU functions are allowed to be used without PCG_LOCK.
1347 * Operations are called by routine of global LRU independently from memcg.
1348 * What we have to take care of here is validness of pc->mem_cgroup.
1349 *
1350 * Changes to pc->mem_cgroup happens when
1351 * 1. charge
1352 * 2. moving account
1353 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1354 * It is added to LRU before charge.
1355 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1356 * When moving account, the page is not on LRU. It's isolated.
1357 */
1358
Johannes Weiner925b7672012-01-12 17:18:15 -08001359/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001360 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001361 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001362 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001363 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001364struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001365{
1366 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001367 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001368 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001369 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001370
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001371 if (mem_cgroup_disabled()) {
1372 lruvec = &zone->lruvec;
1373 goto out;
1374 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001375
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001376 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001377 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001378
1379 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001380 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001381 * an uncharged page off lru does nothing to secure
1382 * its former mem_cgroup from sudden removal.
1383 *
1384 * Our caller holds lru_lock, and PageCgroupUsed is updated
1385 * under page_cgroup lock: between them, they make all uses
1386 * of pc->mem_cgroup safe.
1387 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001388 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001389 pc->mem_cgroup = memcg = root_mem_cgroup;
1390
Johannes Weiner925b7672012-01-12 17:18:15 -08001391 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001392 lruvec = &mz->lruvec;
1393out:
1394 /*
1395 * Since a node can be onlined after the mem_cgroup was created,
1396 * we have to be prepared to initialize lruvec->zone here;
1397 * and if offlined then reonlined, we need to reinitialize it.
1398 */
1399 if (unlikely(lruvec->zone != zone))
1400 lruvec->zone = zone;
1401 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001402}
1403
1404/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001405 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1406 * @lruvec: mem_cgroup per zone lru vector
1407 * @lru: index of lru list the page is sitting on
1408 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001409 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001410 * This function must be called when a page is added to or removed from an
1411 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001412 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001413void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1414 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001415{
1416 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001417 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001418
1419 if (mem_cgroup_disabled())
1420 return;
1421
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001422 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1423 lru_size = mz->lru_size + lru;
1424 *lru_size += nr_pages;
1425 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001426}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001427
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001428/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001429 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001430 * hierarchy subtree
1431 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001432bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1433 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001434{
Johannes Weiner91c637342012-05-29 15:06:24 -07001435 if (root_memcg == memcg)
1436 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001437 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001438 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001439 return css_is_ancestor(&memcg->css, &root_memcg->css);
1440}
1441
1442static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1443 struct mem_cgroup *memcg)
1444{
1445 bool ret;
1446
Johannes Weiner91c637342012-05-29 15:06:24 -07001447 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001448 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001449 rcu_read_unlock();
1450 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001451}
1452
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001453int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001454{
1455 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001456 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001457 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001458
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001459 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001460 if (p) {
1461 curr = try_get_mem_cgroup_from_mm(p->mm);
1462 task_unlock(p);
1463 } else {
1464 /*
1465 * All threads may have already detached their mm's, but the oom
1466 * killer still needs to detect if they have already been oom
1467 * killed to prevent needlessly killing additional tasks.
1468 */
1469 task_lock(task);
1470 curr = mem_cgroup_from_task(task);
1471 if (curr)
1472 css_get(&curr->css);
1473 task_unlock(task);
1474 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001475 if (!curr)
1476 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001477 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001478 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001479 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001480 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1481 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001482 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001483 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001484 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001485 return ret;
1486}
1487
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001488int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001489{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001490 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001491 unsigned long inactive;
1492 unsigned long active;
1493 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001494
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001495 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1496 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001497
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001498 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1499 if (gb)
1500 inactive_ratio = int_sqrt(10 * gb);
1501 else
1502 inactive_ratio = 1;
1503
Johannes Weiner9b272972011-11-02 13:38:23 -07001504 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001505}
1506
Balbir Singh6d61ef42009-01-07 18:08:06 -08001507#define mem_cgroup_from_res_counter(counter, member) \
1508 container_of(counter, struct mem_cgroup, member)
1509
Johannes Weiner19942822011-02-01 15:52:43 -08001510/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001511 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001512 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001513 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001514 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001515 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001516 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001517static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001518{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001519 unsigned long long margin;
1520
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001521 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001522 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001523 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001524 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001525}
1526
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001527int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001528{
1529 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001530
1531 /* root ? */
1532 if (cgrp->parent == NULL)
1533 return vm_swappiness;
1534
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001535 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001536}
1537
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001538/*
1539 * memcg->moving_account is used for checking possibility that some thread is
1540 * calling move_account(). When a thread on CPU-A starts moving pages under
1541 * a memcg, other threads should check memcg->moving_account under
1542 * rcu_read_lock(), like this:
1543 *
1544 * CPU-A CPU-B
1545 * rcu_read_lock()
1546 * memcg->moving_account+1 if (memcg->mocing_account)
1547 * take heavy locks.
1548 * synchronize_rcu() update something.
1549 * rcu_read_unlock()
1550 * start move here.
1551 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001552
1553/* for quick checking without looking up memcg */
1554atomic_t memcg_moving __read_mostly;
1555
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001556static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001557{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001558 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001559 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001560 synchronize_rcu();
1561}
1562
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001563static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001564{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001565 /*
1566 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1567 * We check NULL in callee rather than caller.
1568 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001569 if (memcg) {
1570 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001571 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001572 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001573}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001574
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001575/*
1576 * 2 routines for checking "mem" is under move_account() or not.
1577 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001578 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1579 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001580 * pc->mem_cgroup may be overwritten.
1581 *
1582 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1583 * under hierarchy of moving cgroups. This is for
1584 * waiting at hith-memory prressure caused by "move".
1585 */
1586
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001587static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001588{
1589 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001590 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001591}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001593static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001594{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001595 struct mem_cgroup *from;
1596 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001597 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001598 /*
1599 * Unlike task_move routines, we access mc.to, mc.from not under
1600 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1601 */
1602 spin_lock(&mc.lock);
1603 from = mc.from;
1604 to = mc.to;
1605 if (!from)
1606 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001607
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001608 ret = mem_cgroup_same_or_subtree(memcg, from)
1609 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001610unlock:
1611 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001612 return ret;
1613}
1614
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001615static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001616{
1617 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001618 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001619 DEFINE_WAIT(wait);
1620 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1621 /* moving charge context might have finished. */
1622 if (mc.moving_task)
1623 schedule();
1624 finish_wait(&mc.waitq, &wait);
1625 return true;
1626 }
1627 }
1628 return false;
1629}
1630
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001631/*
1632 * Take this lock when
1633 * - a code tries to modify page's memcg while it's USED.
1634 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001635 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001636 */
1637static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1638 unsigned long *flags)
1639{
1640 spin_lock_irqsave(&memcg->move_lock, *flags);
1641}
1642
1643static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1644 unsigned long *flags)
1645{
1646 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1647}
1648
Sha Zhengju58cf1882013-02-22 16:32:05 -08001649#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001650/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001651 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001652 * @memcg: The memory cgroup that went over limit
1653 * @p: Task that is going to be killed
1654 *
1655 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1656 * enabled
1657 */
1658void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1659{
1660 struct cgroup *task_cgrp;
1661 struct cgroup *mem_cgrp;
1662 /*
1663 * Need a buffer in BSS, can't rely on allocations. The code relies
1664 * on the assumption that OOM is serialized for memory controller.
1665 * If this assumption is broken, revisit this code.
1666 */
1667 static char memcg_name[PATH_MAX];
1668 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001669 struct mem_cgroup *iter;
1670 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001671
Sha Zhengju58cf1882013-02-22 16:32:05 -08001672 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001673 return;
1674
Balbir Singhe2224322009-04-02 16:57:39 -07001675 rcu_read_lock();
1676
1677 mem_cgrp = memcg->css.cgroup;
1678 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1679
1680 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1681 if (ret < 0) {
1682 /*
1683 * Unfortunately, we are unable to convert to a useful name
1684 * But we'll still print out the usage information
1685 */
1686 rcu_read_unlock();
1687 goto done;
1688 }
1689 rcu_read_unlock();
1690
Andrew Mortond0451972013-02-22 16:32:06 -08001691 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001692
1693 rcu_read_lock();
1694 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1695 if (ret < 0) {
1696 rcu_read_unlock();
1697 goto done;
1698 }
1699 rcu_read_unlock();
1700
1701 /*
1702 * Continues from above, so we don't need an KERN_ level
1703 */
Andrew Mortond0451972013-02-22 16:32:06 -08001704 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001705done:
1706
Andrew Mortond0451972013-02-22 16:32:06 -08001707 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001708 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1709 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1710 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001711 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001712 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1713 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1714 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001715 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001716 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1717 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1718 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001719
1720 for_each_mem_cgroup_tree(iter, memcg) {
1721 pr_info("Memory cgroup stats");
1722
1723 rcu_read_lock();
1724 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1725 if (!ret)
1726 pr_cont(" for %s", memcg_name);
1727 rcu_read_unlock();
1728 pr_cont(":");
1729
1730 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1731 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1732 continue;
1733 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1734 K(mem_cgroup_read_stat(iter, i)));
1735 }
1736
1737 for (i = 0; i < NR_LRU_LISTS; i++)
1738 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1739 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1740
1741 pr_cont("\n");
1742 }
Balbir Singhe2224322009-04-02 16:57:39 -07001743}
1744
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001745/*
1746 * This function returns the number of memcg under hierarchy tree. Returns
1747 * 1(self count) if no children.
1748 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001749static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001750{
1751 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001752 struct mem_cgroup *iter;
1753
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001754 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001755 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001756 return num;
1757}
1758
Balbir Singh6d61ef42009-01-07 18:08:06 -08001759/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001760 * Return the memory (and swap, if configured) limit for a memcg.
1761 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001762static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001763{
1764 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001765
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001766 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001767
David Rientjesa63d83f2010-08-09 17:19:46 -07001768 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001769 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001770 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001771 if (mem_cgroup_swappiness(memcg)) {
1772 u64 memsw;
1773
1774 limit += total_swap_pages << PAGE_SHIFT;
1775 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1776
1777 /*
1778 * If memsw is finite and limits the amount of swap space
1779 * available to this memcg, return that limit.
1780 */
1781 limit = min(limit, memsw);
1782 }
1783
1784 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001785}
1786
David Rientjes19965462012-12-11 16:00:26 -08001787static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1788 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001789{
1790 struct mem_cgroup *iter;
1791 unsigned long chosen_points = 0;
1792 unsigned long totalpages;
1793 unsigned int points = 0;
1794 struct task_struct *chosen = NULL;
1795
David Rientjes876aafb2012-07-31 16:43:48 -07001796 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001797 * If current has a pending SIGKILL or is exiting, then automatically
1798 * select it. The goal is to allow it to allocate so that it may
1799 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001800 */
David Rientjes465adcf2013-04-29 15:08:45 -07001801 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001802 set_thread_flag(TIF_MEMDIE);
1803 return;
1804 }
1805
1806 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001807 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1808 for_each_mem_cgroup_tree(iter, memcg) {
1809 struct cgroup *cgroup = iter->css.cgroup;
1810 struct cgroup_iter it;
1811 struct task_struct *task;
1812
1813 cgroup_iter_start(cgroup, &it);
1814 while ((task = cgroup_iter_next(cgroup, &it))) {
1815 switch (oom_scan_process_thread(task, totalpages, NULL,
1816 false)) {
1817 case OOM_SCAN_SELECT:
1818 if (chosen)
1819 put_task_struct(chosen);
1820 chosen = task;
1821 chosen_points = ULONG_MAX;
1822 get_task_struct(chosen);
1823 /* fall through */
1824 case OOM_SCAN_CONTINUE:
1825 continue;
1826 case OOM_SCAN_ABORT:
1827 cgroup_iter_end(cgroup, &it);
1828 mem_cgroup_iter_break(memcg, iter);
1829 if (chosen)
1830 put_task_struct(chosen);
1831 return;
1832 case OOM_SCAN_OK:
1833 break;
1834 };
1835 points = oom_badness(task, memcg, NULL, totalpages);
1836 if (points > chosen_points) {
1837 if (chosen)
1838 put_task_struct(chosen);
1839 chosen = task;
1840 chosen_points = points;
1841 get_task_struct(chosen);
1842 }
1843 }
1844 cgroup_iter_end(cgroup, &it);
1845 }
1846
1847 if (!chosen)
1848 return;
1849 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001850 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1851 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001852}
1853
Johannes Weiner56600482012-01-12 17:17:59 -08001854static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1855 gfp_t gfp_mask,
1856 unsigned long flags)
1857{
1858 unsigned long total = 0;
1859 bool noswap = false;
1860 int loop;
1861
1862 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1863 noswap = true;
1864 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1865 noswap = true;
1866
1867 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1868 if (loop)
1869 drain_all_stock_async(memcg);
1870 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1871 /*
1872 * Allow limit shrinkers, which are triggered directly
1873 * by userspace, to catch signals and stop reclaim
1874 * after minimal progress, regardless of the margin.
1875 */
1876 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1877 break;
1878 if (mem_cgroup_margin(memcg))
1879 break;
1880 /*
1881 * If nothing was reclaimed after two attempts, there
1882 * may be no reclaimable pages in this hierarchy.
1883 */
1884 if (loop && !total)
1885 break;
1886 }
1887 return total;
1888}
1889
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001890/**
1891 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001892 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001893 * @nid: the node ID to be checked.
1894 * @noswap : specify true here if the user wants flle only information.
1895 *
1896 * This function returns whether the specified memcg contains any
1897 * reclaimable pages on a node. Returns true if there are any reclaimable
1898 * pages in the node.
1899 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001900static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001901 int nid, bool noswap)
1902{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001903 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001904 return true;
1905 if (noswap || !total_swap_pages)
1906 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001907 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001908 return true;
1909 return false;
1910
1911}
Ying Han889976d2011-05-26 16:25:33 -07001912#if MAX_NUMNODES > 1
1913
1914/*
1915 * Always updating the nodemask is not very good - even if we have an empty
1916 * list or the wrong list here, we can start from some node and traverse all
1917 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1918 *
1919 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001920static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001921{
1922 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001923 /*
1924 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1925 * pagein/pageout changes since the last update.
1926 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001927 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001928 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001929 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001930 return;
1931
Ying Han889976d2011-05-26 16:25:33 -07001932 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001933 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001934
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001935 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001936
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001937 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1938 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001939 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001940
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001941 atomic_set(&memcg->numainfo_events, 0);
1942 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001943}
1944
1945/*
1946 * Selecting a node where we start reclaim from. Because what we need is just
1947 * reducing usage counter, start from anywhere is O,K. Considering
1948 * memory reclaim from current node, there are pros. and cons.
1949 *
1950 * Freeing memory from current node means freeing memory from a node which
1951 * we'll use or we've used. So, it may make LRU bad. And if several threads
1952 * hit limits, it will see a contention on a node. But freeing from remote
1953 * node means more costs for memory reclaim because of memory latency.
1954 *
1955 * Now, we use round-robin. Better algorithm is welcomed.
1956 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001957int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001958{
1959 int node;
1960
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001961 mem_cgroup_may_update_nodemask(memcg);
1962 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001963
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001964 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001965 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001966 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001967 /*
1968 * We call this when we hit limit, not when pages are added to LRU.
1969 * No LRU may hold pages because all pages are UNEVICTABLE or
1970 * memcg is too small and all pages are not on LRU. In that case,
1971 * we use curret node.
1972 */
1973 if (unlikely(node == MAX_NUMNODES))
1974 node = numa_node_id();
1975
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001976 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001977 return node;
1978}
1979
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001980/*
1981 * Check all nodes whether it contains reclaimable pages or not.
1982 * For quick scan, we make use of scan_nodes. This will allow us to skip
1983 * unused nodes. But scan_nodes is lazily updated and may not cotain
1984 * enough new information. We need to do double check.
1985 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001986static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001987{
1988 int nid;
1989
1990 /*
1991 * quick check...making use of scan_node.
1992 * We can skip unused nodes.
1993 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001994 if (!nodes_empty(memcg->scan_nodes)) {
1995 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001996 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001997 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001998
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001999 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002000 return true;
2001 }
2002 }
2003 /*
2004 * Check rest of nodes.
2005 */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08002006 for_each_node_state(nid, N_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002007 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002008 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002009 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002010 return true;
2011 }
2012 return false;
2013}
2014
Ying Han889976d2011-05-26 16:25:33 -07002015#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002016int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002017{
2018 return 0;
2019}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002020
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002021static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002022{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002023 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002024}
Ying Han889976d2011-05-26 16:25:33 -07002025#endif
2026
Johannes Weiner56600482012-01-12 17:17:59 -08002027static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2028 struct zone *zone,
2029 gfp_t gfp_mask,
2030 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002031{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002032 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08002033 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002034 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002035 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07002036 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08002037 struct mem_cgroup_reclaim_cookie reclaim = {
2038 .zone = zone,
2039 .priority = 0,
2040 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002041
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002042 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002043
Balbir Singh4e416952009-09-23 15:56:39 -07002044 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08002045 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002046 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002047 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07002048 if (loop >= 2) {
2049 /*
2050 * If we have not been able to reclaim
2051 * anything, it might because there are
2052 * no reclaimable pages under this hierarchy
2053 */
Johannes Weiner56600482012-01-12 17:17:59 -08002054 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07002055 break;
Balbir Singh4e416952009-09-23 15:56:39 -07002056 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002057 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07002058 * excess >> 2 is not to excessive so as to
2059 * reclaim too much, nor too less that we keep
2060 * coming back to reclaim from this cgroup
2061 */
2062 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002063 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07002064 break;
Balbir Singh4e416952009-09-23 15:56:39 -07002065 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002066 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07002067 }
Johannes Weiner56600482012-01-12 17:17:59 -08002068 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08002069 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08002070 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2071 zone, &nr_scanned);
2072 *total_scanned += nr_scanned;
2073 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002074 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002075 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002076 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002077 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002078}
2079
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002080/*
2081 * Check OOM-Killer is already running under our hierarchy.
2082 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07002083 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002084 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002085static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002086{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002087 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002088
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002089 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002090 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002091 /*
2092 * this subtree of our hierarchy is already locked
2093 * so we cannot give a lock.
2094 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002095 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002096 mem_cgroup_iter_break(memcg, iter);
2097 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002098 } else
2099 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002100 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002101
Michal Hocko79dfdac2011-07-26 16:08:23 -07002102 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07002103 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002104
2105 /*
2106 * OK, we failed to lock the whole subtree so we have to clean up
2107 * what we set up to the failing subtree
2108 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002109 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002110 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002111 mem_cgroup_iter_break(memcg, iter);
2112 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002113 }
2114 iter->oom_lock = false;
2115 }
Johannes Weiner23751be2011-08-25 15:59:16 -07002116 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002117}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002118
Michal Hocko79dfdac2011-07-26 16:08:23 -07002119/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07002120 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07002121 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002122static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002123{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002124 struct mem_cgroup *iter;
2125
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002126 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002127 iter->oom_lock = false;
2128 return 0;
2129}
2130
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002131static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002132{
2133 struct mem_cgroup *iter;
2134
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002136 atomic_inc(&iter->under_oom);
2137}
2138
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002139static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002140{
2141 struct mem_cgroup *iter;
2142
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002143 /*
2144 * When a new child is created while the hierarchy is under oom,
2145 * mem_cgroup_oom_lock() may not be called. We have to use
2146 * atomic_add_unless() here.
2147 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002148 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002149 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002150}
2151
Michal Hocko1af8efe2011-07-26 16:08:24 -07002152static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002153static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2154
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002155struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002156 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002157 wait_queue_t wait;
2158};
2159
2160static int memcg_oom_wake_function(wait_queue_t *wait,
2161 unsigned mode, int sync, void *arg)
2162{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002163 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2164 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002165 struct oom_wait_info *oom_wait_info;
2166
2167 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002168 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002169
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002170 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002171 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002172 * Then we can use css_is_ancestor without taking care of RCU.
2173 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002174 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2175 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002176 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002177 return autoremove_wake_function(wait, mode, sync, arg);
2178}
2179
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002180static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002181{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002182 /* for filtering, pass "memcg" as argument. */
2183 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002184}
2185
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002186static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002187{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002188 if (memcg && atomic_read(&memcg->under_oom))
2189 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002190}
2191
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002192/*
2193 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
2194 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002195static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
2196 int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002197{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002198 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002199 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002200
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002201 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002202 owait.wait.flags = 0;
2203 owait.wait.func = memcg_oom_wake_function;
2204 owait.wait.private = current;
2205 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002206 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002207 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002208
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002209 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07002210 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002211 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002212 /*
2213 * Even if signal_pending(), we can't quit charge() loop without
2214 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
2215 * under OOM is always welcomed, use TASK_KILLABLE here.
2216 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002217 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002218 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002219 need_to_kill = false;
2220 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002221 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002222 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002223
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002224 if (need_to_kill) {
2225 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07002226 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002227 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002228 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002229 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002230 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07002231 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002232 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002233 mem_cgroup_oom_unlock(memcg);
2234 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002235 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002236
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002237 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002238
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002239 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
2240 return false;
2241 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07002242 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002243 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002244}
2245
Balbir Singhd69b0422009-06-17 16:26:34 -07002246/*
2247 * Currently used to update mapped file statistics, but the routine can be
2248 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002249 *
2250 * Notes: Race condition
2251 *
2252 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2253 * it tends to be costly. But considering some conditions, we doesn't need
2254 * to do so _always_.
2255 *
2256 * Considering "charge", lock_page_cgroup() is not required because all
2257 * file-stat operations happen after a page is attached to radix-tree. There
2258 * are no race with "charge".
2259 *
2260 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2261 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2262 * if there are race with "uncharge". Statistics itself is properly handled
2263 * by flags.
2264 *
2265 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002266 * small, we check mm->moving_account and detect there are possibility of race
2267 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002268 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002269
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002270void __mem_cgroup_begin_update_page_stat(struct page *page,
2271 bool *locked, unsigned long *flags)
2272{
2273 struct mem_cgroup *memcg;
2274 struct page_cgroup *pc;
2275
2276 pc = lookup_page_cgroup(page);
2277again:
2278 memcg = pc->mem_cgroup;
2279 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2280 return;
2281 /*
2282 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002283 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002284 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002285 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002286 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002287 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002288 return;
2289
2290 move_lock_mem_cgroup(memcg, flags);
2291 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2292 move_unlock_mem_cgroup(memcg, flags);
2293 goto again;
2294 }
2295 *locked = true;
2296}
2297
2298void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2299{
2300 struct page_cgroup *pc = lookup_page_cgroup(page);
2301
2302 /*
2303 * It's guaranteed that pc->mem_cgroup never changes while
2304 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002305 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002306 */
2307 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2308}
2309
Greg Thelen2a7106f2011-01-13 15:47:37 -08002310void mem_cgroup_update_page_stat(struct page *page,
2311 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002312{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002313 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002314 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002315 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002316
Johannes Weinercfa44942012-01-12 17:18:38 -08002317 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002318 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002319
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002320 memcg = pc->mem_cgroup;
2321 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002322 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002323
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002324 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08002325 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08002326 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002327 break;
2328 default:
2329 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002330 }
Balbir Singhd69b0422009-06-17 16:26:34 -07002331
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002332 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002333}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002334
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002335/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002336 * size of first charge trial. "32" comes from vmscan.c's magic value.
2337 * TODO: maybe necessary to use big numbers in big irons.
2338 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002339#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002340struct memcg_stock_pcp {
2341 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002342 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002343 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002344 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002345#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002346};
2347static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002348static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002349
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002350/**
2351 * consume_stock: Try to consume stocked charge on this cpu.
2352 * @memcg: memcg to consume from.
2353 * @nr_pages: how many pages to charge.
2354 *
2355 * The charges will only happen if @memcg matches the current cpu's memcg
2356 * stock, and at least @nr_pages are available in that stock. Failure to
2357 * service an allocation will refill the stock.
2358 *
2359 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002360 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002361static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002362{
2363 struct memcg_stock_pcp *stock;
2364 bool ret = true;
2365
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002366 if (nr_pages > CHARGE_BATCH)
2367 return false;
2368
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002370 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2371 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002372 else /* need to call res_counter_charge */
2373 ret = false;
2374 put_cpu_var(memcg_stock);
2375 return ret;
2376}
2377
2378/*
2379 * Returns stocks cached in percpu to res_counter and reset cached information.
2380 */
2381static void drain_stock(struct memcg_stock_pcp *stock)
2382{
2383 struct mem_cgroup *old = stock->cached;
2384
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002385 if (stock->nr_pages) {
2386 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2387
2388 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002389 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002390 res_counter_uncharge(&old->memsw, bytes);
2391 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002392 }
2393 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002394}
2395
2396/*
2397 * This must be called under preempt disabled or must be called by
2398 * a thread which is pinned to local cpu.
2399 */
2400static void drain_local_stock(struct work_struct *dummy)
2401{
2402 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2403 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002404 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002405}
2406
Michal Hockoe4777492013-02-22 16:35:40 -08002407static void __init memcg_stock_init(void)
2408{
2409 int cpu;
2410
2411 for_each_possible_cpu(cpu) {
2412 struct memcg_stock_pcp *stock =
2413 &per_cpu(memcg_stock, cpu);
2414 INIT_WORK(&stock->work, drain_local_stock);
2415 }
2416}
2417
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002418/*
2419 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002420 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002421 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002422static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002423{
2424 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2425
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002426 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002427 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002428 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002429 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002430 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002431 put_cpu_var(memcg_stock);
2432}
2433
2434/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002435 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002436 * of the hierarchy under it. sync flag says whether we should block
2437 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002438 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002439static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002440{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002441 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002442
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002443 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002444 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002445 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002446 for_each_online_cpu(cpu) {
2447 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002448 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002449
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002450 memcg = stock->cached;
2451 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002452 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002453 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002454 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002455 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2456 if (cpu == curcpu)
2457 drain_local_stock(&stock->work);
2458 else
2459 schedule_work_on(cpu, &stock->work);
2460 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002461 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002462 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002463
2464 if (!sync)
2465 goto out;
2466
2467 for_each_online_cpu(cpu) {
2468 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002469 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002470 flush_work(&stock->work);
2471 }
2472out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002473 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002474}
2475
2476/*
2477 * Tries to drain stocked charges in other cpus. This function is asynchronous
2478 * and just put a work per cpu for draining localy on each cpu. Caller can
2479 * expects some charges will be back to res_counter later but cannot wait for
2480 * it.
2481 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002482static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002483{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002484 /*
2485 * If someone calls draining, avoid adding more kworker runs.
2486 */
2487 if (!mutex_trylock(&percpu_charge_mutex))
2488 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002489 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002490 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002491}
2492
2493/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002494static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002495{
2496 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002497 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002498 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002499 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002500}
2501
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002502/*
2503 * This function drains percpu counter value from DEAD cpu and
2504 * move it to local cpu. Note that this function can be preempted.
2505 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002506static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002507{
2508 int i;
2509
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002510 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002511 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002512 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002513
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002514 per_cpu(memcg->stat->count[i], cpu) = 0;
2515 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002516 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002517 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002518 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002519
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002520 per_cpu(memcg->stat->events[i], cpu) = 0;
2521 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002522 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002523 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002524}
2525
2526static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002527 unsigned long action,
2528 void *hcpu)
2529{
2530 int cpu = (unsigned long)hcpu;
2531 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002532 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002533
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002534 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002535 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002536
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002537 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002538 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002539
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002540 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002541 mem_cgroup_drain_pcp_counter(iter, cpu);
2542
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002543 stock = &per_cpu(memcg_stock, cpu);
2544 drain_stock(stock);
2545 return NOTIFY_OK;
2546}
2547
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002548
2549/* See __mem_cgroup_try_charge() for details */
2550enum {
2551 CHARGE_OK, /* success */
2552 CHARGE_RETRY, /* need to retry but retry is not bad */
2553 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2554 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2555 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2556};
2557
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002558static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002559 unsigned int nr_pages, unsigned int min_pages,
2560 bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002561{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002562 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002563 struct mem_cgroup *mem_over_limit;
2564 struct res_counter *fail_res;
2565 unsigned long flags = 0;
2566 int ret;
2567
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002568 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002569
2570 if (likely(!ret)) {
2571 if (!do_swap_account)
2572 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002573 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002574 if (likely(!ret))
2575 return CHARGE_OK;
2576
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002577 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002578 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2579 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2580 } else
2581 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002582 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002583 * Never reclaim on behalf of optional batching, retry with a
2584 * single page instead.
2585 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002586 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002587 return CHARGE_RETRY;
2588
2589 if (!(gfp_mask & __GFP_WAIT))
2590 return CHARGE_WOULDBLOCK;
2591
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002592 if (gfp_mask & __GFP_NORETRY)
2593 return CHARGE_NOMEM;
2594
Johannes Weiner56600482012-01-12 17:17:59 -08002595 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002596 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002597 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002598 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002599 * Even though the limit is exceeded at this point, reclaim
2600 * may have been able to free some pages. Retry the charge
2601 * before killing the task.
2602 *
2603 * Only for regular pages, though: huge pages are rather
2604 * unlikely to succeed so close to the limit, and we fall back
2605 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002606 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002607 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002608 return CHARGE_RETRY;
2609
2610 /*
2611 * At task move, charge accounts can be doubly counted. So, it's
2612 * better to wait until the end of task_move if something is going on.
2613 */
2614 if (mem_cgroup_wait_acct_move(mem_over_limit))
2615 return CHARGE_RETRY;
2616
2617 /* If we don't need to call oom-killer at el, return immediately */
2618 if (!oom_check)
2619 return CHARGE_NOMEM;
2620 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002621 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002622 return CHARGE_OOM_DIE;
2623
2624 return CHARGE_RETRY;
2625}
2626
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002627/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002628 * __mem_cgroup_try_charge() does
2629 * 1. detect memcg to be charged against from passed *mm and *ptr,
2630 * 2. update res_counter
2631 * 3. call memory reclaim if necessary.
2632 *
2633 * In some special case, if the task is fatal, fatal_signal_pending() or
2634 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2635 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2636 * as possible without any hazards. 2: all pages should have a valid
2637 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2638 * pointer, that is treated as a charge to root_mem_cgroup.
2639 *
2640 * So __mem_cgroup_try_charge() will return
2641 * 0 ... on success, filling *ptr with a valid memcg pointer.
2642 * -ENOMEM ... charge failure because of resource limits.
2643 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2644 *
2645 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2646 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002647 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002648static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002649 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002650 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002651 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002652 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002653{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002654 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002655 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002656 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002657 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002658
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002659 /*
2660 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2661 * in system level. So, allow to go ahead dying process in addition to
2662 * MEMDIE process.
2663 */
2664 if (unlikely(test_thread_flag(TIF_MEMDIE)
2665 || fatal_signal_pending(current)))
2666 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002667
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002668 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002669 * We always charge the cgroup the mm_struct belongs to.
2670 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002671 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002672 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002673 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002674 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002675 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002676again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002677 if (*ptr) { /* css should be a valid one */
2678 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002679 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002680 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002681 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002682 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002683 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002684 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002685 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002686
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002687 rcu_read_lock();
2688 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002689 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002690 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002691 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002692 * race with swapoff. Then, we have small risk of mis-accouning.
2693 * But such kind of mis-account by race always happens because
2694 * we don't have cgroup_mutex(). It's overkill and we allo that
2695 * small race, here.
2696 * (*) swapoff at el will charge against mm-struct not against
2697 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002698 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002699 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002700 if (!memcg)
2701 memcg = root_mem_cgroup;
2702 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002703 rcu_read_unlock();
2704 goto done;
2705 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002706 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002707 /*
2708 * It seems dagerous to access memcg without css_get().
2709 * But considering how consume_stok works, it's not
2710 * necessary. If consume_stock success, some charges
2711 * from this memcg are cached on this cpu. So, we
2712 * don't need to call css_get()/css_tryget() before
2713 * calling consume_stock().
2714 */
2715 rcu_read_unlock();
2716 goto done;
2717 }
2718 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002719 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002720 rcu_read_unlock();
2721 goto again;
2722 }
2723 rcu_read_unlock();
2724 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002725
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002726 do {
2727 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002728
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002729 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002730 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002731 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002732 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002733 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002734
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002735 oom_check = false;
2736 if (oom && !nr_oom_retries) {
2737 oom_check = true;
2738 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2739 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002740
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002741 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages,
2742 oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002743 switch (ret) {
2744 case CHARGE_OK:
2745 break;
2746 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002747 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002748 css_put(&memcg->css);
2749 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002750 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002751 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002752 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002753 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002754 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002755 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002756 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002757 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002758 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002759 /* If oom, we never return -ENOMEM */
2760 nr_oom_retries--;
2761 break;
2762 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002763 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002764 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002765 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002766 } while (ret != CHARGE_OK);
2767
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002768 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002769 refill_stock(memcg, batch - nr_pages);
2770 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002771done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002772 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002773 return 0;
2774nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002775 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002776 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002777bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002778 *ptr = root_mem_cgroup;
2779 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002780}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002781
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002782/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002783 * Somemtimes we have to undo a charge we got by try_charge().
2784 * This function is for that and do uncharge, put css's refcnt.
2785 * gotten by try_charge().
2786 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002787static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002788 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002789{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002790 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002791 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002792
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002793 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002794 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002795 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002796 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002797}
2798
2799/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002800 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2801 * This is useful when moving usage to parent cgroup.
2802 */
2803static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2804 unsigned int nr_pages)
2805{
2806 unsigned long bytes = nr_pages * PAGE_SIZE;
2807
2808 if (mem_cgroup_is_root(memcg))
2809 return;
2810
2811 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2812 if (do_swap_account)
2813 res_counter_uncharge_until(&memcg->memsw,
2814 memcg->memsw.parent, bytes);
2815}
2816
2817/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002818 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002819 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2820 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2821 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002822 */
2823static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2824{
2825 struct cgroup_subsys_state *css;
2826
2827 /* ID 0 is unused ID */
2828 if (!id)
2829 return NULL;
2830 css = css_lookup(&mem_cgroup_subsys, id);
2831 if (!css)
2832 return NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -07002833 return mem_cgroup_from_css(css);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002834}
2835
Wu Fengguange42d9d52009-12-16 12:19:59 +01002836struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002837{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002838 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002839 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002840 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002841 swp_entry_t ent;
2842
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002843 VM_BUG_ON(!PageLocked(page));
2844
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002845 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002846 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002847 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002848 memcg = pc->mem_cgroup;
2849 if (memcg && !css_tryget(&memcg->css))
2850 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002851 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002852 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002853 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002854 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002855 memcg = mem_cgroup_lookup(id);
2856 if (memcg && !css_tryget(&memcg->css))
2857 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002858 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002859 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002860 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002861 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002862}
2863
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002864static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002865 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002866 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002867 enum charge_type ctype,
2868 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002869{
Johannes Weinerce587e62012-04-24 20:22:33 +02002870 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002871 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002872 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002873 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002874 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002875
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002876 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002877 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002878 /*
2879 * we don't need page_cgroup_lock about tail pages, becase they are not
2880 * accessed by any other context at this point.
2881 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002882
2883 /*
2884 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2885 * may already be on some other mem_cgroup's LRU. Take care of it.
2886 */
2887 if (lrucare) {
2888 zone = page_zone(page);
2889 spin_lock_irq(&zone->lru_lock);
2890 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002891 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002892 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002893 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002894 was_on_lru = true;
2895 }
2896 }
2897
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002898 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002899 /*
2900 * We access a page_cgroup asynchronously without lock_page_cgroup().
2901 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2902 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2903 * before USED bit, we need memory barrier here.
2904 * See mem_cgroup_add_lru_list(), etc.
2905 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002906 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002907 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002908
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002909 if (lrucare) {
2910 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002911 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002912 VM_BUG_ON(PageLRU(page));
2913 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002914 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002915 }
2916 spin_unlock_irq(&zone->lru_lock);
2917 }
2918
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002919 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002920 anon = true;
2921 else
2922 anon = false;
2923
David Rientjesb070e652013-05-07 16:18:09 -07002924 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002925 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002926
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002927 /*
2928 * "charge_statistics" updated event counter. Then, check it.
2929 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2930 * if they exceeds softlimit.
2931 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002932 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002933}
2934
Glauber Costa7cf27982012-12-18 14:22:55 -08002935static DEFINE_MUTEX(set_limit_mutex);
2936
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002937#ifdef CONFIG_MEMCG_KMEM
2938static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2939{
2940 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
2941 (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK);
2942}
2943
Glauber Costa1f458cb2012-12-18 14:22:50 -08002944/*
2945 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2946 * in the memcg_cache_params struct.
2947 */
2948static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2949{
2950 struct kmem_cache *cachep;
2951
2952 VM_BUG_ON(p->is_root_cache);
2953 cachep = p->root_cache;
2954 return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)];
2955}
2956
Glauber Costa749c5412012-12-18 14:23:01 -08002957#ifdef CONFIG_SLABINFO
2958static int mem_cgroup_slabinfo_read(struct cgroup *cont, struct cftype *cft,
2959 struct seq_file *m)
2960{
2961 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2962 struct memcg_cache_params *params;
2963
2964 if (!memcg_can_account_kmem(memcg))
2965 return -EIO;
2966
2967 print_slabinfo_header(m);
2968
2969 mutex_lock(&memcg->slab_caches_mutex);
2970 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2971 cache_show(memcg_params_to_cache(params), m);
2972 mutex_unlock(&memcg->slab_caches_mutex);
2973
2974 return 0;
2975}
2976#endif
2977
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002978static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2979{
2980 struct res_counter *fail_res;
2981 struct mem_cgroup *_memcg;
2982 int ret = 0;
2983 bool may_oom;
2984
2985 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2986 if (ret)
2987 return ret;
2988
2989 /*
2990 * Conditions under which we can wait for the oom_killer. Those are
2991 * the same conditions tested by the core page allocator
2992 */
2993 may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY);
2994
2995 _memcg = memcg;
2996 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
2997 &_memcg, may_oom);
2998
2999 if (ret == -EINTR) {
3000 /*
3001 * __mem_cgroup_try_charge() chosed to bypass to root due to
3002 * OOM kill or fatal signal. Since our only options are to
3003 * either fail the allocation or charge it to this cgroup, do
3004 * it as a temporary condition. But we can't fail. From a
3005 * kmem/slab perspective, the cache has already been selected,
3006 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3007 * our minds.
3008 *
3009 * This condition will only trigger if the task entered
3010 * memcg_charge_kmem in a sane state, but was OOM-killed during
3011 * __mem_cgroup_try_charge() above. Tasks that were already
3012 * dying when the allocation triggers should have been already
3013 * directed to the root cgroup in memcontrol.h
3014 */
3015 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3016 if (do_swap_account)
3017 res_counter_charge_nofail(&memcg->memsw, size,
3018 &fail_res);
3019 ret = 0;
3020 } else if (ret)
3021 res_counter_uncharge(&memcg->kmem, size);
3022
3023 return ret;
3024}
3025
3026static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3027{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003028 res_counter_uncharge(&memcg->res, size);
3029 if (do_swap_account)
3030 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003031
3032 /* Not down to 0 */
3033 if (res_counter_uncharge(&memcg->kmem, size))
3034 return;
3035
3036 if (memcg_kmem_test_and_clear_dead(memcg))
3037 mem_cgroup_put(memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003038}
3039
Glauber Costa2633d7a2012-12-18 14:22:34 -08003040void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
3041{
3042 if (!memcg)
3043 return;
3044
3045 mutex_lock(&memcg->slab_caches_mutex);
3046 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
3047 mutex_unlock(&memcg->slab_caches_mutex);
3048}
3049
3050/*
3051 * helper for acessing a memcg's index. It will be used as an index in the
3052 * child cache array in kmem_cache, and also to derive its name. This function
3053 * will return -1 when this is not a kmem-limited memcg.
3054 */
3055int memcg_cache_id(struct mem_cgroup *memcg)
3056{
3057 return memcg ? memcg->kmemcg_id : -1;
3058}
3059
Glauber Costa55007d82012-12-18 14:22:38 -08003060/*
3061 * This ends up being protected by the set_limit mutex, during normal
3062 * operation, because that is its main call site.
3063 *
3064 * But when we create a new cache, we can call this as well if its parent
3065 * is kmem-limited. That will have to hold set_limit_mutex as well.
3066 */
3067int memcg_update_cache_sizes(struct mem_cgroup *memcg)
3068{
3069 int num, ret;
3070
3071 num = ida_simple_get(&kmem_limited_groups,
3072 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
3073 if (num < 0)
3074 return num;
3075 /*
3076 * After this point, kmem_accounted (that we test atomically in
3077 * the beginning of this conditional), is no longer 0. This
3078 * guarantees only one process will set the following boolean
3079 * to true. We don't need test_and_set because we're protected
3080 * by the set_limit_mutex anyway.
3081 */
3082 memcg_kmem_set_activated(memcg);
3083
3084 ret = memcg_update_all_caches(num+1);
3085 if (ret) {
3086 ida_simple_remove(&kmem_limited_groups, num);
3087 memcg_kmem_clear_activated(memcg);
3088 return ret;
3089 }
3090
3091 memcg->kmemcg_id = num;
3092 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
3093 mutex_init(&memcg->slab_caches_mutex);
3094 return 0;
3095}
3096
3097static size_t memcg_caches_array_size(int num_groups)
3098{
3099 ssize_t size;
3100 if (num_groups <= 0)
3101 return 0;
3102
3103 size = 2 * num_groups;
3104 if (size < MEMCG_CACHES_MIN_SIZE)
3105 size = MEMCG_CACHES_MIN_SIZE;
3106 else if (size > MEMCG_CACHES_MAX_SIZE)
3107 size = MEMCG_CACHES_MAX_SIZE;
3108
3109 return size;
3110}
3111
3112/*
3113 * We should update the current array size iff all caches updates succeed. This
3114 * can only be done from the slab side. The slab mutex needs to be held when
3115 * calling this.
3116 */
3117void memcg_update_array_size(int num)
3118{
3119 if (num > memcg_limited_groups_array_size)
3120 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3121}
3122
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003123static void kmem_cache_destroy_work_func(struct work_struct *w);
3124
Glauber Costa55007d82012-12-18 14:22:38 -08003125int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3126{
3127 struct memcg_cache_params *cur_params = s->memcg_params;
3128
3129 VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache);
3130
3131 if (num_groups > memcg_limited_groups_array_size) {
3132 int i;
3133 ssize_t size = memcg_caches_array_size(num_groups);
3134
3135 size *= sizeof(void *);
3136 size += sizeof(struct memcg_cache_params);
3137
3138 s->memcg_params = kzalloc(size, GFP_KERNEL);
3139 if (!s->memcg_params) {
3140 s->memcg_params = cur_params;
3141 return -ENOMEM;
3142 }
3143
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003144 INIT_WORK(&s->memcg_params->destroy,
3145 kmem_cache_destroy_work_func);
Glauber Costa55007d82012-12-18 14:22:38 -08003146 s->memcg_params->is_root_cache = true;
3147
3148 /*
3149 * There is the chance it will be bigger than
3150 * memcg_limited_groups_array_size, if we failed an allocation
3151 * in a cache, in which case all caches updated before it, will
3152 * have a bigger array.
3153 *
3154 * But if that is the case, the data after
3155 * memcg_limited_groups_array_size is certainly unused
3156 */
3157 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3158 if (!cur_params->memcg_caches[i])
3159 continue;
3160 s->memcg_params->memcg_caches[i] =
3161 cur_params->memcg_caches[i];
3162 }
3163
3164 /*
3165 * Ideally, we would wait until all caches succeed, and only
3166 * then free the old one. But this is not worth the extra
3167 * pointer per-cache we'd have to have for this.
3168 *
3169 * It is not a big deal if some caches are left with a size
3170 * bigger than the others. And all updates will reset this
3171 * anyway.
3172 */
3173 kfree(cur_params);
3174 }
3175 return 0;
3176}
3177
Glauber Costa943a4512012-12-18 14:23:03 -08003178int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
3179 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003180{
3181 size_t size = sizeof(struct memcg_cache_params);
3182
3183 if (!memcg_kmem_enabled())
3184 return 0;
3185
Glauber Costa55007d82012-12-18 14:22:38 -08003186 if (!memcg)
3187 size += memcg_limited_groups_array_size * sizeof(void *);
3188
Glauber Costa2633d7a2012-12-18 14:22:34 -08003189 s->memcg_params = kzalloc(size, GFP_KERNEL);
3190 if (!s->memcg_params)
3191 return -ENOMEM;
3192
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003193 INIT_WORK(&s->memcg_params->destroy,
3194 kmem_cache_destroy_work_func);
Glauber Costa943a4512012-12-18 14:23:03 -08003195 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003196 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003197 s->memcg_params->root_cache = root_cache;
Glauber Costa4ba902b2013-02-12 13:46:22 -08003198 } else
3199 s->memcg_params->is_root_cache = true;
3200
Glauber Costa2633d7a2012-12-18 14:22:34 -08003201 return 0;
3202}
3203
3204void memcg_release_cache(struct kmem_cache *s)
3205{
Glauber Costad7f25f82012-12-18 14:22:40 -08003206 struct kmem_cache *root;
3207 struct mem_cgroup *memcg;
3208 int id;
3209
3210 /*
3211 * This happens, for instance, when a root cache goes away before we
3212 * add any memcg.
3213 */
3214 if (!s->memcg_params)
3215 return;
3216
3217 if (s->memcg_params->is_root_cache)
3218 goto out;
3219
3220 memcg = s->memcg_params->memcg;
3221 id = memcg_cache_id(memcg);
3222
3223 root = s->memcg_params->root_cache;
3224 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08003225
3226 mutex_lock(&memcg->slab_caches_mutex);
3227 list_del(&s->memcg_params->list);
3228 mutex_unlock(&memcg->slab_caches_mutex);
3229
Li Zefanfd0ccaf2013-04-29 15:08:43 -07003230 mem_cgroup_put(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08003231out:
Glauber Costa2633d7a2012-12-18 14:22:34 -08003232 kfree(s->memcg_params);
3233}
3234
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003235/*
3236 * During the creation a new cache, we need to disable our accounting mechanism
3237 * altogether. This is true even if we are not creating, but rather just
3238 * enqueing new caches to be created.
3239 *
3240 * This is because that process will trigger allocations; some visible, like
3241 * explicit kmallocs to auxiliary data structures, name strings and internal
3242 * cache structures; some well concealed, like INIT_WORK() that can allocate
3243 * objects during debug.
3244 *
3245 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3246 * to it. This may not be a bounded recursion: since the first cache creation
3247 * failed to complete (waiting on the allocation), we'll just try to create the
3248 * cache again, failing at the same point.
3249 *
3250 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3251 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3252 * inside the following two functions.
3253 */
3254static inline void memcg_stop_kmem_account(void)
3255{
3256 VM_BUG_ON(!current->mm);
3257 current->memcg_kmem_skip_account++;
3258}
3259
3260static inline void memcg_resume_kmem_account(void)
3261{
3262 VM_BUG_ON(!current->mm);
3263 current->memcg_kmem_skip_account--;
3264}
3265
Glauber Costa1f458cb2012-12-18 14:22:50 -08003266static void kmem_cache_destroy_work_func(struct work_struct *w)
3267{
3268 struct kmem_cache *cachep;
3269 struct memcg_cache_params *p;
3270
3271 p = container_of(w, struct memcg_cache_params, destroy);
3272
3273 cachep = memcg_params_to_cache(p);
3274
Glauber Costa22933152012-12-18 14:22:59 -08003275 /*
3276 * If we get down to 0 after shrink, we could delete right away.
3277 * However, memcg_release_pages() already puts us back in the workqueue
3278 * in that case. If we proceed deleting, we'll get a dangling
3279 * reference, and removing the object from the workqueue in that case
3280 * is unnecessary complication. We are not a fast path.
3281 *
3282 * Note that this case is fundamentally different from racing with
3283 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3284 * kmem_cache_shrink, not only we would be reinserting a dead cache
3285 * into the queue, but doing so from inside the worker racing to
3286 * destroy it.
3287 *
3288 * So if we aren't down to zero, we'll just schedule a worker and try
3289 * again
3290 */
3291 if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
3292 kmem_cache_shrink(cachep);
3293 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
3294 return;
3295 } else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003296 kmem_cache_destroy(cachep);
3297}
3298
3299void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3300{
3301 if (!cachep->memcg_params->dead)
3302 return;
3303
3304 /*
Glauber Costa22933152012-12-18 14:22:59 -08003305 * There are many ways in which we can get here.
3306 *
3307 * We can get to a memory-pressure situation while the delayed work is
3308 * still pending to run. The vmscan shrinkers can then release all
3309 * cache memory and get us to destruction. If this is the case, we'll
3310 * be executed twice, which is a bug (the second time will execute over
3311 * bogus data). In this case, cancelling the work should be fine.
3312 *
3313 * But we can also get here from the worker itself, if
3314 * kmem_cache_shrink is enough to shake all the remaining objects and
3315 * get the page count to 0. In this case, we'll deadlock if we try to
3316 * cancel the work (the worker runs with an internal lock held, which
3317 * is the same lock we would hold for cancel_work_sync().)
3318 *
3319 * Since we can't possibly know who got us here, just refrain from
3320 * running if there is already work pending
3321 */
3322 if (work_pending(&cachep->memcg_params->destroy))
3323 return;
3324 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003325 * We have to defer the actual destroying to a workqueue, because
3326 * we might currently be in a context that cannot sleep.
3327 */
3328 schedule_work(&cachep->memcg_params->destroy);
3329}
3330
Glauber Costad7f25f82012-12-18 14:22:40 -08003331/*
3332 * This lock protects updaters, not readers. We want readers to be as fast as
3333 * they can, and they will either see NULL or a valid cache value. Our model
3334 * allow them to see NULL, in which case the root memcg will be selected.
3335 *
3336 * We need this lock because multiple allocations to the same cache from a non
3337 * will span more than one worker. Only one of them can create the cache.
3338 */
3339static DEFINE_MUTEX(memcg_cache_mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003340
3341/*
3342 * Called with memcg_cache_mutex held
3343 */
3344static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
3345 struct kmem_cache *s)
3346{
3347 struct kmem_cache *new;
3348 static char *tmp_name = NULL;
3349
3350 lockdep_assert_held(&memcg_cache_mutex);
3351
3352 /*
3353 * kmem_cache_create_memcg duplicates the given name and
3354 * cgroup_name for this name requires RCU context.
3355 * This static temporary buffer is used to prevent from
3356 * pointless shortliving allocation.
3357 */
3358 if (!tmp_name) {
3359 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3360 if (!tmp_name)
3361 return NULL;
3362 }
3363
3364 rcu_read_lock();
3365 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3366 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3367 rcu_read_unlock();
3368
3369 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3370 (s->flags & ~SLAB_PANIC), s->ctor, s);
3371
3372 if (new)
3373 new->allocflags |= __GFP_KMEMCG;
3374
3375 return new;
3376}
3377
Glauber Costad7f25f82012-12-18 14:22:40 -08003378static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3379 struct kmem_cache *cachep)
3380{
3381 struct kmem_cache *new_cachep;
3382 int idx;
3383
3384 BUG_ON(!memcg_can_account_kmem(memcg));
3385
3386 idx = memcg_cache_id(memcg);
3387
3388 mutex_lock(&memcg_cache_mutex);
3389 new_cachep = cachep->memcg_params->memcg_caches[idx];
3390 if (new_cachep)
3391 goto out;
3392
3393 new_cachep = kmem_cache_dup(memcg, cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003394 if (new_cachep == NULL) {
3395 new_cachep = cachep;
3396 goto out;
3397 }
3398
3399 mem_cgroup_get(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08003400 atomic_set(&new_cachep->memcg_params->nr_pages , 0);
Glauber Costad7f25f82012-12-18 14:22:40 -08003401
3402 cachep->memcg_params->memcg_caches[idx] = new_cachep;
3403 /*
3404 * the readers won't lock, make sure everybody sees the updated value,
3405 * so they won't put stuff in the queue again for no reason
3406 */
3407 wmb();
3408out:
3409 mutex_unlock(&memcg_cache_mutex);
3410 return new_cachep;
3411}
3412
Glauber Costa7cf27982012-12-18 14:22:55 -08003413void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3414{
3415 struct kmem_cache *c;
3416 int i;
3417
3418 if (!s->memcg_params)
3419 return;
3420 if (!s->memcg_params->is_root_cache)
3421 return;
3422
3423 /*
3424 * If the cache is being destroyed, we trust that there is no one else
3425 * requesting objects from it. Even if there are, the sanity checks in
3426 * kmem_cache_destroy should caught this ill-case.
3427 *
3428 * Still, we don't want anyone else freeing memcg_caches under our
3429 * noses, which can happen if a new memcg comes to life. As usual,
3430 * we'll take the set_limit_mutex to protect ourselves against this.
3431 */
3432 mutex_lock(&set_limit_mutex);
3433 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3434 c = s->memcg_params->memcg_caches[i];
3435 if (!c)
3436 continue;
3437
3438 /*
3439 * We will now manually delete the caches, so to avoid races
3440 * we need to cancel all pending destruction workers and
3441 * proceed with destruction ourselves.
3442 *
3443 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3444 * and that could spawn the workers again: it is likely that
3445 * the cache still have active pages until this very moment.
3446 * This would lead us back to mem_cgroup_destroy_cache.
3447 *
3448 * But that will not execute at all if the "dead" flag is not
3449 * set, so flip it down to guarantee we are in control.
3450 */
3451 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003452 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003453 kmem_cache_destroy(c);
3454 }
3455 mutex_unlock(&set_limit_mutex);
3456}
3457
Glauber Costad7f25f82012-12-18 14:22:40 -08003458struct create_work {
3459 struct mem_cgroup *memcg;
3460 struct kmem_cache *cachep;
3461 struct work_struct work;
3462};
3463
Glauber Costa1f458cb2012-12-18 14:22:50 -08003464static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3465{
3466 struct kmem_cache *cachep;
3467 struct memcg_cache_params *params;
3468
3469 if (!memcg_kmem_is_active(memcg))
3470 return;
3471
3472 mutex_lock(&memcg->slab_caches_mutex);
3473 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3474 cachep = memcg_params_to_cache(params);
3475 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003476 schedule_work(&cachep->memcg_params->destroy);
3477 }
3478 mutex_unlock(&memcg->slab_caches_mutex);
3479}
3480
Glauber Costad7f25f82012-12-18 14:22:40 -08003481static void memcg_create_cache_work_func(struct work_struct *w)
3482{
3483 struct create_work *cw;
3484
3485 cw = container_of(w, struct create_work, work);
3486 memcg_create_kmem_cache(cw->memcg, cw->cachep);
3487 /* Drop the reference gotten when we enqueued. */
3488 css_put(&cw->memcg->css);
3489 kfree(cw);
3490}
3491
3492/*
3493 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003494 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003495static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3496 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003497{
3498 struct create_work *cw;
3499
3500 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003501 if (cw == NULL) {
3502 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003503 return;
3504 }
3505
3506 cw->memcg = memcg;
3507 cw->cachep = cachep;
3508
3509 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3510 schedule_work(&cw->work);
3511}
3512
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003513static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3514 struct kmem_cache *cachep)
3515{
3516 /*
3517 * We need to stop accounting when we kmalloc, because if the
3518 * corresponding kmalloc cache is not yet created, the first allocation
3519 * in __memcg_create_cache_enqueue will recurse.
3520 *
3521 * However, it is better to enclose the whole function. Depending on
3522 * the debugging options enabled, INIT_WORK(), for instance, can
3523 * trigger an allocation. This too, will make us recurse. Because at
3524 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3525 * the safest choice is to do it like this, wrapping the whole function.
3526 */
3527 memcg_stop_kmem_account();
3528 __memcg_create_cache_enqueue(memcg, cachep);
3529 memcg_resume_kmem_account();
3530}
Glauber Costad7f25f82012-12-18 14:22:40 -08003531/*
3532 * Return the kmem_cache we're supposed to use for a slab allocation.
3533 * We try to use the current memcg's version of the cache.
3534 *
3535 * If the cache does not exist yet, if we are the first user of it,
3536 * we either create it immediately, if possible, or create it asynchronously
3537 * in a workqueue.
3538 * In the latter case, we will let the current allocation go through with
3539 * the original cache.
3540 *
3541 * Can't be called in interrupt context or from kernel threads.
3542 * This function needs to be called with rcu_read_lock() held.
3543 */
3544struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3545 gfp_t gfp)
3546{
3547 struct mem_cgroup *memcg;
3548 int idx;
3549
3550 VM_BUG_ON(!cachep->memcg_params);
3551 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3552
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003553 if (!current->mm || current->memcg_kmem_skip_account)
3554 return cachep;
3555
Glauber Costad7f25f82012-12-18 14:22:40 -08003556 rcu_read_lock();
3557 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003558
3559 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003560 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003561
3562 idx = memcg_cache_id(memcg);
3563
3564 /*
3565 * barrier to mare sure we're always seeing the up to date value. The
3566 * code updating memcg_caches will issue a write barrier to match this.
3567 */
3568 read_barrier_depends();
Li Zefanca0dde92013-04-29 15:08:57 -07003569 if (likely(cachep->memcg_params->memcg_caches[idx])) {
3570 cachep = cachep->memcg_params->memcg_caches[idx];
3571 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003572 }
3573
Li Zefanca0dde92013-04-29 15:08:57 -07003574 /* The corresponding put will be done in the workqueue. */
3575 if (!css_tryget(&memcg->css))
3576 goto out;
3577 rcu_read_unlock();
3578
3579 /*
3580 * If we are in a safe context (can wait, and not in interrupt
3581 * context), we could be be predictable and return right away.
3582 * This would guarantee that the allocation being performed
3583 * already belongs in the new cache.
3584 *
3585 * However, there are some clashes that can arrive from locking.
3586 * For instance, because we acquire the slab_mutex while doing
3587 * kmem_cache_dup, this means no further allocation could happen
3588 * with the slab_mutex held.
3589 *
3590 * Also, because cache creation issue get_online_cpus(), this
3591 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3592 * that ends up reversed during cpu hotplug. (cpuset allocates
3593 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3594 * better to defer everything.
3595 */
3596 memcg_create_cache_enqueue(memcg, cachep);
3597 return cachep;
3598out:
3599 rcu_read_unlock();
3600 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003601}
3602EXPORT_SYMBOL(__memcg_kmem_get_cache);
3603
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003604/*
3605 * We need to verify if the allocation against current->mm->owner's memcg is
3606 * possible for the given order. But the page is not allocated yet, so we'll
3607 * need a further commit step to do the final arrangements.
3608 *
3609 * It is possible for the task to switch cgroups in this mean time, so at
3610 * commit time, we can't rely on task conversion any longer. We'll then use
3611 * the handle argument to return to the caller which cgroup we should commit
3612 * against. We could also return the memcg directly and avoid the pointer
3613 * passing, but a boolean return value gives better semantics considering
3614 * the compiled-out case as well.
3615 *
3616 * Returning true means the allocation is possible.
3617 */
3618bool
3619__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3620{
3621 struct mem_cgroup *memcg;
3622 int ret;
3623
3624 *_memcg = NULL;
3625 memcg = try_get_mem_cgroup_from_mm(current->mm);
3626
3627 /*
3628 * very rare case described in mem_cgroup_from_task. Unfortunately there
3629 * isn't much we can do without complicating this too much, and it would
3630 * be gfp-dependent anyway. Just let it go
3631 */
3632 if (unlikely(!memcg))
3633 return true;
3634
3635 if (!memcg_can_account_kmem(memcg)) {
3636 css_put(&memcg->css);
3637 return true;
3638 }
3639
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003640 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3641 if (!ret)
3642 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003643
3644 css_put(&memcg->css);
3645 return (ret == 0);
3646}
3647
3648void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3649 int order)
3650{
3651 struct page_cgroup *pc;
3652
3653 VM_BUG_ON(mem_cgroup_is_root(memcg));
3654
3655 /* The page allocation failed. Revert */
3656 if (!page) {
3657 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003658 return;
3659 }
3660
3661 pc = lookup_page_cgroup(page);
3662 lock_page_cgroup(pc);
3663 pc->mem_cgroup = memcg;
3664 SetPageCgroupUsed(pc);
3665 unlock_page_cgroup(pc);
3666}
3667
3668void __memcg_kmem_uncharge_pages(struct page *page, int order)
3669{
3670 struct mem_cgroup *memcg = NULL;
3671 struct page_cgroup *pc;
3672
3673
3674 pc = lookup_page_cgroup(page);
3675 /*
3676 * Fast unlocked return. Theoretically might have changed, have to
3677 * check again after locking.
3678 */
3679 if (!PageCgroupUsed(pc))
3680 return;
3681
3682 lock_page_cgroup(pc);
3683 if (PageCgroupUsed(pc)) {
3684 memcg = pc->mem_cgroup;
3685 ClearPageCgroupUsed(pc);
3686 }
3687 unlock_page_cgroup(pc);
3688
3689 /*
3690 * We trust that only if there is a memcg associated with the page, it
3691 * is a valid allocation
3692 */
3693 if (!memcg)
3694 return;
3695
3696 VM_BUG_ON(mem_cgroup_is_root(memcg));
3697 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003698}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003699#else
3700static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3701{
3702}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003703#endif /* CONFIG_MEMCG_KMEM */
3704
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003705#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3706
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003707#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003708/*
3709 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003710 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3711 * charge/uncharge will be never happen and move_account() is done under
3712 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003713 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003714void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003715{
3716 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003717 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003718 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003719 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003720
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003721 if (mem_cgroup_disabled())
3722 return;
David Rientjesb070e652013-05-07 16:18:09 -07003723
3724 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003725 for (i = 1; i < HPAGE_PMD_NR; i++) {
3726 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003727 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003728 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003729 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3730 }
David Rientjesb070e652013-05-07 16:18:09 -07003731 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3732 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003733}
Hugh Dickins12d27102012-01-12 17:19:52 -08003734#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003735
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003736/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003737 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003738 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003739 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003740 * @pc: page_cgroup of the page.
3741 * @from: mem_cgroup which the page is moved from.
3742 * @to: mem_cgroup which the page is moved to. @from != @to.
3743 *
3744 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003745 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003746 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003747 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003748 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3749 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003750 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003751static int mem_cgroup_move_account(struct page *page,
3752 unsigned int nr_pages,
3753 struct page_cgroup *pc,
3754 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003755 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003756{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003757 unsigned long flags;
3758 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003759 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003760
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003761 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07003762 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07003763 /*
3764 * The page is isolated from LRU. So, collapse function
3765 * will not handle this page. But page splitting can happen.
3766 * Do this check under compound_page_lock(). The caller should
3767 * hold it.
3768 */
3769 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003770 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003771 goto out;
3772
3773 lock_page_cgroup(pc);
3774
3775 ret = -EINVAL;
3776 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3777 goto unlock;
3778
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003779 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003780
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07003781 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003782 /* Update mapped_file data for mem_cgroup */
3783 preempt_disable();
3784 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3785 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3786 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07003787 }
David Rientjesb070e652013-05-07 16:18:09 -07003788 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003789
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003790 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003791 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003792 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003793 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003794 ret = 0;
3795unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003796 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003797 /*
3798 * check events
3799 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003800 memcg_check_events(to, page);
3801 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003802out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003803 return ret;
3804}
3805
Michal Hocko2ef37d32012-10-26 13:37:30 +02003806/**
3807 * mem_cgroup_move_parent - moves page to the parent group
3808 * @page: the page to move
3809 * @pc: page_cgroup of the page
3810 * @child: page's cgroup
3811 *
3812 * move charges to its parent or the root cgroup if the group has no
3813 * parent (aka use_hierarchy==0).
3814 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3815 * mem_cgroup_move_account fails) the failure is always temporary and
3816 * it signals a race with a page removal/uncharge or migration. In the
3817 * first case the page is on the way out and it will vanish from the LRU
3818 * on the next attempt and the call should be retried later.
3819 * Isolation from the LRU fails only if page has been isolated from
3820 * the LRU since we looked at it and that usually means either global
3821 * reclaim or migration going on. The page will either get back to the
3822 * LRU or vanish.
3823 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3824 * (!PageCgroupUsed) or moved to a different group. The page will
3825 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003826 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003827static int mem_cgroup_move_parent(struct page *page,
3828 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003829 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003830{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003831 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003832 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003833 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003834 int ret;
3835
Michal Hockod8423012012-10-26 13:37:29 +02003836 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003837
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003838 ret = -EBUSY;
3839 if (!get_page_unless_zero(page))
3840 goto out;
3841 if (isolate_lru_page(page))
3842 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003843
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003844 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003845
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003846 parent = parent_mem_cgroup(child);
3847 /*
3848 * If no parent, move charges to root cgroup.
3849 */
3850 if (!parent)
3851 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003852
Michal Hocko2ef37d32012-10-26 13:37:30 +02003853 if (nr_pages > 1) {
3854 VM_BUG_ON(!PageTransHuge(page));
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003855 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003856 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003857
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003858 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003859 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003860 if (!ret)
3861 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003862
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003863 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003864 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003865 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003866put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003867 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003868out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003869 return ret;
3870}
3871
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003872/*
3873 * Charge the memory controller for page usage.
3874 * Return
3875 * 0 if the charge was successful
3876 * < 0 if the cgroup is over its limit
3877 */
3878static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003879 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003880{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003881 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003882 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003883 bool oom = true;
3884 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003885
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003886 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003887 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003888 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08003889 /*
3890 * Never OOM-kill a process for a huge page. The
3891 * fault handler will fall back to regular pages.
3892 */
3893 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003894 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003895
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003896 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003897 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003898 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003899 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003900 return 0;
3901}
3902
3903int mem_cgroup_newpage_charge(struct page *page,
3904 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003905{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003906 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003907 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003908 VM_BUG_ON(page_mapped(page));
3909 VM_BUG_ON(page->mapping && !PageAnon(page));
3910 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003911 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003912 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003913}
3914
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003915/*
3916 * While swap-in, try_charge -> commit or cancel, the page is locked.
3917 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003918 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003919 * "commit()" or removed by "cancel()"
3920 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003921static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3922 struct page *page,
3923 gfp_t mask,
3924 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003925{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003926 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003927 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003928 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003929
Johannes Weiner90deb782012-07-31 16:45:47 -07003930 pc = lookup_page_cgroup(page);
3931 /*
3932 * Every swap fault against a single page tries to charge the
3933 * page, bail as early as possible. shmem_unuse() encounters
3934 * already charged pages, too. The USED bit is protected by
3935 * the page lock, which serializes swap cache removal, which
3936 * in turn serializes uncharging.
3937 */
3938 if (PageCgroupUsed(pc))
3939 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003940 if (!do_swap_account)
3941 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003942 memcg = try_get_mem_cgroup_from_page(page);
3943 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003944 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08003945 *memcgp = memcg;
3946 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003947 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003948 if (ret == -EINTR)
3949 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003950 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003951charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003952 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
3953 if (ret == -EINTR)
3954 ret = 0;
3955 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003956}
3957
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003958int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
3959 gfp_t gfp_mask, struct mem_cgroup **memcgp)
3960{
3961 *memcgp = NULL;
3962 if (mem_cgroup_disabled())
3963 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003964 /*
3965 * A racing thread's fault, or swapoff, may have already
3966 * updated the pte, and even removed page from swap cache: in
3967 * those cases unuse_pte()'s pte_same() test will fail; but
3968 * there's also a KSM case which does need to charge the page.
3969 */
3970 if (!PageSwapCache(page)) {
3971 int ret;
3972
3973 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
3974 if (ret == -EINTR)
3975 ret = 0;
3976 return ret;
3977 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003978 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
3979}
3980
Johannes Weiner827a03d2012-07-31 16:45:36 -07003981void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
3982{
3983 if (mem_cgroup_disabled())
3984 return;
3985 if (!memcg)
3986 return;
3987 __mem_cgroup_cancel_charge(memcg, 1);
3988}
3989
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003990static void
Johannes Weiner72835c82012-01-12 17:18:32 -08003991__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003992 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003993{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003994 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003995 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08003996 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003997 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07003998
Johannes Weinerce587e62012-04-24 20:22:33 +02003999 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004000 /*
4001 * Now swap is on-memory. This means this page may be
4002 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004003 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
4004 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
4005 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004006 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004007 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004008 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004009 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004010 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004011}
4012
Johannes Weiner72835c82012-01-12 17:18:32 -08004013void mem_cgroup_commit_charge_swapin(struct page *page,
4014 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004015{
Johannes Weiner72835c82012-01-12 17:18:32 -08004016 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004017 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004018}
4019
Johannes Weiner827a03d2012-07-31 16:45:36 -07004020int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4021 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004022{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004023 struct mem_cgroup *memcg = NULL;
4024 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4025 int ret;
4026
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004027 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004028 return 0;
4029 if (PageCompound(page))
4030 return 0;
4031
Johannes Weiner827a03d2012-07-31 16:45:36 -07004032 if (!PageSwapCache(page))
4033 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4034 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004035 ret = __mem_cgroup_try_charge_swapin(mm, page,
4036 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004037 if (!ret)
4038 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4039 }
4040 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004041}
4042
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004043static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004044 unsigned int nr_pages,
4045 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004046{
4047 struct memcg_batch_info *batch = NULL;
4048 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004049
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004050 /* If swapout, usage of swap doesn't decrease */
4051 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4052 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004053
4054 batch = &current->memcg_batch;
4055 /*
4056 * In usual, we do css_get() when we remember memcg pointer.
4057 * But in this case, we keep res->usage until end of a series of
4058 * uncharges. Then, it's ok to ignore memcg's refcnt.
4059 */
4060 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004061 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004062 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004063 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004064 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004065 * the same cgroup and we have chance to coalesce uncharges.
4066 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4067 * because we want to do uncharge as soon as possible.
4068 */
4069
4070 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4071 goto direct_uncharge;
4072
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004073 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004074 goto direct_uncharge;
4075
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004076 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004077 * In typical case, batch->memcg == mem. This means we can
4078 * merge a series of uncharges to an uncharge of res_counter.
4079 * If not, we uncharge res_counter ony by one.
4080 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004081 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004082 goto direct_uncharge;
4083 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004084 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004085 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004086 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004087 return;
4088direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004089 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004090 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004091 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4092 if (unlikely(batch->memcg != memcg))
4093 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004094}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004095
Balbir Singh8697d332008-02-07 00:13:59 -08004096/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004097 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004098 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004099static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004100__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4101 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004102{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004103 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004104 unsigned int nr_pages = 1;
4105 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004106 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004107
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004108 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004109 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004110
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004111 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004112 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004113 VM_BUG_ON(!PageTransHuge(page));
4114 }
Balbir Singh8697d332008-02-07 00:13:59 -08004115 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004116 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004117 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004118 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004119 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004120 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004121
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004122 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004123
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004124 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004125
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004126 if (!PageCgroupUsed(pc))
4127 goto unlock_out;
4128
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004129 anon = PageAnon(page);
4130
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004131 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004132 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004133 /*
4134 * Generally PageAnon tells if it's the anon statistics to be
4135 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4136 * used before page reached the stage of being marked PageAnon.
4137 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004138 anon = true;
4139 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004140 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004141 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004142 if (page_mapped(page))
4143 goto unlock_out;
4144 /*
4145 * Pages under migration may not be uncharged. But
4146 * end_migration() /must/ be the one uncharging the
4147 * unused post-migration page and so it has to call
4148 * here with the migration bit still set. See the
4149 * res_counter handling below.
4150 */
4151 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004152 goto unlock_out;
4153 break;
4154 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4155 if (!PageAnon(page)) { /* Shared memory */
4156 if (page->mapping && !page_is_file_cache(page))
4157 goto unlock_out;
4158 } else if (page_mapped(page)) /* Anon */
4159 goto unlock_out;
4160 break;
4161 default:
4162 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004163 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004164
David Rientjesb070e652013-05-07 16:18:09 -07004165 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004166
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004167 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004168 /*
4169 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4170 * freed from LRU. This is safe because uncharged page is expected not
4171 * to be reused (freed soon). Exception is SwapCache, it's handled by
4172 * special functions.
4173 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004174
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004175 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004176 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004177 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004178 * will never be freed.
4179 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004180 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004181 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004182 mem_cgroup_swap_statistics(memcg, true);
4183 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004184 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004185 /*
4186 * Migration does not charge the res_counter for the
4187 * replacement page, so leave it alone when phasing out the
4188 * page that is unused after the migration.
4189 */
4190 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004191 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004192
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004193 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004194
4195unlock_out:
4196 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004197 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004198}
4199
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004200void mem_cgroup_uncharge_page(struct page *page)
4201{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004202 /* early check. */
4203 if (page_mapped(page))
4204 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08004205 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004206 /*
4207 * If the page is in swap cache, uncharge should be deferred
4208 * to the swap path, which also properly accounts swap usage
4209 * and handles memcg lifetime.
4210 *
4211 * Note that this check is not stable and reclaim may add the
4212 * page to swap cache at any time after this. However, if the
4213 * page is not in swap cache by the time page->mapcount hits
4214 * 0, there won't be any page table references to the swap
4215 * slot, and reclaim will free it and not actually write the
4216 * page to disk.
4217 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004218 if (PageSwapCache(page))
4219 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004220 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004221}
4222
4223void mem_cgroup_uncharge_cache_page(struct page *page)
4224{
4225 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07004226 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07004227 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004228}
4229
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004230/*
4231 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4232 * In that cases, pages are freed continuously and we can expect pages
4233 * are in the same memcg. All these calls itself limits the number of
4234 * pages freed at once, then uncharge_start/end() is called properly.
4235 * This may be called prural(2) times in a context,
4236 */
4237
4238void mem_cgroup_uncharge_start(void)
4239{
4240 current->memcg_batch.do_batch++;
4241 /* We can do nest. */
4242 if (current->memcg_batch.do_batch == 1) {
4243 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004244 current->memcg_batch.nr_pages = 0;
4245 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004246 }
4247}
4248
4249void mem_cgroup_uncharge_end(void)
4250{
4251 struct memcg_batch_info *batch = &current->memcg_batch;
4252
4253 if (!batch->do_batch)
4254 return;
4255
4256 batch->do_batch--;
4257 if (batch->do_batch) /* If stacked, do nothing. */
4258 return;
4259
4260 if (!batch->memcg)
4261 return;
4262 /*
4263 * This "batch->memcg" is valid without any css_get/put etc...
4264 * bacause we hide charges behind us.
4265 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004266 if (batch->nr_pages)
4267 res_counter_uncharge(&batch->memcg->res,
4268 batch->nr_pages * PAGE_SIZE);
4269 if (batch->memsw_nr_pages)
4270 res_counter_uncharge(&batch->memcg->memsw,
4271 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004272 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004273 /* forget this pointer (for sanity check) */
4274 batch->memcg = NULL;
4275}
4276
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004277#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004278/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004279 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004280 * memcg information is recorded to swap_cgroup of "ent"
4281 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004282void
4283mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004284{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004285 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004286 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004287
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004288 if (!swapout) /* this was a swap cache but the swap is unused ! */
4289 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4290
Johannes Weiner0030f532012-07-31 16:45:25 -07004291 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004292
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004293 /*
4294 * record memcg information, if swapout && memcg != NULL,
4295 * mem_cgroup_get() was called in uncharge().
4296 */
4297 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004298 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004299}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004300#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004301
Andrew Mortonc255a452012-07-31 16:43:02 -07004302#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004303/*
4304 * called from swap_entry_free(). remove record in swap_cgroup and
4305 * uncharge "memsw" account.
4306 */
4307void mem_cgroup_uncharge_swap(swp_entry_t ent)
4308{
4309 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004310 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004311
4312 if (!do_swap_account)
4313 return;
4314
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004315 id = swap_cgroup_record(ent, 0);
4316 rcu_read_lock();
4317 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004318 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004319 /*
4320 * We uncharge this because swap is freed.
4321 * This memcg can be obsolete one. We avoid calling css_tryget
4322 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004323 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004324 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004325 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004326 mem_cgroup_put(memcg);
4327 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004328 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004329}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004330
4331/**
4332 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4333 * @entry: swap entry to be moved
4334 * @from: mem_cgroup which the entry is moved from
4335 * @to: mem_cgroup which the entry is moved to
4336 *
4337 * It succeeds only when the swap_cgroup's record for this entry is the same
4338 * as the mem_cgroup's id of @from.
4339 *
4340 * Returns 0 on success, -EINVAL on failure.
4341 *
4342 * The caller must have charged to @to, IOW, called res_counter_charge() about
4343 * both res and memsw, and called css_get().
4344 */
4345static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004346 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004347{
4348 unsigned short old_id, new_id;
4349
4350 old_id = css_id(&from->css);
4351 new_id = css_id(&to->css);
4352
4353 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004354 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004355 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004356 /*
4357 * This function is only called from task migration context now.
4358 * It postpones res_counter and refcount handling till the end
4359 * of task migration(mem_cgroup_clear_mc()) for performance
4360 * improvement. But we cannot postpone mem_cgroup_get(to)
4361 * because if the process that has been moved to @to does
4362 * swap-in, the refcount of @to might be decreased to 0.
4363 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08004364 mem_cgroup_get(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004365 return 0;
4366 }
4367 return -EINVAL;
4368}
4369#else
4370static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004371 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004372{
4373 return -EINVAL;
4374}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004375#endif
4376
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004377/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004378 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4379 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004380 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004381void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4382 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004383{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004384 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004385 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004386 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004387 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004388
Johannes Weiner72835c82012-01-12 17:18:32 -08004389 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004390
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004391 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004392 return;
Balbir Singh40779602008-04-04 14:29:59 -07004393
Mel Gormanb32967f2012-11-19 12:35:47 +00004394 if (PageTransHuge(page))
4395 nr_pages <<= compound_order(page);
4396
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004397 pc = lookup_page_cgroup(page);
4398 lock_page_cgroup(pc);
4399 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004400 memcg = pc->mem_cgroup;
4401 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004402 /*
4403 * At migrating an anonymous page, its mapcount goes down
4404 * to 0 and uncharge() will be called. But, even if it's fully
4405 * unmapped, migration may fail and this page has to be
4406 * charged again. We set MIGRATION flag here and delay uncharge
4407 * until end_migration() is called
4408 *
4409 * Corner Case Thinking
4410 * A)
4411 * When the old page was mapped as Anon and it's unmap-and-freed
4412 * while migration was ongoing.
4413 * If unmap finds the old page, uncharge() of it will be delayed
4414 * until end_migration(). If unmap finds a new page, it's
4415 * uncharged when it make mapcount to be 1->0. If unmap code
4416 * finds swap_migration_entry, the new page will not be mapped
4417 * and end_migration() will find it(mapcount==0).
4418 *
4419 * B)
4420 * When the old page was mapped but migraion fails, the kernel
4421 * remaps it. A charge for it is kept by MIGRATION flag even
4422 * if mapcount goes down to 0. We can do remap successfully
4423 * without charging it again.
4424 *
4425 * C)
4426 * The "old" page is under lock_page() until the end of
4427 * migration, so, the old page itself will not be swapped-out.
4428 * If the new page is swapped out before end_migraton, our
4429 * hook to usual swap-out path will catch the event.
4430 */
4431 if (PageAnon(page))
4432 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004433 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004434 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004435 /*
4436 * If the page is not charged at this point,
4437 * we return here.
4438 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004439 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004440 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004441
Johannes Weiner72835c82012-01-12 17:18:32 -08004442 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004443 /*
4444 * We charge new page before it's used/mapped. So, even if unlock_page()
4445 * is called before end_migration, we can catch all events on this new
4446 * page. In the case new page is migrated but not remapped, new page's
4447 * mapcount will be finally 0 and we call uncharge in end_migration().
4448 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004449 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004450 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004451 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004452 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004453 /*
4454 * The page is committed to the memcg, but it's not actually
4455 * charged to the res_counter since we plan on replacing the
4456 * old one and only one page is going to be left afterwards.
4457 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004458 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004459}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004460
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004461/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004462void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004463 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004464{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004465 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004466 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004467 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004468
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004469 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004470 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004471
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004472 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004473 used = oldpage;
4474 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004475 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004476 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004477 unused = oldpage;
4478 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004479 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004480 __mem_cgroup_uncharge_common(unused,
4481 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4482 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4483 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004484 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004485 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004486 * We disallowed uncharge of pages under migration because mapcount
4487 * of the page goes down to zero, temporarly.
4488 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004489 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004490 pc = lookup_page_cgroup(oldpage);
4491 lock_page_cgroup(pc);
4492 ClearPageCgroupMigration(pc);
4493 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004494
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004495 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004496 * If a page is a file cache, radix-tree replacement is very atomic
4497 * and we can skip this check. When it was an Anon page, its mapcount
4498 * goes down to 0. But because we added MIGRATION flage, it's not
4499 * uncharged yet. There are several case but page->mapcount check
4500 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4501 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004502 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004503 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004504 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004505}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004506
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004507/*
4508 * At replace page cache, newpage is not under any memcg but it's on
4509 * LRU. So, this function doesn't touch res_counter but handles LRU
4510 * in correct way. Both pages are locked so we cannot race with uncharge.
4511 */
4512void mem_cgroup_replace_page_cache(struct page *oldpage,
4513 struct page *newpage)
4514{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004515 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004516 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004517 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004518
4519 if (mem_cgroup_disabled())
4520 return;
4521
4522 pc = lookup_page_cgroup(oldpage);
4523 /* fix accounting on old pages */
4524 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004525 if (PageCgroupUsed(pc)) {
4526 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004527 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004528 ClearPageCgroupUsed(pc);
4529 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004530 unlock_page_cgroup(pc);
4531
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004532 /*
4533 * When called from shmem_replace_page(), in some cases the
4534 * oldpage has already been charged, and in some cases not.
4535 */
4536 if (!memcg)
4537 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004538 /*
4539 * Even if newpage->mapping was NULL before starting replacement,
4540 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4541 * LRU while we overwrite pc->mem_cgroup.
4542 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004543 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004544}
4545
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004546#ifdef CONFIG_DEBUG_VM
4547static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4548{
4549 struct page_cgroup *pc;
4550
4551 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004552 /*
4553 * Can be NULL while feeding pages into the page allocator for
4554 * the first time, i.e. during boot or memory hotplug;
4555 * or when mem_cgroup_disabled().
4556 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004557 if (likely(pc) && PageCgroupUsed(pc))
4558 return pc;
4559 return NULL;
4560}
4561
4562bool mem_cgroup_bad_page_check(struct page *page)
4563{
4564 if (mem_cgroup_disabled())
4565 return false;
4566
4567 return lookup_page_cgroup_used(page) != NULL;
4568}
4569
4570void mem_cgroup_print_bad_page(struct page *page)
4571{
4572 struct page_cgroup *pc;
4573
4574 pc = lookup_page_cgroup_used(page);
4575 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004576 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4577 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004578 }
4579}
4580#endif
4581
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004582static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004583 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004584{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004585 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004586 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004587 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004588 int children = mem_cgroup_count_children(memcg);
4589 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004590 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004591
4592 /*
4593 * For keeping hierarchical_reclaim simple, how long we should retry
4594 * is depends on callers. We set our retry-count to be function
4595 * of # of children which we should visit in this loop.
4596 */
4597 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4598
4599 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004600
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004601 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004602 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004603 if (signal_pending(current)) {
4604 ret = -EINTR;
4605 break;
4606 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004607 /*
4608 * Rather than hide all in some function, I do this in
4609 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004610 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004611 */
4612 mutex_lock(&set_limit_mutex);
4613 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4614 if (memswlimit < val) {
4615 ret = -EINVAL;
4616 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004617 break;
4618 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004619
4620 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4621 if (memlimit < val)
4622 enlarge = 1;
4623
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004624 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004625 if (!ret) {
4626 if (memswlimit == val)
4627 memcg->memsw_is_minimum = true;
4628 else
4629 memcg->memsw_is_minimum = false;
4630 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004631 mutex_unlock(&set_limit_mutex);
4632
4633 if (!ret)
4634 break;
4635
Johannes Weiner56600482012-01-12 17:17:59 -08004636 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4637 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004638 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4639 /* Usage is reduced ? */
4640 if (curusage >= oldusage)
4641 retry_count--;
4642 else
4643 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004644 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004645 if (!ret && enlarge)
4646 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004647
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004648 return ret;
4649}
4650
Li Zefan338c8432009-06-17 16:27:15 -07004651static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4652 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004653{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004654 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004655 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004656 int children = mem_cgroup_count_children(memcg);
4657 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004658 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004659
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004660 /* see mem_cgroup_resize_res_limit */
4661 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
4662 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004663 while (retry_count) {
4664 if (signal_pending(current)) {
4665 ret = -EINTR;
4666 break;
4667 }
4668 /*
4669 * Rather than hide all in some function, I do this in
4670 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004671 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004672 */
4673 mutex_lock(&set_limit_mutex);
4674 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4675 if (memlimit > val) {
4676 ret = -EINVAL;
4677 mutex_unlock(&set_limit_mutex);
4678 break;
4679 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004680 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4681 if (memswlimit < val)
4682 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004683 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004684 if (!ret) {
4685 if (memlimit == val)
4686 memcg->memsw_is_minimum = true;
4687 else
4688 memcg->memsw_is_minimum = false;
4689 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004690 mutex_unlock(&set_limit_mutex);
4691
4692 if (!ret)
4693 break;
4694
Johannes Weiner56600482012-01-12 17:17:59 -08004695 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4696 MEM_CGROUP_RECLAIM_NOSWAP |
4697 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004698 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004699 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004700 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004701 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004702 else
4703 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004704 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004705 if (!ret && enlarge)
4706 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004707 return ret;
4708}
4709
Balbir Singh4e416952009-09-23 15:56:39 -07004710unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07004711 gfp_t gfp_mask,
4712 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07004713{
4714 unsigned long nr_reclaimed = 0;
4715 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4716 unsigned long reclaimed;
4717 int loop = 0;
4718 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07004719 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07004720 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07004721
4722 if (order > 0)
4723 return 0;
4724
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07004725 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07004726 /*
4727 * This loop can run a while, specially if mem_cgroup's continuously
4728 * keep exceeding their soft limit and putting the system under
4729 * pressure
4730 */
4731 do {
4732 if (next_mz)
4733 mz = next_mz;
4734 else
4735 mz = mem_cgroup_largest_soft_limit_node(mctz);
4736 if (!mz)
4737 break;
4738
Ying Han0ae5e892011-05-26 16:25:25 -07004739 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004740 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08004741 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07004742 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07004743 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07004744 spin_lock(&mctz->lock);
4745
4746 /*
4747 * If we failed to reclaim anything from this memory cgroup
4748 * it is time to move on to the next cgroup
4749 */
4750 next_mz = NULL;
4751 if (!reclaimed) {
4752 do {
4753 /*
4754 * Loop until we find yet another one.
4755 *
4756 * By the time we get the soft_limit lock
4757 * again, someone might have aded the
4758 * group back on the RB tree. Iterate to
4759 * make sure we get a different mem.
4760 * mem_cgroup_largest_soft_limit_node returns
4761 * NULL if no other cgroup is present on
4762 * the tree
4763 */
4764 next_mz =
4765 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07004766 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004767 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07004768 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07004769 break;
4770 } while (1);
4771 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004772 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4773 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07004774 /*
4775 * One school of thought says that we should not add
4776 * back the node to the tree if reclaim returns 0.
4777 * But our reclaim could return 0, simply because due
4778 * to priority we are exposing a smaller subset of
4779 * memory to reclaim from. Consider this as a longer
4780 * term TODO.
4781 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07004782 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004783 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07004784 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004785 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07004786 loop++;
4787 /*
4788 * Could not reclaim anything and there are no more
4789 * mem cgroups to try or we seem to be looping without
4790 * reclaiming anything.
4791 */
4792 if (!nr_reclaimed &&
4793 (next_mz == NULL ||
4794 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4795 break;
4796 } while (!nr_reclaimed);
4797 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004798 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07004799 return nr_reclaimed;
4800}
4801
Michal Hocko2ef37d32012-10-26 13:37:30 +02004802/**
4803 * mem_cgroup_force_empty_list - clears LRU of a group
4804 * @memcg: group to clear
4805 * @node: NUMA node
4806 * @zid: zone id
4807 * @lru: lru to to clear
4808 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004809 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004810 * reclaim the pages page themselves - pages are moved to the parent (or root)
4811 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004812 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004813static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004814 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004815{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004816 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004817 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004818 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004819 struct page *busy;
4820 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004821
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004822 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004823 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4824 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004825
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004826 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004827 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004828 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004829 struct page *page;
4830
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004831 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004832 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004833 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004834 break;
4835 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004836 page = list_entry(list->prev, struct page, lru);
4837 if (busy == page) {
4838 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004839 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004840 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004841 continue;
4842 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004843 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004844
Johannes Weiner925b7672012-01-12 17:18:15 -08004845 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004846
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004847 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004848 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004849 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004850 cond_resched();
4851 } else
4852 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004853 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004854}
4855
4856/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004857 * make mem_cgroup's charge to be 0 if there is no task by moving
4858 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004859 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004860 *
4861 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004862 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004863static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004864{
Michal Hockoc26251f2012-10-26 13:37:28 +02004865 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004866 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004867
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004868 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004869 /* This is for making all *used* pages to be on LRU. */
4870 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004871 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004872 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004873 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004874 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004875 enum lru_list lru;
4876 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004877 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004878 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004879 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004880 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004881 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004882 mem_cgroup_end_move(memcg);
4883 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004884 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004885
Michal Hocko2ef37d32012-10-26 13:37:30 +02004886 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004887 * Kernel memory may not necessarily be trackable to a specific
4888 * process. So they are not migrated, and therefore we can't
4889 * expect their value to drop to 0 here.
4890 * Having res filled up with kmem only is enough.
4891 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004892 * This is a safety check because mem_cgroup_force_empty_list
4893 * could have raced with mem_cgroup_replace_page_cache callers
4894 * so the lru seemed empty but the page could have been added
4895 * right after the check. RES_USAGE should be safe as we always
4896 * charge before adding to the LRU.
4897 */
Glauber Costabea207c2012-12-18 14:22:11 -08004898 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4899 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4900 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004901}
4902
4903/*
Glauber Costab5f99b52013-02-22 16:34:53 -08004904 * This mainly exists for tests during the setting of set of use_hierarchy.
4905 * Since this is the very setting we are changing, the current hierarchy value
4906 * is meaningless
4907 */
4908static inline bool __memcg_has_children(struct mem_cgroup *memcg)
4909{
4910 struct cgroup *pos;
4911
4912 /* bounce at first found */
4913 cgroup_for_each_child(pos, memcg->css.cgroup)
4914 return true;
4915 return false;
4916}
4917
4918/*
Glauber Costa09998212013-02-22 16:34:55 -08004919 * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed
4920 * to be already dead (as in mem_cgroup_force_empty, for instance). This is
Glauber Costab5f99b52013-02-22 16:34:53 -08004921 * from mem_cgroup_count_children(), in the sense that we don't really care how
4922 * many children we have; we only need to know if we have any. It also counts
4923 * any memcg without hierarchy as infertile.
4924 */
4925static inline bool memcg_has_children(struct mem_cgroup *memcg)
4926{
4927 return memcg->use_hierarchy && __memcg_has_children(memcg);
4928}
4929
4930/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004931 * Reclaims as many pages from the given memcg as possible and moves
4932 * the rest to the parent.
4933 *
4934 * Caller is responsible for holding css reference for memcg.
4935 */
4936static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4937{
4938 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4939 struct cgroup *cgrp = memcg->css.cgroup;
4940
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004941 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004942 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4943 return -EBUSY;
4944
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004945 /* we call try-to-free pages for make this cgroup empty */
4946 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004947 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004948 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004949 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004950
Michal Hockoc26251f2012-10-26 13:37:28 +02004951 if (signal_pending(current))
4952 return -EINTR;
4953
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004954 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07004955 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004956 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004957 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004958 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02004959 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004960 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004961
4962 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004963 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02004964 mem_cgroup_reparent_charges(memcg);
4965
4966 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004967}
4968
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07004969static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004970{
Michal Hockoc26251f2012-10-26 13:37:28 +02004971 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4972 int ret;
4973
Michal Hockod8423012012-10-26 13:37:29 +02004974 if (mem_cgroup_is_root(memcg))
4975 return -EINVAL;
Michal Hockoc26251f2012-10-26 13:37:28 +02004976 css_get(&memcg->css);
4977 ret = mem_cgroup_force_empty(memcg);
4978 css_put(&memcg->css);
4979
4980 return ret;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004981}
4982
4983
Balbir Singh18f59ea2009-01-07 18:08:07 -08004984static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
4985{
4986 return mem_cgroup_from_cont(cont)->use_hierarchy;
4987}
4988
4989static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
4990 u64 val)
4991{
4992 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004993 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004994 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004995 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004996
4997 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004998 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004999
Glauber Costa09998212013-02-22 16:34:55 -08005000 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07005001
5002 if (memcg->use_hierarchy == val)
5003 goto out;
5004
Balbir Singh18f59ea2009-01-07 18:08:07 -08005005 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005006 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08005007 * in the child subtrees. If it is unset, then the change can
5008 * occur, provided the current cgroup has no children.
5009 *
5010 * For the root cgroup, parent_mem is NULL, we allow value to be
5011 * set if there are no children.
5012 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005013 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08005014 (val == 1 || val == 0)) {
Glauber Costab5f99b52013-02-22 16:34:53 -08005015 if (!__memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005016 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005017 else
5018 retval = -EBUSY;
5019 } else
5020 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005021
5022out:
Glauber Costa09998212013-02-22 16:34:55 -08005023 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005024
5025 return retval;
5026}
5027
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005028
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005029static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005030 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005031{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005032 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005033 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005034
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005035 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005036 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005037 val += mem_cgroup_read_stat(iter, idx);
5038
5039 if (val < 0) /* race ? */
5040 val = 0;
5041 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005042}
5043
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005044static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005045{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005046 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005047
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005048 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005049 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005050 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005051 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005052 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005053 }
5054
David Rientjesb070e652013-05-07 16:18:09 -07005055 /*
5056 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5057 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5058 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005059 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5060 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005061
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005062 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005063 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005064
5065 return val << PAGE_SHIFT;
5066}
5067
Tejun Heoaf36f902012-04-01 12:09:55 -07005068static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
5069 struct file *file, char __user *buf,
5070 size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005071{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Tejun Heoaf36f902012-04-01 12:09:55 -07005073 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005074 u64 val;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005075 int name, len;
5076 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005077
5078 type = MEMFILE_TYPE(cft->private);
5079 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005080
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005081 switch (type) {
5082 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005083 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005084 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005085 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005086 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005087 break;
5088 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005089 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005090 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005091 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005092 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005093 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005094 case _KMEM:
5095 val = res_counter_read_u64(&memcg->kmem, name);
5096 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005097 default:
5098 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005099 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005100
5101 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
5102 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005103}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005104
5105static int memcg_update_kmem_limit(struct cgroup *cont, u64 val)
5106{
5107 int ret = -EINVAL;
5108#ifdef CONFIG_MEMCG_KMEM
5109 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5110 /*
5111 * For simplicity, we won't allow this to be disabled. It also can't
5112 * be changed if the cgroup has children already, or if tasks had
5113 * already joined.
5114 *
5115 * If tasks join before we set the limit, a person looking at
5116 * kmem.usage_in_bytes will have no way to determine when it took
5117 * place, which makes the value quite meaningless.
5118 *
5119 * After it first became limited, changes in the value of the limit are
5120 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005121 */
Glauber Costa09998212013-02-22 16:34:55 -08005122 mutex_lock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005123 mutex_lock(&set_limit_mutex);
5124 if (!memcg->kmem_account_flags && val != RESOURCE_MAX) {
Glauber Costab5f99b52013-02-22 16:34:53 -08005125 if (cgroup_task_count(cont) || memcg_has_children(memcg)) {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005126 ret = -EBUSY;
5127 goto out;
5128 }
5129 ret = res_counter_set_limit(&memcg->kmem, val);
5130 VM_BUG_ON(ret);
5131
Glauber Costa55007d82012-12-18 14:22:38 -08005132 ret = memcg_update_cache_sizes(memcg);
5133 if (ret) {
5134 res_counter_set_limit(&memcg->kmem, RESOURCE_MAX);
5135 goto out;
5136 }
Glauber Costa692e89a2013-02-22 16:34:56 -08005137 static_key_slow_inc(&memcg_kmem_enabled_key);
5138 /*
5139 * setting the active bit after the inc will guarantee no one
5140 * starts accounting before all call sites are patched
5141 */
5142 memcg_kmem_set_active(memcg);
5143
Glauber Costa7de37682012-12-18 14:22:07 -08005144 /*
5145 * kmem charges can outlive the cgroup. In the case of slab
5146 * pages, for instance, a page contain objects from various
5147 * processes, so it is unfeasible to migrate them away. We
5148 * need to reference count the memcg because of that.
5149 */
5150 mem_cgroup_get(memcg);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005151 } else
5152 ret = res_counter_set_limit(&memcg->kmem, val);
5153out:
5154 mutex_unlock(&set_limit_mutex);
Glauber Costa09998212013-02-22 16:34:55 -08005155 mutex_unlock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005156#endif
5157 return ret;
5158}
5159
Hugh Dickins6d0439902013-02-22 16:35:50 -08005160#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -08005161static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005162{
Glauber Costa55007d82012-12-18 14:22:38 -08005163 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005164 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5165 if (!parent)
Glauber Costa55007d82012-12-18 14:22:38 -08005166 goto out;
5167
Glauber Costa510fc4e2012-12-18 14:21:47 -08005168 memcg->kmem_account_flags = parent->kmem_account_flags;
Glauber Costaa8964b92012-12-18 14:22:09 -08005169 /*
5170 * When that happen, we need to disable the static branch only on those
5171 * memcgs that enabled it. To achieve this, we would be forced to
5172 * complicate the code by keeping track of which memcgs were the ones
5173 * that actually enabled limits, and which ones got it from its
5174 * parents.
5175 *
5176 * It is a lot simpler just to do static_key_slow_inc() on every child
5177 * that is accounted.
5178 */
Glauber Costa55007d82012-12-18 14:22:38 -08005179 if (!memcg_kmem_is_active(memcg))
5180 goto out;
5181
5182 /*
5183 * destroy(), called if we fail, will issue static_key_slow_inc() and
5184 * mem_cgroup_put() if kmem is enabled. We have to either call them
5185 * unconditionally, or clear the KMEM_ACTIVE flag. I personally find
5186 * this more consistent, since it always leads to the same destroy path
5187 */
5188 mem_cgroup_get(memcg);
5189 static_key_slow_inc(&memcg_kmem_enabled_key);
5190
5191 mutex_lock(&set_limit_mutex);
5192 ret = memcg_update_cache_sizes(memcg);
5193 mutex_unlock(&set_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005194out:
5195 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005196}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005197#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005198
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005199/*
5200 * The user of this function is...
5201 * RES_LIMIT.
5202 */
Paul Menage856c13a2008-07-25 01:47:04 -07005203static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
5204 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005205{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005206 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005207 enum res_type type;
5208 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005209 unsigned long long val;
5210 int ret;
5211
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005212 type = MEMFILE_TYPE(cft->private);
5213 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005214
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005215 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005216 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005217 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5218 ret = -EINVAL;
5219 break;
5220 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005221 /* This function does all necessary parse...reuse it */
5222 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005223 if (ret)
5224 break;
5225 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005226 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005227 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005228 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005229 else if (type == _KMEM)
5230 ret = memcg_update_kmem_limit(cont, val);
5231 else
5232 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005233 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005234 case RES_SOFT_LIMIT:
5235 ret = res_counter_memparse_write_strategy(buffer, &val);
5236 if (ret)
5237 break;
5238 /*
5239 * For memsw, soft limits are hard to implement in terms
5240 * of semantics, for now, we support soft limits for
5241 * control without swap
5242 */
5243 if (type == _MEM)
5244 ret = res_counter_set_soft_limit(&memcg->res, val);
5245 else
5246 ret = -EINVAL;
5247 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005248 default:
5249 ret = -EINVAL; /* should be BUG() ? */
5250 break;
5251 }
5252 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005253}
5254
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005255static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5256 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5257{
5258 struct cgroup *cgroup;
5259 unsigned long long min_limit, min_memsw_limit, tmp;
5260
5261 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5262 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5263 cgroup = memcg->css.cgroup;
5264 if (!memcg->use_hierarchy)
5265 goto out;
5266
5267 while (cgroup->parent) {
5268 cgroup = cgroup->parent;
5269 memcg = mem_cgroup_from_cont(cgroup);
5270 if (!memcg->use_hierarchy)
5271 break;
5272 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5273 min_limit = min(min_limit, tmp);
5274 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5275 min_memsw_limit = min(min_memsw_limit, tmp);
5276 }
5277out:
5278 *mem_limit = min_limit;
5279 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005280}
5281
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005282static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005283{
Tejun Heoaf36f902012-04-01 12:09:55 -07005284 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005285 int name;
5286 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005287
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005288 type = MEMFILE_TYPE(event);
5289 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005290
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005291 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005292 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005293 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005294 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005295 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005296 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005297 else if (type == _KMEM)
5298 res_counter_reset_max(&memcg->kmem);
5299 else
5300 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005301 break;
5302 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005303 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005304 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005305 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005306 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005307 else if (type == _KMEM)
5308 res_counter_reset_failcnt(&memcg->kmem);
5309 else
5310 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005311 break;
5312 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005313
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005314 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005315}
5316
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005317static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
5318 struct cftype *cft)
5319{
5320 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
5321}
5322
Daisuke Nishimura02491442010-03-10 15:22:17 -08005323#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005324static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
5325 struct cftype *cft, u64 val)
5326{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005327 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005328
5329 if (val >= (1 << NR_MOVE_TYPE))
5330 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005331
Glauber Costaee5e8472013-02-22 16:34:50 -08005332 /*
5333 * No kind of locking is needed in here, because ->can_attach() will
5334 * check this value once in the beginning of the process, and then carry
5335 * on with stale data. This means that changes to this value will only
5336 * affect task migrations starting after the change.
5337 */
5338 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005339 return 0;
5340}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005341#else
5342static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
5343 struct cftype *cft, u64 val)
5344{
5345 return -ENOSYS;
5346}
5347#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005348
Ying Han406eb0c2011-05-26 16:25:37 -07005349#ifdef CONFIG_NUMA
Wanpeng Liab215882012-07-31 16:43:09 -07005350static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weinerfada52c2012-05-29 15:07:06 -07005351 struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07005352{
5353 int nid;
5354 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
5355 unsigned long node_nr;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005356 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07005357
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005358 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005359 seq_printf(m, "total=%lu", total_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005360 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005361 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005362 seq_printf(m, " N%d=%lu", nid, node_nr);
5363 }
5364 seq_putc(m, '\n');
5365
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005366 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005367 seq_printf(m, "file=%lu", file_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005368 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005369 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005370 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005371 seq_printf(m, " N%d=%lu", nid, node_nr);
5372 }
5373 seq_putc(m, '\n');
5374
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005375 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005376 seq_printf(m, "anon=%lu", anon_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005377 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005378 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005379 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005380 seq_printf(m, " N%d=%lu", nid, node_nr);
5381 }
5382 seq_putc(m, '\n');
5383
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005384 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005385 seq_printf(m, "unevictable=%lu", unevictable_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005386 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005387 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005388 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005389 seq_printf(m, " N%d=%lu", nid, node_nr);
5390 }
5391 seq_putc(m, '\n');
5392 return 0;
5393}
5394#endif /* CONFIG_NUMA */
5395
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005396static inline void mem_cgroup_lru_names_not_uptodate(void)
5397{
5398 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5399}
5400
Wanpeng Liab215882012-07-31 16:43:09 -07005401static int memcg_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005402 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005403{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005404 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005405 struct mem_cgroup *mi;
5406 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005407
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005408 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005409 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005410 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005411 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5412 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005413 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005414
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005415 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5416 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5417 mem_cgroup_read_events(memcg, i));
5418
5419 for (i = 0; i < NR_LRU_LISTS; i++)
5420 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5421 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5422
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005423 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005424 {
5425 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005426 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005427 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005428 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005429 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5430 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005431 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005432
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005433 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5434 long long val = 0;
5435
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005436 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005437 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005438 for_each_mem_cgroup_tree(mi, memcg)
5439 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5440 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5441 }
5442
5443 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5444 unsigned long long val = 0;
5445
5446 for_each_mem_cgroup_tree(mi, memcg)
5447 val += mem_cgroup_read_events(mi, i);
5448 seq_printf(m, "total_%s %llu\n",
5449 mem_cgroup_events_names[i], val);
5450 }
5451
5452 for (i = 0; i < NR_LRU_LISTS; i++) {
5453 unsigned long long val = 0;
5454
5455 for_each_mem_cgroup_tree(mi, memcg)
5456 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5457 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005458 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005459
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005460#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005461 {
5462 int nid, zid;
5463 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005464 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005465 unsigned long recent_rotated[2] = {0, 0};
5466 unsigned long recent_scanned[2] = {0, 0};
5467
5468 for_each_online_node(nid)
5469 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005470 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005471 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005472
Hugh Dickins89abfab2012-05-29 15:06:53 -07005473 recent_rotated[0] += rstat->recent_rotated[0];
5474 recent_rotated[1] += rstat->recent_rotated[1];
5475 recent_scanned[0] += rstat->recent_scanned[0];
5476 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005477 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005478 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5479 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5480 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5481 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005482 }
5483#endif
5484
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005485 return 0;
5486}
5487
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005488static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
5489{
5490 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5491
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005492 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005493}
5494
5495static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
5496 u64 val)
5497{
5498 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5499 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08005500
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005501 if (val > 100)
5502 return -EINVAL;
5503
5504 if (cgrp->parent == NULL)
5505 return -EINVAL;
5506
5507 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08005508
Glauber Costa09998212013-02-22 16:34:55 -08005509 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005510
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005511 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005512 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005513 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005514 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005515 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005516
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005517 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005518
Glauber Costa09998212013-02-22 16:34:55 -08005519 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005520
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005521 return 0;
5522}
5523
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005524static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5525{
5526 struct mem_cgroup_threshold_ary *t;
5527 u64 usage;
5528 int i;
5529
5530 rcu_read_lock();
5531 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005532 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005533 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005534 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005535
5536 if (!t)
5537 goto unlock;
5538
5539 usage = mem_cgroup_usage(memcg, swap);
5540
5541 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005542 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005543 * If it's not true, a threshold was crossed after last
5544 * call of __mem_cgroup_threshold().
5545 */
Phil Carmody5407a562010-05-26 14:42:42 -07005546 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005547
5548 /*
5549 * Iterate backward over array of thresholds starting from
5550 * current_threshold and check if a threshold is crossed.
5551 * If none of thresholds below usage is crossed, we read
5552 * only one element of the array here.
5553 */
5554 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5555 eventfd_signal(t->entries[i].eventfd, 1);
5556
5557 /* i = current_threshold + 1 */
5558 i++;
5559
5560 /*
5561 * Iterate forward over array of thresholds starting from
5562 * current_threshold+1 and check if a threshold is crossed.
5563 * If none of thresholds above usage is crossed, we read
5564 * only one element of the array here.
5565 */
5566 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5567 eventfd_signal(t->entries[i].eventfd, 1);
5568
5569 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005570 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005571unlock:
5572 rcu_read_unlock();
5573}
5574
5575static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5576{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005577 while (memcg) {
5578 __mem_cgroup_threshold(memcg, false);
5579 if (do_swap_account)
5580 __mem_cgroup_threshold(memcg, true);
5581
5582 memcg = parent_mem_cgroup(memcg);
5583 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005584}
5585
5586static int compare_thresholds(const void *a, const void *b)
5587{
5588 const struct mem_cgroup_threshold *_a = a;
5589 const struct mem_cgroup_threshold *_b = b;
5590
5591 return _a->threshold - _b->threshold;
5592}
5593
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005594static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005595{
5596 struct mem_cgroup_eventfd_list *ev;
5597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005598 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005599 eventfd_signal(ev->eventfd, 1);
5600 return 0;
5601}
5602
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005603static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005604{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005605 struct mem_cgroup *iter;
5606
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005607 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005608 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005609}
5610
5611static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
5612 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005613{
5614 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005615 struct mem_cgroup_thresholds *thresholds;
5616 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005617 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005618 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005619 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005620
5621 ret = res_counter_memparse_write_strategy(args, &threshold);
5622 if (ret)
5623 return ret;
5624
5625 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005626
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005627 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005628 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005629 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005630 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005631 else
5632 BUG();
5633
5634 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5635
5636 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005637 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005638 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5639
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005640 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005641
5642 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005643 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005644 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005645 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005646 ret = -ENOMEM;
5647 goto unlock;
5648 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005649 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005650
5651 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005652 if (thresholds->primary) {
5653 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005654 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005655 }
5656
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005657 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005658 new->entries[size - 1].eventfd = eventfd;
5659 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005660
5661 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005662 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005663 compare_thresholds, NULL);
5664
5665 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005666 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005667 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005668 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005669 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005670 * new->current_threshold will not be used until
5671 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005672 * it here.
5673 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005674 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005675 } else
5676 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005677 }
5678
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005679 /* Free old spare buffer and save old primary buffer as spare */
5680 kfree(thresholds->spare);
5681 thresholds->spare = thresholds->primary;
5682
5683 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005684
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005685 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005686 synchronize_rcu();
5687
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005688unlock:
5689 mutex_unlock(&memcg->thresholds_lock);
5690
5691 return ret;
5692}
5693
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005694static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005695 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005696{
5697 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005698 struct mem_cgroup_thresholds *thresholds;
5699 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005700 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005701 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005702 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005703
5704 mutex_lock(&memcg->thresholds_lock);
5705 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005706 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005707 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005708 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005709 else
5710 BUG();
5711
Anton Vorontsov371528c2012-02-24 05:14:46 +04005712 if (!thresholds->primary)
5713 goto unlock;
5714
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005715 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5716
5717 /* Check if a threshold crossed before removing */
5718 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5719
5720 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005721 size = 0;
5722 for (i = 0; i < thresholds->primary->size; i++) {
5723 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005724 size++;
5725 }
5726
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005727 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005728
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005729 /* Set thresholds array to NULL if we don't have thresholds */
5730 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005731 kfree(new);
5732 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005733 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005734 }
5735
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005736 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005737
5738 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005739 new->current_threshold = -1;
5740 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5741 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005742 continue;
5743
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005744 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005745 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005746 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005747 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005748 * until rcu_assign_pointer(), so it's safe to increment
5749 * it here.
5750 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005751 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005752 }
5753 j++;
5754 }
5755
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005756swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005757 /* Swap primary and spare array */
5758 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005759 /* If all events are unregistered, free the spare array */
5760 if (!new) {
5761 kfree(thresholds->spare);
5762 thresholds->spare = NULL;
5763 }
5764
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005765 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005766
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005767 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005768 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005769unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005770 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005771}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005772
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005773static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
5774 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5775{
5776 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5777 struct mem_cgroup_eventfd_list *event;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005778 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005779
5780 BUG_ON(type != _OOM_TYPE);
5781 event = kmalloc(sizeof(*event), GFP_KERNEL);
5782 if (!event)
5783 return -ENOMEM;
5784
Michal Hocko1af8efe2011-07-26 16:08:24 -07005785 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005786
5787 event->eventfd = eventfd;
5788 list_add(&event->list, &memcg->oom_notify);
5789
5790 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005791 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005792 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005793 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005794
5795 return 0;
5796}
5797
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005798static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005799 struct cftype *cft, struct eventfd_ctx *eventfd)
5800{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005801 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005802 struct mem_cgroup_eventfd_list *ev, *tmp;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005803 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005804
5805 BUG_ON(type != _OOM_TYPE);
5806
Michal Hocko1af8efe2011-07-26 16:08:24 -07005807 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005808
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005809 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005810 if (ev->eventfd == eventfd) {
5811 list_del(&ev->list);
5812 kfree(ev);
5813 }
5814 }
5815
Michal Hocko1af8efe2011-07-26 16:08:24 -07005816 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005817}
5818
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005819static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
5820 struct cftype *cft, struct cgroup_map_cb *cb)
5821{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005822 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005823
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005824 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005825
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005826 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005827 cb->fill(cb, "under_oom", 1);
5828 else
5829 cb->fill(cb, "under_oom", 0);
5830 return 0;
5831}
5832
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005833static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
5834 struct cftype *cft, u64 val)
5835{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005836 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005837 struct mem_cgroup *parent;
5838
5839 /* cannot set to root cgroup and only 0 and 1 are allowed */
5840 if (!cgrp->parent || !((val == 0) || (val == 1)))
5841 return -EINVAL;
5842
5843 parent = mem_cgroup_from_cont(cgrp->parent);
5844
Glauber Costa09998212013-02-22 16:34:55 -08005845 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005846 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005847 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005848 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005849 return -EINVAL;
5850 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005851 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005852 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005853 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005854 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005855 return 0;
5856}
5857
Andrew Mortonc255a452012-07-31 16:43:02 -07005858#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005859static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005860{
Glauber Costa55007d82012-12-18 14:22:38 -08005861 int ret;
5862
Glauber Costa2633d7a2012-12-18 14:22:34 -08005863 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005864 ret = memcg_propagate_kmem(memcg);
5865 if (ret)
5866 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005867
Glauber Costa1d62e432012-04-09 19:36:33 -03005868 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005869}
Glauber Costae5671df2011-12-11 21:47:01 +00005870
Glauber Costa1d62e432012-04-09 19:36:33 -03005871static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005872{
Glauber Costa1d62e432012-04-09 19:36:33 -03005873 mem_cgroup_sockets_destroy(memcg);
Glauber Costa7de37682012-12-18 14:22:07 -08005874
5875 memcg_kmem_mark_dead(memcg);
5876
5877 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5878 return;
5879
5880 /*
5881 * Charges already down to 0, undo mem_cgroup_get() done in the charge
5882 * path here, being careful not to race with memcg_uncharge_kmem: it is
5883 * possible that the charges went down to 0 between mark_dead and the
5884 * res_counter read, so in that case, we don't need the put
5885 */
5886 if (memcg_kmem_test_and_clear_dead(memcg))
5887 mem_cgroup_put(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005888}
Glauber Costae5671df2011-12-11 21:47:01 +00005889#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005890static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005891{
5892 return 0;
5893}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005894
Glauber Costa1d62e432012-04-09 19:36:33 -03005895static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005896{
5897}
Glauber Costae5671df2011-12-11 21:47:01 +00005898#endif
5899
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005900static struct cftype mem_cgroup_files[] = {
5901 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005902 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005903 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07005904 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005905 .register_event = mem_cgroup_usage_register_event,
5906 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005907 },
5908 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005909 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005910 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005911 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005912 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005913 },
5914 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005915 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005916 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07005917 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005918 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005919 },
5920 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005921 .name = "soft_limit_in_bytes",
5922 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
5923 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005924 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07005925 },
5926 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005927 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005928 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005929 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005930 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005931 },
Balbir Singh8697d332008-02-07 00:13:59 -08005932 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005933 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005934 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005935 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005936 {
5937 .name = "force_empty",
5938 .trigger = mem_cgroup_force_empty_write,
5939 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005940 {
5941 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07005942 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08005943 .write_u64 = mem_cgroup_hierarchy_write,
5944 .read_u64 = mem_cgroup_hierarchy_read,
5945 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005946 {
5947 .name = "swappiness",
5948 .read_u64 = mem_cgroup_swappiness_read,
5949 .write_u64 = mem_cgroup_swappiness_write,
5950 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005951 {
5952 .name = "move_charge_at_immigrate",
5953 .read_u64 = mem_cgroup_move_charge_read,
5954 .write_u64 = mem_cgroup_move_charge_write,
5955 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005956 {
5957 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005958 .read_map = mem_cgroup_oom_control_read,
5959 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005960 .register_event = mem_cgroup_oom_register_event,
5961 .unregister_event = mem_cgroup_oom_unregister_event,
5962 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
5963 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005964 {
5965 .name = "pressure_level",
5966 .register_event = vmpressure_register_event,
5967 .unregister_event = vmpressure_unregister_event,
5968 },
Ying Han406eb0c2011-05-26 16:25:37 -07005969#ifdef CONFIG_NUMA
5970 {
5971 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005972 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07005973 },
5974#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005975#ifdef CONFIG_MEMCG_KMEM
5976 {
5977 .name = "kmem.limit_in_bytes",
5978 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
5979 .write_string = mem_cgroup_write,
5980 .read = mem_cgroup_read,
5981 },
5982 {
5983 .name = "kmem.usage_in_bytes",
5984 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
5985 .read = mem_cgroup_read,
5986 },
5987 {
5988 .name = "kmem.failcnt",
5989 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
5990 .trigger = mem_cgroup_reset,
5991 .read = mem_cgroup_read,
5992 },
5993 {
5994 .name = "kmem.max_usage_in_bytes",
5995 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
5996 .trigger = mem_cgroup_reset,
5997 .read = mem_cgroup_read,
5998 },
Glauber Costa749c5412012-12-18 14:23:01 -08005999#ifdef CONFIG_SLABINFO
6000 {
6001 .name = "kmem.slabinfo",
6002 .read_seq_string = mem_cgroup_slabinfo_read,
6003 },
6004#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006005#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006006 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006007};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006008
Michal Hocko2d110852013-02-22 16:34:43 -08006009#ifdef CONFIG_MEMCG_SWAP
6010static struct cftype memsw_cgroup_files[] = {
6011 {
6012 .name = "memsw.usage_in_bytes",
6013 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6014 .read = mem_cgroup_read,
6015 .register_event = mem_cgroup_usage_register_event,
6016 .unregister_event = mem_cgroup_usage_unregister_event,
6017 },
6018 {
6019 .name = "memsw.max_usage_in_bytes",
6020 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6021 .trigger = mem_cgroup_reset,
6022 .read = mem_cgroup_read,
6023 },
6024 {
6025 .name = "memsw.limit_in_bytes",
6026 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6027 .write_string = mem_cgroup_write,
6028 .read = mem_cgroup_read,
6029 },
6030 {
6031 .name = "memsw.failcnt",
6032 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6033 .trigger = mem_cgroup_reset,
6034 .read = mem_cgroup_read,
6035 },
6036 { }, /* terminate */
6037};
6038#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006039static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006040{
6041 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006042 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006043 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006044 /*
6045 * This routine is called against possible nodes.
6046 * But it's BUG to call kmalloc() against offline node.
6047 *
6048 * TODO: this routine can waste much memory for nodes which will
6049 * never be onlined. It's better to use memory hotplug callback
6050 * function.
6051 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006052 if (!node_state(node, N_NORMAL_MEMORY))
6053 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006054 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006055 if (!pn)
6056 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006057
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006058 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6059 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006060 lruvec_init(&mz->lruvec);
Balbir Singhf64c3f52009-09-23 15:56:37 -07006061 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07006062 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006063 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006064 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07006065 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006066 return 0;
6067}
6068
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006069static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006070{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006071 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006072}
6073
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006074static struct mem_cgroup *mem_cgroup_alloc(void)
6075{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006076 struct mem_cgroup *memcg;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006077 size_t size = memcg_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006078
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006079 /* Can be very big if nr_node_ids is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08006080 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006081 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006082 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006083 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006084
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006085 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006086 return NULL;
6087
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006088 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6089 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006090 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006091 spin_lock_init(&memcg->pcp_counter_lock);
6092 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006093
6094out_free:
6095 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006096 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006097 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006098 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006099 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006100}
6101
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006102/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006103 * At destroying mem_cgroup, references from swap_cgroup can remain.
6104 * (scanning all at force_empty is too costly...)
6105 *
6106 * Instead of clearing all references at force_empty, we remember
6107 * the number of reference from swap_cgroup and free mem_cgroup when
6108 * it goes down to 0.
6109 *
6110 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006111 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006112
6113static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006114{
Glauber Costac8b2a362012-12-18 14:22:13 -08006115 int node;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006116 size_t size = memcg_size();
Hugh Dickins59927fb2012-03-15 15:17:07 -07006117
Glauber Costac8b2a362012-12-18 14:22:13 -08006118 mem_cgroup_remove_from_trees(memcg);
6119 free_css_id(&mem_cgroup_subsys, &memcg->css);
6120
6121 for_each_node(node)
6122 free_mem_cgroup_per_zone_info(memcg, node);
6123
6124 free_percpu(memcg->stat);
6125
Glauber Costa3f134612012-05-29 15:07:11 -07006126 /*
6127 * We need to make sure that (at least for now), the jump label
6128 * destruction code runs outside of the cgroup lock. This is because
6129 * get_online_cpus(), which is called from the static_branch update,
6130 * can't be called inside the cgroup_lock. cpusets are the ones
6131 * enforcing this dependency, so if they ever change, we might as well.
6132 *
6133 * schedule_work() will guarantee this happens. Be careful if you need
6134 * to move this code around, and make sure it is outside
6135 * the cgroup_lock.
6136 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006137 disarm_static_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006138 if (size < PAGE_SIZE)
6139 kfree(memcg);
6140 else
6141 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006142}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006143
Glauber Costac8b2a362012-12-18 14:22:13 -08006144
6145/*
6146 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
6147 * but in process context. The work_freeing structure is overlaid
6148 * on the rcu_freeing structure, which itself is overlaid on memsw.
6149 */
6150static void free_work(struct work_struct *work)
6151{
6152 struct mem_cgroup *memcg;
6153
6154 memcg = container_of(work, struct mem_cgroup, work_freeing);
6155 __mem_cgroup_free(memcg);
6156}
6157
Glauber Costa3afe36b2012-05-29 15:07:10 -07006158static void free_rcu(struct rcu_head *rcu_head)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006159{
6160 struct mem_cgroup *memcg;
6161
6162 memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006163 INIT_WORK(&memcg->work_freeing, free_work);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006164 schedule_work(&memcg->work_freeing);
6165}
6166
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006167static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006168{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006169 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006170}
6171
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006172static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006173{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006174 if (atomic_sub_and_test(count, &memcg->refcnt)) {
6175 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006176 call_rcu(&memcg->rcu_freeing, free_rcu);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006177 if (parent)
6178 mem_cgroup_put(parent);
6179 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006180}
6181
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006182static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006183{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006184 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006185}
6186
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006187/*
6188 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6189 */
Glauber Costae1aab162011-12-11 21:47:03 +00006190struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006191{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006192 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006193 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006194 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006195}
Glauber Costae1aab162011-12-11 21:47:03 +00006196EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006197
Michal Hocko8787a1d2013-02-22 16:35:39 -08006198static void __init mem_cgroup_soft_limit_tree_init(void)
Balbir Singhf64c3f52009-09-23 15:56:37 -07006199{
6200 struct mem_cgroup_tree_per_node *rtpn;
6201 struct mem_cgroup_tree_per_zone *rtpz;
6202 int tmp, node, zone;
6203
Bob Liu3ed28fa2012-01-12 17:19:04 -08006204 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07006205 tmp = node;
6206 if (!node_state(node, N_NORMAL_MEMORY))
6207 tmp = -1;
6208 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
Michal Hocko8787a1d2013-02-22 16:35:39 -08006209 BUG_ON(!rtpn);
Balbir Singhf64c3f52009-09-23 15:56:37 -07006210
6211 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6212
6213 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6214 rtpz = &rtpn->rb_tree_per_zone[zone];
6215 rtpz->rb_root = RB_ROOT;
6216 spin_lock_init(&rtpz->lock);
6217 }
6218 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07006219}
6220
Li Zefan0eb253e2009-01-15 13:51:25 -08006221static struct cgroup_subsys_state * __ref
Tejun Heo92fb9742012-11-19 08:13:38 -08006222mem_cgroup_css_alloc(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006223{
Glauber Costad142e3e2013-02-22 16:34:52 -08006224 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006225 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006226 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006227
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006228 memcg = mem_cgroup_alloc();
6229 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006230 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006231
Bob Liu3ed28fa2012-01-12 17:19:04 -08006232 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006233 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006234 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006235
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006236 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006237 if (cont->parent == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006238 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006239 res_counter_init(&memcg->res, NULL);
6240 res_counter_init(&memcg->memsw, NULL);
6241 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006242 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006243
Glauber Costad142e3e2013-02-22 16:34:52 -08006244 memcg->last_scanned_node = MAX_NUMNODES;
6245 INIT_LIST_HEAD(&memcg->oom_notify);
6246 atomic_set(&memcg->refcnt, 1);
6247 memcg->move_charge_at_immigrate = 0;
6248 mutex_init(&memcg->thresholds_lock);
6249 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006250 vmpressure_init(&memcg->vmpressure);
Glauber Costad142e3e2013-02-22 16:34:52 -08006251
6252 return &memcg->css;
6253
6254free_out:
6255 __mem_cgroup_free(memcg);
6256 return ERR_PTR(error);
6257}
6258
6259static int
6260mem_cgroup_css_online(struct cgroup *cont)
6261{
6262 struct mem_cgroup *memcg, *parent;
6263 int error = 0;
6264
6265 if (!cont->parent)
6266 return 0;
6267
Glauber Costa09998212013-02-22 16:34:55 -08006268 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006269 memcg = mem_cgroup_from_cont(cont);
6270 parent = mem_cgroup_from_cont(cont->parent);
6271
6272 memcg->use_hierarchy = parent->use_hierarchy;
6273 memcg->oom_kill_disable = parent->oom_kill_disable;
6274 memcg->swappiness = mem_cgroup_swappiness(parent);
6275
6276 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006277 res_counter_init(&memcg->res, &parent->res);
6278 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006279 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006280
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006281 /*
6282 * We increment refcnt of the parent to ensure that we can
6283 * safely access it on res_counter_charge/uncharge.
6284 * This refcnt will be decremented when freeing this
6285 * mem_cgroup(see mem_cgroup_put).
6286 */
6287 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006288 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006289 res_counter_init(&memcg->res, NULL);
6290 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006291 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006292 /*
6293 * Deeper hierachy with use_hierarchy == false doesn't make
6294 * much sense so let cgroup subsystem know about this
6295 * unfortunate state in our controller.
6296 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006297 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006298 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006299 }
Glauber Costacbe128e32012-04-09 19:36:34 -03006300
6301 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
Glauber Costa09998212013-02-22 16:34:55 -08006302 mutex_unlock(&memcg_create_mutex);
Glauber Costacbe128e32012-04-09 19:36:34 -03006303 if (error) {
6304 /*
6305 * We call put now because our (and parent's) refcnts
6306 * are already in place. mem_cgroup_put() will internally
6307 * call __mem_cgroup_free, so return directly
6308 */
6309 mem_cgroup_put(memcg);
Glauber Costae4715f02013-02-22 16:34:57 -08006310 if (parent->use_hierarchy)
6311 mem_cgroup_put(parent);
Glauber Costacbe128e32012-04-09 19:36:34 -03006312 }
Glauber Costad142e3e2013-02-22 16:34:52 -08006313 return error;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006314}
6315
Michal Hocko5f578162013-04-29 15:07:17 -07006316/*
6317 * Announce all parents that a group from their hierarchy is gone.
6318 */
6319static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6320{
6321 struct mem_cgroup *parent = memcg;
6322
6323 while ((parent = parent_mem_cgroup(parent)))
6324 atomic_inc(&parent->dead_count);
6325
6326 /*
6327 * if the root memcg is not hierarchical we have to check it
6328 * explicitely.
6329 */
6330 if (!root_mem_cgroup->use_hierarchy)
6331 atomic_inc(&root_mem_cgroup->dead_count);
6332}
6333
Tejun Heo92fb9742012-11-19 08:13:38 -08006334static void mem_cgroup_css_offline(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006335{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006336 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006337
Michal Hocko5f578162013-04-29 15:07:17 -07006338 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006339 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006340 mem_cgroup_destroy_all_caches(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006341}
6342
Tejun Heo92fb9742012-11-19 08:13:38 -08006343static void mem_cgroup_css_free(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006344{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006345 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006346
Glauber Costa1d62e432012-04-09 19:36:33 -03006347 kmem_cgroup_destroy(memcg);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00006348
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006349 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006350}
6351
Daisuke Nishimura02491442010-03-10 15:22:17 -08006352#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006353/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006354#define PRECHARGE_COUNT_AT_ONCE 256
6355static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006356{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006357 int ret = 0;
6358 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006359 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006360
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006361 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006362 mc.precharge += count;
6363 /* we don't need css_get for root */
6364 return ret;
6365 }
6366 /* try to charge at once */
6367 if (count > 1) {
6368 struct res_counter *dummy;
6369 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006370 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006371 * by cgroup_lock_live_cgroup() that it is not removed and we
6372 * are still under the same cgroup_mutex. So we can postpone
6373 * css_get().
6374 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006375 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006376 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006377 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006378 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006379 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006380 goto one_by_one;
6381 }
6382 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006383 return ret;
6384 }
6385one_by_one:
6386 /* fall back to one by one charge */
6387 while (count--) {
6388 if (signal_pending(current)) {
6389 ret = -EINTR;
6390 break;
6391 }
6392 if (!batch_count--) {
6393 batch_count = PRECHARGE_COUNT_AT_ONCE;
6394 cond_resched();
6395 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006396 ret = __mem_cgroup_try_charge(NULL,
6397 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006398 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006399 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006400 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006401 mc.precharge++;
6402 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006403 return ret;
6404}
6405
6406/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006407 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006408 * @vma: the vma the pte to be checked belongs
6409 * @addr: the address corresponding to the pte to be checked
6410 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006411 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006412 *
6413 * Returns
6414 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6415 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6416 * move charge. if @target is not NULL, the page is stored in target->page
6417 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006418 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6419 * target for charge migration. if @target is not NULL, the entry is stored
6420 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006421 *
6422 * Called with pte lock held.
6423 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006424union mc_target {
6425 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006426 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006427};
6428
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006429enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006430 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006431 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006432 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006433};
6434
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006435static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6436 unsigned long addr, pte_t ptent)
6437{
6438 struct page *page = vm_normal_page(vma, addr, ptent);
6439
6440 if (!page || !page_mapped(page))
6441 return NULL;
6442 if (PageAnon(page)) {
6443 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006444 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006445 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006446 } else if (!move_file())
6447 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006448 return NULL;
6449 if (!get_page_unless_zero(page))
6450 return NULL;
6451
6452 return page;
6453}
6454
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006455#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006456static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6457 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6458{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006459 struct page *page = NULL;
6460 swp_entry_t ent = pte_to_swp_entry(ptent);
6461
6462 if (!move_anon() || non_swap_entry(ent))
6463 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006464 /*
6465 * Because lookup_swap_cache() updates some statistics counter,
6466 * we call find_get_page() with swapper_space directly.
6467 */
Shaohua Li33806f02013-02-22 16:34:37 -08006468 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006469 if (do_swap_account)
6470 entry->val = ent.val;
6471
6472 return page;
6473}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006474#else
6475static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6476 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6477{
6478 return NULL;
6479}
6480#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006481
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006482static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6483 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6484{
6485 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006486 struct address_space *mapping;
6487 pgoff_t pgoff;
6488
6489 if (!vma->vm_file) /* anonymous vma */
6490 return NULL;
6491 if (!move_file())
6492 return NULL;
6493
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006494 mapping = vma->vm_file->f_mapping;
6495 if (pte_none(ptent))
6496 pgoff = linear_page_index(vma, addr);
6497 else /* pte_file(ptent) is true */
6498 pgoff = pte_to_pgoff(ptent);
6499
6500 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006501 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006502
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006503#ifdef CONFIG_SWAP
6504 /* shmem/tmpfs may report page out on swap: account for that too. */
6505 if (radix_tree_exceptional_entry(page)) {
6506 swp_entry_t swap = radix_to_swp_entry(page);
6507 if (do_swap_account)
6508 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006509 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006510 }
6511#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006512 return page;
6513}
6514
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006515static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006516 unsigned long addr, pte_t ptent, union mc_target *target)
6517{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006518 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006519 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006520 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006521 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006522
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006523 if (pte_present(ptent))
6524 page = mc_handle_present_pte(vma, addr, ptent);
6525 else if (is_swap_pte(ptent))
6526 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006527 else if (pte_none(ptent) || pte_file(ptent))
6528 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006529
6530 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006531 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006532 if (page) {
6533 pc = lookup_page_cgroup(page);
6534 /*
6535 * Do only loose check w/o page_cgroup lock.
6536 * mem_cgroup_move_account() checks the pc is valid or not under
6537 * the lock.
6538 */
6539 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6540 ret = MC_TARGET_PAGE;
6541 if (target)
6542 target->page = page;
6543 }
6544 if (!ret || !target)
6545 put_page(page);
6546 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006547 /* There is a swap entry and a page doesn't exist or isn't charged */
6548 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08006549 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006550 ret = MC_TARGET_SWAP;
6551 if (target)
6552 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006553 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006554 return ret;
6555}
6556
Naoya Horiguchi12724852012-03-21 16:34:28 -07006557#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6558/*
6559 * We don't consider swapping or file mapped pages because THP does not
6560 * support them for now.
6561 * Caller should make sure that pmd_trans_huge(pmd) is true.
6562 */
6563static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6564 unsigned long addr, pmd_t pmd, union mc_target *target)
6565{
6566 struct page *page = NULL;
6567 struct page_cgroup *pc;
6568 enum mc_target_type ret = MC_TARGET_NONE;
6569
6570 page = pmd_page(pmd);
6571 VM_BUG_ON(!page || !PageHead(page));
6572 if (!move_anon())
6573 return ret;
6574 pc = lookup_page_cgroup(page);
6575 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6576 ret = MC_TARGET_PAGE;
6577 if (target) {
6578 get_page(page);
6579 target->page = page;
6580 }
6581 }
6582 return ret;
6583}
6584#else
6585static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6586 unsigned long addr, pmd_t pmd, union mc_target *target)
6587{
6588 return MC_TARGET_NONE;
6589}
6590#endif
6591
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006592static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6593 unsigned long addr, unsigned long end,
6594 struct mm_walk *walk)
6595{
6596 struct vm_area_struct *vma = walk->private;
6597 pte_t *pte;
6598 spinlock_t *ptl;
6599
Naoya Horiguchi12724852012-03-21 16:34:28 -07006600 if (pmd_trans_huge_lock(pmd, vma) == 1) {
6601 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6602 mc.precharge += HPAGE_PMD_NR;
6603 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006604 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006605 }
Dave Hansen03319322011-03-22 16:32:56 -07006606
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006607 if (pmd_trans_unstable(pmd))
6608 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006609 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6610 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006611 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006612 mc.precharge++; /* increment precharge temporarily */
6613 pte_unmap_unlock(pte - 1, ptl);
6614 cond_resched();
6615
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006616 return 0;
6617}
6618
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006619static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6620{
6621 unsigned long precharge;
6622 struct vm_area_struct *vma;
6623
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006624 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006625 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6626 struct mm_walk mem_cgroup_count_precharge_walk = {
6627 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6628 .mm = mm,
6629 .private = vma,
6630 };
6631 if (is_vm_hugetlb_page(vma))
6632 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006633 walk_page_range(vma->vm_start, vma->vm_end,
6634 &mem_cgroup_count_precharge_walk);
6635 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006636 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006637
6638 precharge = mc.precharge;
6639 mc.precharge = 0;
6640
6641 return precharge;
6642}
6643
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006644static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6645{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006646 unsigned long precharge = mem_cgroup_count_precharge(mm);
6647
6648 VM_BUG_ON(mc.moving_task);
6649 mc.moving_task = current;
6650 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006651}
6652
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006653/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6654static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006655{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006656 struct mem_cgroup *from = mc.from;
6657 struct mem_cgroup *to = mc.to;
6658
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006659 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006660 if (mc.precharge) {
6661 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6662 mc.precharge = 0;
6663 }
6664 /*
6665 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6666 * we must uncharge here.
6667 */
6668 if (mc.moved_charge) {
6669 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6670 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006671 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006672 /* we must fixup refcnts and charges */
6673 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006674 /* uncharge swap account from the old cgroup */
6675 if (!mem_cgroup_is_root(mc.from))
6676 res_counter_uncharge(&mc.from->memsw,
6677 PAGE_SIZE * mc.moved_swap);
6678 __mem_cgroup_put(mc.from, mc.moved_swap);
6679
6680 if (!mem_cgroup_is_root(mc.to)) {
6681 /*
6682 * we charged both to->res and to->memsw, so we should
6683 * uncharge to->res.
6684 */
6685 res_counter_uncharge(&mc.to->res,
6686 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006687 }
6688 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006689 mc.moved_swap = 0;
6690 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006691 memcg_oom_recover(from);
6692 memcg_oom_recover(to);
6693 wake_up_all(&mc.waitq);
6694}
6695
6696static void mem_cgroup_clear_mc(void)
6697{
6698 struct mem_cgroup *from = mc.from;
6699
6700 /*
6701 * we must clear moving_task before waking up waiters at the end of
6702 * task migration.
6703 */
6704 mc.moving_task = NULL;
6705 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006706 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006707 mc.from = NULL;
6708 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006709 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006710 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006711}
6712
Li Zefan761b3ef2012-01-31 13:47:36 +08006713static int mem_cgroup_can_attach(struct cgroup *cgroup,
6714 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006715{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006716 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006717 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006718 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Glauber Costaee5e8472013-02-22 16:34:50 -08006719 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006720
Glauber Costaee5e8472013-02-22 16:34:50 -08006721 /*
6722 * We are now commited to this value whatever it is. Changes in this
6723 * tunable will only affect upcoming migrations, not the current one.
6724 * So we need to save it, and keep it going.
6725 */
6726 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6727 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006728 struct mm_struct *mm;
6729 struct mem_cgroup *from = mem_cgroup_from_task(p);
6730
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006731 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006732
6733 mm = get_task_mm(p);
6734 if (!mm)
6735 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006736 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006737 if (mm->owner == p) {
6738 VM_BUG_ON(mc.from);
6739 VM_BUG_ON(mc.to);
6740 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006741 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006742 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006743 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006744 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006745 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006746 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006747 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006748 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006749 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006750
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006751 ret = mem_cgroup_precharge_mc(mm);
6752 if (ret)
6753 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006754 }
6755 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006756 }
6757 return ret;
6758}
6759
Li Zefan761b3ef2012-01-31 13:47:36 +08006760static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
6761 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006762{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006763 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006764}
6765
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006766static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6767 unsigned long addr, unsigned long end,
6768 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006769{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006770 int ret = 0;
6771 struct vm_area_struct *vma = walk->private;
6772 pte_t *pte;
6773 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006774 enum mc_target_type target_type;
6775 union mc_target target;
6776 struct page *page;
6777 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006778
Naoya Horiguchi12724852012-03-21 16:34:28 -07006779 /*
6780 * We don't take compound_lock() here but no race with splitting thp
6781 * happens because:
6782 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6783 * under splitting, which means there's no concurrent thp split,
6784 * - if another thread runs into split_huge_page() just after we
6785 * entered this if-block, the thread must wait for page table lock
6786 * to be unlocked in __split_huge_page_splitting(), where the main
6787 * part of thp split is not executed yet.
6788 */
6789 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006790 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006791 spin_unlock(&vma->vm_mm->page_table_lock);
6792 return 0;
6793 }
6794 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6795 if (target_type == MC_TARGET_PAGE) {
6796 page = target.page;
6797 if (!isolate_lru_page(page)) {
6798 pc = lookup_page_cgroup(page);
6799 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006800 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006801 mc.precharge -= HPAGE_PMD_NR;
6802 mc.moved_charge += HPAGE_PMD_NR;
6803 }
6804 putback_lru_page(page);
6805 }
6806 put_page(page);
6807 }
6808 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006809 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006810 }
6811
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006812 if (pmd_trans_unstable(pmd))
6813 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006814retry:
6815 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6816 for (; addr != end; addr += PAGE_SIZE) {
6817 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006818 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006819
6820 if (!mc.precharge)
6821 break;
6822
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006823 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006824 case MC_TARGET_PAGE:
6825 page = target.page;
6826 if (isolate_lru_page(page))
6827 goto put;
6828 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006829 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006830 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006831 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006832 /* we uncharge from mc.from later. */
6833 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006834 }
6835 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006836put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006837 put_page(page);
6838 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006839 case MC_TARGET_SWAP:
6840 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006841 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006842 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006843 /* we fixup refcnts and charges later. */
6844 mc.moved_swap++;
6845 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006846 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006847 default:
6848 break;
6849 }
6850 }
6851 pte_unmap_unlock(pte - 1, ptl);
6852 cond_resched();
6853
6854 if (addr != end) {
6855 /*
6856 * We have consumed all precharges we got in can_attach().
6857 * We try charge one by one, but don't do any additional
6858 * charges to mc.to if we have failed in charge once in attach()
6859 * phase.
6860 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006861 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006862 if (!ret)
6863 goto retry;
6864 }
6865
6866 return ret;
6867}
6868
6869static void mem_cgroup_move_charge(struct mm_struct *mm)
6870{
6871 struct vm_area_struct *vma;
6872
6873 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006874retry:
6875 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6876 /*
6877 * Someone who are holding the mmap_sem might be waiting in
6878 * waitq. So we cancel all extra charges, wake up all waiters,
6879 * and retry. Because we cancel precharges, we might not be able
6880 * to move enough charges, but moving charge is a best-effort
6881 * feature anyway, so it wouldn't be a big problem.
6882 */
6883 __mem_cgroup_clear_mc();
6884 cond_resched();
6885 goto retry;
6886 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006887 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6888 int ret;
6889 struct mm_walk mem_cgroup_move_charge_walk = {
6890 .pmd_entry = mem_cgroup_move_charge_pte_range,
6891 .mm = mm,
6892 .private = vma,
6893 };
6894 if (is_vm_hugetlb_page(vma))
6895 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006896 ret = walk_page_range(vma->vm_start, vma->vm_end,
6897 &mem_cgroup_move_charge_walk);
6898 if (ret)
6899 /*
6900 * means we have consumed all precharges and failed in
6901 * doing additional charge. Just abandon here.
6902 */
6903 break;
6904 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006905 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006906}
6907
Li Zefan761b3ef2012-01-31 13:47:36 +08006908static void mem_cgroup_move_task(struct cgroup *cont,
6909 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006910{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006911 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006912 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006913
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006914 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006915 if (mc.to)
6916 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006917 mmput(mm);
6918 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006919 if (mc.to)
6920 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006921}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006922#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08006923static int mem_cgroup_can_attach(struct cgroup *cgroup,
6924 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006925{
6926 return 0;
6927}
Li Zefan761b3ef2012-01-31 13:47:36 +08006928static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
6929 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006930{
6931}
Li Zefan761b3ef2012-01-31 13:47:36 +08006932static void mem_cgroup_move_task(struct cgroup *cont,
6933 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006934{
6935}
6936#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006937
Tejun Heof00baae2013-04-15 13:41:15 -07006938/*
6939 * Cgroup retains root cgroups across [un]mount cycles making it necessary
6940 * to verify sane_behavior flag on each mount attempt.
6941 */
6942static void mem_cgroup_bind(struct cgroup *root)
6943{
6944 /*
6945 * use_hierarchy is forced with sane_behavior. cgroup core
6946 * guarantees that @root doesn't have any children, so turning it
6947 * on for the root memcg is enough.
6948 */
6949 if (cgroup_sane_behavior(root))
6950 mem_cgroup_from_cont(root)->use_hierarchy = true;
6951}
6952
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006953struct cgroup_subsys mem_cgroup_subsys = {
6954 .name = "memory",
6955 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08006956 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006957 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006958 .css_offline = mem_cgroup_css_offline,
6959 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006960 .can_attach = mem_cgroup_can_attach,
6961 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08006962 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006963 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07006964 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006965 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006966 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006967};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006968
Andrew Mortonc255a452012-07-31 16:43:02 -07006969#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08006970static int __init enable_swap_account(char *s)
6971{
6972 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07006973 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006974 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07006975 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006976 really_do_swap_account = 0;
6977 return 1;
6978}
Michal Hockoa2c89902011-05-24 17:12:50 -07006979__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006980
Michal Hocko2d110852013-02-22 16:34:43 -08006981static void __init memsw_file_init(void)
6982{
Michal Hocko6acc8b02013-02-22 16:34:45 -08006983 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08006984}
Michal Hocko6acc8b02013-02-22 16:34:45 -08006985
6986static void __init enable_swap_cgroup(void)
6987{
6988 if (!mem_cgroup_disabled() && really_do_swap_account) {
6989 do_swap_account = 1;
6990 memsw_file_init();
6991 }
6992}
6993
Michal Hocko2d110852013-02-22 16:34:43 -08006994#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08006995static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08006996{
6997}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006998#endif
Michal Hocko2d110852013-02-22 16:34:43 -08006999
7000/*
Michal Hocko10813122013-02-22 16:35:41 -08007001 * subsys_initcall() for memory controller.
7002 *
7003 * Some parts like hotcpu_notifier() have to be initialized from this context
7004 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7005 * everything that doesn't depend on a specific mem_cgroup structure should
7006 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007007 */
7008static int __init mem_cgroup_init(void)
7009{
7010 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007011 enable_swap_cgroup();
Michal Hocko8787a1d2013-02-22 16:35:39 -08007012 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007013 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007014 return 0;
7015}
7016subsys_initcall(mem_cgroup_init);