blob: c3b674f9774f317e8188dbaaedfb7cbb8c8c4058 [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>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -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>
Tejun Heo79bd9812013-11-22 18:20:42 -050048#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080049#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080051#include <linux/seq_file.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>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070057#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050058#include <linux/file.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080059#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000060#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070061#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000062#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080063#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080064
Balbir Singh8697d332008-02-07 00:13:59 -080065#include <asm/uaccess.h>
66
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070067#include <trace/events/vmscan.h>
68
Tejun Heo073219e2014-02-08 10:36:58 -050069struct cgroup_subsys memory_cgrp_subsys __read_mostly;
70EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080071
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070072#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070073static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080074
Andrew Mortonc255a452012-07-31 16:43:02 -070075#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070076/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080077int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080078
79/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070080#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080081static int really_do_swap_account __initdata = 1;
82#else
83static int really_do_swap_account __initdata = 0;
84#endif
85
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080086#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070087#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088#endif
89
90
Johannes Weineraf7c4b02012-05-29 15:07:08 -070091static const char * const mem_cgroup_stat_names[] = {
92 "cache",
93 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070094 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070095 "mapped_file",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070096 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070097 "swap",
98};
99
Johannes Weinere9f89742011-03-23 16:42:37 -0700100enum mem_cgroup_events_index {
101 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
102 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700103 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
104 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700105 MEM_CGROUP_EVENTS_NSTATS,
106};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107
108static const char * const mem_cgroup_events_names[] = {
109 "pgpgin",
110 "pgpgout",
111 "pgfault",
112 "pgmajfault",
113};
114
Sha Zhengju58cf1882013-02-22 16:32:05 -0800115static const char * const mem_cgroup_lru_names[] = {
116 "inactive_anon",
117 "active_anon",
118 "inactive_file",
119 "active_file",
120 "unevictable",
121};
122
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700123/*
124 * Per memcg event counter is incremented at every pagein/pageout. With THP,
125 * it will be incremated by the number of pages. This counter is used for
126 * for trigger some periodic events. This is straightforward and better
127 * than using jiffies etc. to handle periodic memcg event.
128 */
129enum mem_cgroup_events_target {
130 MEM_CGROUP_TARGET_THRESH,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700131 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700132 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700133 MEM_CGROUP_NTARGETS,
134};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700135#define THRESHOLDS_EVENTS_TARGET 128
136#define SOFTLIMIT_EVENTS_TARGET 1024
137#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700138
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800139struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700140 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700141 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700142 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700143 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800144};
145
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800146struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700147 /*
148 * last scanned hierarchy member. Valid only if last_dead_count
149 * matches memcg->dead_count of the hierarchy root group.
150 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700151 struct mem_cgroup *last_visited;
Hugh Dickinsd2ab70a2014-01-23 15:53:30 -0800152 int last_dead_count;
Michal Hocko5f578162013-04-29 15:07:17 -0700153
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800154 /* scan generation, increased every round-trip */
155 unsigned int generation;
156};
157
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800158/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800159 * per-zone information in memory controller.
160 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800161struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800162 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700163 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800164
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800165 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
166
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700167 struct rb_node tree_node; /* RB tree node */
168 unsigned long long usage_in_excess;/* Set to the value by which */
169 /* the soft limit is exceeded*/
170 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700171 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700172 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800173};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800174
175struct mem_cgroup_per_node {
176 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
177};
178
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700179/*
180 * Cgroups above their limits are maintained in a RB-Tree, independent of
181 * their hierarchy representation
182 */
183
184struct mem_cgroup_tree_per_zone {
185 struct rb_root rb_root;
186 spinlock_t lock;
187};
188
189struct mem_cgroup_tree_per_node {
190 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
191};
192
193struct mem_cgroup_tree {
194 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
195};
196
197static struct mem_cgroup_tree soft_limit_tree __read_mostly;
198
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800199struct mem_cgroup_threshold {
200 struct eventfd_ctx *eventfd;
201 u64 threshold;
202};
203
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700204/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800205struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700206 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700207 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800208 /* Size of entries[] */
209 unsigned int size;
210 /* Array of thresholds */
211 struct mem_cgroup_threshold entries[0];
212};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700213
214struct mem_cgroup_thresholds {
215 /* Primary thresholds array */
216 struct mem_cgroup_threshold_ary *primary;
217 /*
218 * Spare threshold array.
219 * This is needed to make mem_cgroup_unregister_event() "never fail".
220 * It must be able to store at least primary->size - 1 entries.
221 */
222 struct mem_cgroup_threshold_ary *spare;
223};
224
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700225/* for OOM */
226struct mem_cgroup_eventfd_list {
227 struct list_head list;
228 struct eventfd_ctx *eventfd;
229};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800230
Tejun Heo79bd9812013-11-22 18:20:42 -0500231/*
232 * cgroup_event represents events which userspace want to receive.
233 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500234struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500235 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500236 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500237 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500238 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500239 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500240 * eventfd to signal userspace about the event.
241 */
242 struct eventfd_ctx *eventfd;
243 /*
244 * Each of these stored in a list by the cgroup.
245 */
246 struct list_head list;
247 /*
Tejun Heofba94802013-11-22 18:20:43 -0500248 * register_event() callback will be used to add new userspace
249 * waiter for changes related to this event. Use eventfd_signal()
250 * on eventfd to send notification to userspace.
251 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500252 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500253 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500254 /*
255 * unregister_event() callback will be called when userspace closes
256 * the eventfd or on cgroup removing. This callback must be set,
257 * if you want provide notification functionality.
258 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500259 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500260 struct eventfd_ctx *eventfd);
261 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500262 * All fields below needed to unregister event when
263 * userspace closes eventfd.
264 */
265 poll_table pt;
266 wait_queue_head_t *wqh;
267 wait_queue_t wait;
268 struct work_struct remove;
269};
270
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700271static void mem_cgroup_threshold(struct mem_cgroup *memcg);
272static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800273
Balbir Singhf64c3f52009-09-23 15:56:37 -0700274/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800275 * The memory controller data structure. The memory controller controls both
276 * page cache and RSS per cgroup. We would eventually like to provide
277 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
278 * to help the administrator determine what knobs to tune.
279 *
280 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800281 * we hit the water mark. May be even add a low water mark, such that
282 * no reclaim occurs from a cgroup at it's low water mark, this is
283 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800284 */
285struct mem_cgroup {
286 struct cgroup_subsys_state css;
287 /*
288 * the counter to account for memory usage
289 */
290 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700291
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700292 /* vmpressure notifications */
293 struct vmpressure vmpressure;
294
Li Zefan465939a2013-07-08 16:00:38 -0700295 /*
296 * the counter to account for mem+swap usage.
297 */
298 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700299
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800300 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800301 * the counter to account for kernel memory usage.
302 */
303 struct res_counter kmem;
304 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800305 * Should the accounting and control be hierarchical, per subtree?
306 */
307 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800308 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700309
310 bool oom_lock;
311 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700312 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700313
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700314 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700315 /* OOM-Killer disable */
316 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800317
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700318 /* set when res.limit == memsw.limit */
319 bool memsw_is_minimum;
320
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800321 /* protect arrays of thresholds */
322 struct mutex thresholds_lock;
323
324 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700325 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700326
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800327 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700328 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700329
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700330 /* For oom notifier event fd */
331 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700332
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800333 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800334 * Should we move charges of a task when a task is moved into this
335 * mem_cgroup ? And what type of charges should we move ?
336 */
Andrew Mortonf894ffa2013-09-12 15:13:35 -0700337 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800338 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700339 * set > 0 if pages under this cgroup are moving to other cgroup.
340 */
341 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700342 /* taken only while moving_account > 0 */
343 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700344 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800345 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800346 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700347 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700348 /*
349 * used when a cpu is offlined or other synchronizations
350 * See mem_cgroup_read_stat().
351 */
352 struct mem_cgroup_stat_cpu nocpu_base;
353 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000354
Michal Hocko5f578162013-04-29 15:07:17 -0700355 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700356#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700357 struct cg_proto tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000358#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800359#if defined(CONFIG_MEMCG_KMEM)
360 /* analogous to slab_common's slab_caches list. per-memcg */
361 struct list_head memcg_slab_caches;
362 /* Not a spinlock, we can take a lot of time walking the list */
363 struct mutex slab_caches_mutex;
364 /* Index in the kmem_cache->memcg_params->memcg_caches array */
365 int kmemcg_id;
366#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800367
368 int last_scanned_node;
369#if MAX_NUMNODES > 1
370 nodemask_t scan_nodes;
371 atomic_t numainfo_events;
372 atomic_t numainfo_updating;
373#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700374
Tejun Heofba94802013-11-22 18:20:43 -0500375 /* List of events which userspace want to receive */
376 struct list_head event_list;
377 spinlock_t event_list_lock;
378
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700379 struct mem_cgroup_per_node *nodeinfo[0];
380 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800381};
382
Glauber Costa510fc4e2012-12-18 14:21:47 -0800383/* internal only representation about the status of kmem accounting. */
384enum {
Vladimir Davydov6de64be2014-01-23 15:53:08 -0800385 KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */
Glauber Costa7de37682012-12-18 14:22:07 -0800386 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800387};
388
Glauber Costa510fc4e2012-12-18 14:21:47 -0800389#ifdef CONFIG_MEMCG_KMEM
390static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
391{
392 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
393}
Glauber Costa7de37682012-12-18 14:22:07 -0800394
395static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
396{
397 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
398}
399
400static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
401{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700402 /*
403 * Our caller must use css_get() first, because memcg_uncharge_kmem()
404 * will call css_put() if it sees the memcg is dead.
405 */
406 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800407 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
408 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
409}
410
411static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
412{
413 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
414 &memcg->kmem_account_flags);
415}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800416#endif
417
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800418/* Stuffs for move charges at task migration. */
419/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800420 * Types of charges to be moved. "move_charge_at_immitgrate" and
421 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800422 */
423enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800424 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700425 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800426 NR_MOVE_TYPE,
427};
428
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800429/* "mc" and its members are protected by cgroup_mutex */
430static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800431 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800432 struct mem_cgroup *from;
433 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800434 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800435 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800436 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800437 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800438 struct task_struct *moving_task; /* a task moving charges */
439 wait_queue_head_t waitq; /* a waitq for other context */
440} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700441 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800442 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
443};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800444
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700445static bool move_anon(void)
446{
Glauber Costaee5e8472013-02-22 16:34:50 -0800447 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700448}
449
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700450static bool move_file(void)
451{
Glauber Costaee5e8472013-02-22 16:34:50 -0800452 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700453}
454
Balbir Singh4e416952009-09-23 15:56:39 -0700455/*
456 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
457 * limit reclaim to prevent infinite loops, if they ever occur.
458 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700459#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700460#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700461
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800462enum charge_type {
463 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700464 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800465 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700466 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700467 NR_CHARGE_TYPE,
468};
469
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800470/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800471enum res_type {
472 _MEM,
473 _MEMSWAP,
474 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800475 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800476};
477
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700478#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
479#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800480#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700481/* Used for OOM nofiier */
482#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800483
Balbir Singh75822b42009-09-23 15:56:38 -0700484/*
485 * Reclaim flags for mem_cgroup_hierarchical_reclaim
486 */
487#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
488#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
489#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
490#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
491
Glauber Costa09998212013-02-22 16:34:55 -0800492/*
493 * The memcg_create_mutex will be held whenever a new cgroup is created.
494 * As a consequence, any change that needs to protect against new child cgroups
495 * appearing has to hold it as well.
496 */
497static DEFINE_MUTEX(memcg_create_mutex);
498
Wanpeng Lib2145142012-07-31 16:46:01 -0700499struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
500{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400501 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700502}
503
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700504/* Some nice accessors for the vmpressure. */
505struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
506{
507 if (!memcg)
508 memcg = root_mem_cgroup;
509 return &memcg->vmpressure;
510}
511
512struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
513{
514 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
515}
516
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700517static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
518{
519 return (memcg == root_mem_cgroup);
520}
521
Li Zefan4219b2d2013-09-23 16:56:29 +0800522/*
523 * We restrict the id in the range of [1, 65535], so it can fit into
524 * an unsigned short.
525 */
526#define MEM_CGROUP_ID_MAX USHRT_MAX
527
Li Zefan34c00c32013-09-23 16:56:01 +0800528static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
529{
530 /*
531 * The ID of the root cgroup is 0, but memcg treat 0 as an
532 * invalid ID, so we return (cgroup_id + 1).
533 */
534 return memcg->css.cgroup->id + 1;
535}
536
537static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
538{
539 struct cgroup_subsys_state *css;
540
Tejun Heo073219e2014-02-08 10:36:58 -0500541 css = css_from_id(id - 1, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800542 return mem_cgroup_from_css(css);
543}
544
Glauber Costae1aab162011-12-11 21:47:03 +0000545/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700546#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000547
Glauber Costae1aab162011-12-11 21:47:03 +0000548void sock_update_memcg(struct sock *sk)
549{
Glauber Costa376be5f2012-01-20 04:57:14 +0000550 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000551 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700552 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000553
554 BUG_ON(!sk->sk_prot->proto_cgroup);
555
Glauber Costaf3f511e2012-01-05 20:16:39 +0000556 /* Socket cloning can throw us here with sk_cgrp already
557 * filled. It won't however, necessarily happen from
558 * process context. So the test for root memcg given
559 * the current task's memcg won't help us in this case.
560 *
561 * Respecting the original socket's memcg is a better
562 * decision in this case.
563 */
564 if (sk->sk_cgrp) {
565 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700566 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000567 return;
568 }
569
Glauber Costae1aab162011-12-11 21:47:03 +0000570 rcu_read_lock();
571 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700572 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700573 if (!mem_cgroup_is_root(memcg) &&
574 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700575 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000576 }
577 rcu_read_unlock();
578 }
579}
580EXPORT_SYMBOL(sock_update_memcg);
581
582void sock_release_memcg(struct sock *sk)
583{
Glauber Costa376be5f2012-01-20 04:57:14 +0000584 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000585 struct mem_cgroup *memcg;
586 WARN_ON(!sk->sk_cgrp->memcg);
587 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700588 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000589 }
590}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000591
592struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
593{
594 if (!memcg || mem_cgroup_is_root(memcg))
595 return NULL;
596
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700597 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000598}
599EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000600
Glauber Costa3f134612012-05-29 15:07:11 -0700601static void disarm_sock_keys(struct mem_cgroup *memcg)
602{
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700603 if (!memcg_proto_activated(&memcg->tcp_mem))
Glauber Costa3f134612012-05-29 15:07:11 -0700604 return;
605 static_key_slow_dec(&memcg_socket_limit_enabled);
606}
607#else
608static void disarm_sock_keys(struct mem_cgroup *memcg)
609{
610}
611#endif
612
Glauber Costaa8964b92012-12-18 14:22:09 -0800613#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800614/*
615 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800616 * The main reason for not using cgroup id for this:
617 * this works better in sparse environments, where we have a lot of memcgs,
618 * but only a few kmem-limited. Or also, if we have, for instance, 200
619 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
620 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800621 *
622 * The current size of the caches array is stored in
623 * memcg_limited_groups_array_size. It will double each time we have to
624 * increase it.
625 */
626static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800627int memcg_limited_groups_array_size;
628
Glauber Costa55007d82012-12-18 14:22:38 -0800629/*
630 * MIN_SIZE is different than 1, because we would like to avoid going through
631 * the alloc/free process all the time. In a small machine, 4 kmem-limited
632 * cgroups is a reasonable guess. In the future, it could be a parameter or
633 * tunable, but that is strictly not necessary.
634 *
Li Zefanb8627832013-09-23 16:56:47 +0800635 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800636 * this constant directly from cgroup, but it is understandable that this is
637 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800638 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800639 * increase ours as well if it increases.
640 */
641#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800642#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800643
Glauber Costad7f25f82012-12-18 14:22:40 -0800644/*
645 * A lot of the calls to the cache allocation functions are expected to be
646 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
647 * conditional to this static branch, we'll have to allow modules that does
648 * kmem_cache_alloc and the such to see this symbol as well
649 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800650struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800651EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800652
653static void disarm_kmem_keys(struct mem_cgroup *memcg)
654{
Glauber Costa55007d82012-12-18 14:22:38 -0800655 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800656 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800657 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
658 }
Glauber Costabea207c2012-12-18 14:22:11 -0800659 /*
660 * This check can't live in kmem destruction function,
661 * since the charges will outlive the cgroup
662 */
663 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800664}
665#else
666static void disarm_kmem_keys(struct mem_cgroup *memcg)
667{
668}
669#endif /* CONFIG_MEMCG_KMEM */
670
671static void disarm_static_keys(struct mem_cgroup *memcg)
672{
673 disarm_sock_keys(memcg);
674 disarm_kmem_keys(memcg);
675}
676
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700677static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800678
Balbir Singhf64c3f52009-09-23 15:56:37 -0700679static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700681{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800682 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700683 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700684}
685
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100687{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100689}
690
Balbir Singhf64c3f52009-09-23 15:56:37 -0700691static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700692page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700693{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700694 int nid = page_to_nid(page);
695 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700696
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700697 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700698}
699
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700static struct mem_cgroup_tree_per_zone *
701soft_limit_tree_node_zone(int nid, int zid)
702{
703 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
704}
705
706static struct mem_cgroup_tree_per_zone *
707soft_limit_tree_from_page(struct page *page)
708{
709 int nid = page_to_nid(page);
710 int zid = page_zonenum(page);
711
712 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
713}
714
715static void
716__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
717 struct mem_cgroup_per_zone *mz,
718 struct mem_cgroup_tree_per_zone *mctz,
719 unsigned long long new_usage_in_excess)
720{
721 struct rb_node **p = &mctz->rb_root.rb_node;
722 struct rb_node *parent = NULL;
723 struct mem_cgroup_per_zone *mz_node;
724
725 if (mz->on_tree)
726 return;
727
728 mz->usage_in_excess = new_usage_in_excess;
729 if (!mz->usage_in_excess)
730 return;
731 while (*p) {
732 parent = *p;
733 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
734 tree_node);
735 if (mz->usage_in_excess < mz_node->usage_in_excess)
736 p = &(*p)->rb_left;
737 /*
738 * We can't avoid mem cgroups that are over their soft
739 * limit by the same amount
740 */
741 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
742 p = &(*p)->rb_right;
743 }
744 rb_link_node(&mz->tree_node, parent, p);
745 rb_insert_color(&mz->tree_node, &mctz->rb_root);
746 mz->on_tree = true;
747}
748
749static void
750__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
751 struct mem_cgroup_per_zone *mz,
752 struct mem_cgroup_tree_per_zone *mctz)
753{
754 if (!mz->on_tree)
755 return;
756 rb_erase(&mz->tree_node, &mctz->rb_root);
757 mz->on_tree = false;
758}
759
760static void
761mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
762 struct mem_cgroup_per_zone *mz,
763 struct mem_cgroup_tree_per_zone *mctz)
764{
765 spin_lock(&mctz->lock);
766 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
767 spin_unlock(&mctz->lock);
768}
769
770
771static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
772{
773 unsigned long long excess;
774 struct mem_cgroup_per_zone *mz;
775 struct mem_cgroup_tree_per_zone *mctz;
776 int nid = page_to_nid(page);
777 int zid = page_zonenum(page);
778 mctz = soft_limit_tree_from_page(page);
779
780 /*
781 * Necessary to update all ancestors when hierarchy is used.
782 * because their event counter is not touched.
783 */
784 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
785 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
786 excess = res_counter_soft_limit_excess(&memcg->res);
787 /*
788 * We have to update the tree if mz is on RB-tree or
789 * mem is over its softlimit.
790 */
791 if (excess || mz->on_tree) {
792 spin_lock(&mctz->lock);
793 /* if on-tree, remove it */
794 if (mz->on_tree)
795 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
796 /*
797 * Insert again. mz->usage_in_excess will be updated.
798 * If excess is 0, no tree ops.
799 */
800 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
801 spin_unlock(&mctz->lock);
802 }
803 }
804}
805
806static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
807{
808 int node, zone;
809 struct mem_cgroup_per_zone *mz;
810 struct mem_cgroup_tree_per_zone *mctz;
811
812 for_each_node(node) {
813 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
814 mz = mem_cgroup_zoneinfo(memcg, node, zone);
815 mctz = soft_limit_tree_node_zone(node, zone);
816 mem_cgroup_remove_exceeded(memcg, mz, mctz);
817 }
818 }
819}
820
821static struct mem_cgroup_per_zone *
822__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
823{
824 struct rb_node *rightmost = NULL;
825 struct mem_cgroup_per_zone *mz;
826
827retry:
828 mz = NULL;
829 rightmost = rb_last(&mctz->rb_root);
830 if (!rightmost)
831 goto done; /* Nothing to reclaim from */
832
833 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
834 /*
835 * Remove the node now but someone else can add it back,
836 * we will to add it back at the end of reclaim to its correct
837 * position in the tree.
838 */
839 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
840 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
841 !css_tryget(&mz->memcg->css))
842 goto retry;
843done:
844 return mz;
845}
846
847static struct mem_cgroup_per_zone *
848mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
849{
850 struct mem_cgroup_per_zone *mz;
851
852 spin_lock(&mctz->lock);
853 mz = __mem_cgroup_largest_soft_limit_node(mctz);
854 spin_unlock(&mctz->lock);
855 return mz;
856}
857
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700858/*
859 * Implementation Note: reading percpu statistics for memcg.
860 *
861 * Both of vmstat[] and percpu_counter has threshold and do periodic
862 * synchronization to implement "quick" read. There are trade-off between
863 * reading cost and precision of value. Then, we may have a chance to implement
864 * a periodic synchronizion of counter in memcg's counter.
865 *
866 * But this _read() function is used for user interface now. The user accounts
867 * memory usage by memory cgroup and he _always_ requires exact value because
868 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
869 * have to visit all online cpus and make sum. So, for now, unnecessary
870 * synchronization is not implemented. (just implemented for cpu hotplug)
871 *
872 * If there are kernel internal actions which can make use of some not-exact
873 * value, and reading all cpu value can be performance bottleneck in some
874 * common workload, threashold and synchonization as vmstat[] should be
875 * implemented.
876 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700877static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700878 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800879{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700880 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800881 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800882
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700883 get_online_cpus();
884 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700885 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700886#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700887 spin_lock(&memcg->pcp_counter_lock);
888 val += memcg->nocpu_base.count[idx];
889 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700890#endif
891 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800892 return val;
893}
894
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700895static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700896 bool charge)
897{
898 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700899 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700900}
901
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700902static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700903 enum mem_cgroup_events_index idx)
904{
905 unsigned long val = 0;
906 int cpu;
907
David Rientjes9c567512013-10-16 13:46:43 -0700908 get_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700909 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700910 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700911#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700912 spin_lock(&memcg->pcp_counter_lock);
913 val += memcg->nocpu_base.events[idx];
914 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700915#endif
David Rientjes9c567512013-10-16 13:46:43 -0700916 put_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700917 return val;
918}
919
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700920static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700921 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700922 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800923{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700924 /*
925 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
926 * counted as CACHE even if it's on ANON LRU.
927 */
928 if (anon)
929 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700930 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800931 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700932 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700933 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700934
David Rientjesb070e652013-05-07 16:18:09 -0700935 if (PageTransHuge(page))
936 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
937 nr_pages);
938
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800939 /* pagein of a big page is an event. So, ignore page size */
940 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700941 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800942 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700943 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800944 nr_pages = -nr_pages; /* for event */
945 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800946
Johannes Weiner13114712012-05-29 15:07:07 -0700947 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800948}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800949
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700950unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700951mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700952{
953 struct mem_cgroup_per_zone *mz;
954
955 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
956 return mz->lru_size[lru];
957}
958
959static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700960mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700961 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700962{
963 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700964 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700965 unsigned long ret = 0;
966
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700967 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700968
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700969 for_each_lru(lru) {
970 if (BIT(lru) & lru_mask)
971 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700972 }
973 return ret;
974}
975
976static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700977mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700978 int nid, unsigned int lru_mask)
979{
Ying Han889976d2011-05-26 16:25:33 -0700980 u64 total = 0;
981 int zid;
982
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700983 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700984 total += mem_cgroup_zone_nr_lru_pages(memcg,
985 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700986
Ying Han889976d2011-05-26 16:25:33 -0700987 return total;
988}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700989
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700990static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700991 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800992{
Ying Han889976d2011-05-26 16:25:33 -0700993 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800994 u64 total = 0;
995
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800996 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700997 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800998 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800999}
1000
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001001static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
1002 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001003{
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001004 unsigned long val, next;
1005
Johannes Weiner13114712012-05-29 15:07:07 -07001006 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -07001007 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001008 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001009 if ((long)next - (long)val < 0) {
1010 switch (target) {
1011 case MEM_CGROUP_TARGET_THRESH:
1012 next = val + THRESHOLDS_EVENTS_TARGET;
1013 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001014 case MEM_CGROUP_TARGET_SOFTLIMIT:
1015 next = val + SOFTLIMIT_EVENTS_TARGET;
1016 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001017 case MEM_CGROUP_TARGET_NUMAINFO:
1018 next = val + NUMAINFO_EVENTS_TARGET;
1019 break;
1020 default:
1021 break;
1022 }
1023 __this_cpu_write(memcg->stat->targets[target], next);
1024 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001025 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001026 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001027}
1028
1029/*
1030 * Check events in order.
1031 *
1032 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001033static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001034{
Steven Rostedt47994012011-11-02 13:38:33 -07001035 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001036 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001037 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1038 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001039 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001040 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001041
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001042 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1043 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001044#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001045 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1046 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001047#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001048 preempt_enable();
1049
1050 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001051 if (unlikely(do_softlimit))
1052 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001053#if MAX_NUMNODES > 1
1054 if (unlikely(do_numainfo))
1055 atomic_inc(&memcg->numainfo_events);
1056#endif
1057 } else
1058 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001059}
1060
Balbir Singhcf475ad2008-04-29 01:00:16 -07001061struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001062{
Balbir Singh31a78f22008-09-28 23:09:31 +01001063 /*
1064 * mm_update_next_owner() may clear mm->owner to NULL
1065 * if it races with swapoff, page migration, etc.
1066 * So this can be called with p == NULL.
1067 */
1068 if (unlikely(!p))
1069 return NULL;
1070
Tejun Heo073219e2014-02-08 10:36:58 -05001071 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001072}
1073
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001074struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001075{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001076 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001077
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001078 rcu_read_lock();
1079 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001080 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1081 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001082 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001083 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001084 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001085 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001086}
1087
Michal Hocko16248d82013-04-29 15:07:19 -07001088/*
1089 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1090 * ref. count) or NULL if the whole root's subtree has been visited.
1091 *
1092 * helper function to be used by mem_cgroup_iter
1093 */
1094static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001095 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001096{
Tejun Heo492eb212013-08-08 20:11:25 -04001097 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001098
Tejun Heobd8815a2013-08-08 20:11:27 -04001099 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001100skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001101 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001102
1103 /*
1104 * Even if we found a group we have to make sure it is
1105 * alive. css && !memcg means that the groups should be
1106 * skipped and we should continue the tree walk.
1107 * last_visited css is safe to use because it is
1108 * protected by css_get and the tree walk is rcu safe.
Michal Hocko0eef6152014-01-23 15:53:37 -08001109 *
1110 * We do not take a reference on the root of the tree walk
1111 * because we might race with the root removal when it would
1112 * be the only node in the iterated hierarchy and mem_cgroup_iter
1113 * would end up in an endless loop because it expects that at
1114 * least one valid node will be returned. Root cannot disappear
1115 * because caller of the iterator should hold it already so
1116 * skipping css reference should be safe.
Michal Hocko16248d82013-04-29 15:07:19 -07001117 */
Tejun Heo492eb212013-08-08 20:11:25 -04001118 if (next_css) {
Hugh Dickinsce482252014-03-03 15:38:24 -08001119 if ((next_css == &root->css) ||
1120 ((next_css->flags & CSS_ONLINE) && css_tryget(next_css)))
Hugh Dickinsd8ad3052014-01-23 15:53:32 -08001121 return mem_cgroup_from_css(next_css);
Michal Hocko0eef6152014-01-23 15:53:37 -08001122
1123 prev_css = next_css;
1124 goto skip_node;
Michal Hocko16248d82013-04-29 15:07:19 -07001125 }
1126
1127 return NULL;
1128}
1129
Johannes Weiner519ebea2013-07-03 15:04:51 -07001130static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1131{
1132 /*
1133 * When a group in the hierarchy below root is destroyed, the
1134 * hierarchy iterator can no longer be trusted since it might
1135 * have pointed to the destroyed group. Invalidate it.
1136 */
1137 atomic_inc(&root->dead_count);
1138}
1139
1140static struct mem_cgroup *
1141mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1142 struct mem_cgroup *root,
1143 int *sequence)
1144{
1145 struct mem_cgroup *position = NULL;
1146 /*
1147 * A cgroup destruction happens in two stages: offlining and
1148 * release. They are separated by a RCU grace period.
1149 *
1150 * If the iterator is valid, we may still race with an
1151 * offlining. The RCU lock ensures the object won't be
1152 * released, tryget will fail if we lost the race.
1153 */
1154 *sequence = atomic_read(&root->dead_count);
1155 if (iter->last_dead_count == *sequence) {
1156 smp_rmb();
1157 position = iter->last_visited;
Michal Hockoecc736f2014-01-23 15:53:35 -08001158
1159 /*
1160 * We cannot take a reference to root because we might race
1161 * with root removal and returning NULL would end up in
1162 * an endless loop on the iterator user level when root
1163 * would be returned all the time.
1164 */
1165 if (position && position != root &&
1166 !css_tryget(&position->css))
Johannes Weiner519ebea2013-07-03 15:04:51 -07001167 position = NULL;
1168 }
1169 return position;
1170}
1171
1172static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1173 struct mem_cgroup *last_visited,
1174 struct mem_cgroup *new_position,
Michal Hockoecc736f2014-01-23 15:53:35 -08001175 struct mem_cgroup *root,
Johannes Weiner519ebea2013-07-03 15:04:51 -07001176 int sequence)
1177{
Michal Hockoecc736f2014-01-23 15:53:35 -08001178 /* root reference counting symmetric to mem_cgroup_iter_load */
1179 if (last_visited && last_visited != root)
Johannes Weiner519ebea2013-07-03 15:04:51 -07001180 css_put(&last_visited->css);
1181 /*
1182 * We store the sequence count from the time @last_visited was
1183 * loaded successfully instead of rereading it here so that we
1184 * don't lose destruction events in between. We could have
1185 * raced with the destruction of @new_position after all.
1186 */
1187 iter->last_visited = new_position;
1188 smp_wmb();
1189 iter->last_dead_count = sequence;
1190}
1191
Johannes Weiner56600482012-01-12 17:17:59 -08001192/**
1193 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1194 * @root: hierarchy root
1195 * @prev: previously returned memcg, NULL on first invocation
1196 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1197 *
1198 * Returns references to children of the hierarchy below @root, or
1199 * @root itself, or %NULL after a full round-trip.
1200 *
1201 * Caller must pass the return value in @prev on subsequent
1202 * invocations for reference counting, or use mem_cgroup_iter_break()
1203 * to cancel a hierarchy walk before the round-trip is complete.
1204 *
1205 * Reclaimers can specify a zone and a priority level in @reclaim to
1206 * divide up the memcgs in the hierarchy among all concurrent
1207 * reclaimers operating on the same zone and priority.
1208 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001209struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001210 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001211 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001212{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001213 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001214 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001215
Andrew Morton694fbc02013-09-24 15:27:37 -07001216 if (mem_cgroup_disabled())
1217 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001218
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001219 if (!root)
1220 root = root_mem_cgroup;
1221
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001222 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001223 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001224
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001225 if (!root->use_hierarchy && root != root_mem_cgroup) {
1226 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001227 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001228 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001229 }
1230
Michal Hocko542f85f2013-04-29 15:07:15 -07001231 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001232 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001233 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001234 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001235
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001236 if (reclaim) {
1237 int nid = zone_to_nid(reclaim->zone);
1238 int zid = zone_idx(reclaim->zone);
1239 struct mem_cgroup_per_zone *mz;
1240
1241 mz = mem_cgroup_zoneinfo(root, nid, zid);
1242 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001243 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001244 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001245 goto out_unlock;
1246 }
Michal Hocko5f578162013-04-29 15:07:17 -07001247
Johannes Weiner519ebea2013-07-03 15:04:51 -07001248 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001249 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001250
Andrew Morton694fbc02013-09-24 15:27:37 -07001251 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001252
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001253 if (reclaim) {
Michal Hockoecc736f2014-01-23 15:53:35 -08001254 mem_cgroup_iter_update(iter, last_visited, memcg, root,
1255 seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001256
Michal Hocko19f39402013-04-29 15:07:18 -07001257 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001258 iter->generation++;
1259 else if (!prev && memcg)
1260 reclaim->generation = iter->generation;
1261 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001262
Andrew Morton694fbc02013-09-24 15:27:37 -07001263 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001264 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001265 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001266out_unlock:
1267 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001268out_css_put:
1269 if (prev && prev != root)
1270 css_put(&prev->css);
1271
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001272 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001273}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001274
Johannes Weiner56600482012-01-12 17:17:59 -08001275/**
1276 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1277 * @root: hierarchy root
1278 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1279 */
1280void mem_cgroup_iter_break(struct mem_cgroup *root,
1281 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001282{
1283 if (!root)
1284 root = root_mem_cgroup;
1285 if (prev && prev != root)
1286 css_put(&prev->css);
1287}
1288
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001289/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001290 * Iteration constructs for visiting all cgroups (under a tree). If
1291 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1292 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001293 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001294#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001295 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001296 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001297 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001298
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001299#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001300 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001301 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001302 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001303
David Rientjes68ae5642012-12-12 13:51:57 -08001304void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001305{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001306 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001307
Ying Han456f9982011-05-26 16:25:38 -07001308 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001309 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1310 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001311 goto out;
1312
1313 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001314 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001315 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1316 break;
1317 case PGMAJFAULT:
1318 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001319 break;
1320 default:
1321 BUG();
1322 }
1323out:
1324 rcu_read_unlock();
1325}
David Rientjes68ae5642012-12-12 13:51:57 -08001326EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001327
Johannes Weiner925b7672012-01-12 17:18:15 -08001328/**
1329 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1330 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001331 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001332 *
1333 * Returns the lru list vector holding pages for the given @zone and
1334 * @mem. This can be the global zone lruvec, if the memory controller
1335 * is disabled.
1336 */
1337struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1338 struct mem_cgroup *memcg)
1339{
1340 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001341 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001342
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001343 if (mem_cgroup_disabled()) {
1344 lruvec = &zone->lruvec;
1345 goto out;
1346 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001347
1348 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001349 lruvec = &mz->lruvec;
1350out:
1351 /*
1352 * Since a node can be onlined after the mem_cgroup was created,
1353 * we have to be prepared to initialize lruvec->zone here;
1354 * and if offlined then reonlined, we need to reinitialize it.
1355 */
1356 if (unlikely(lruvec->zone != zone))
1357 lruvec->zone = zone;
1358 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001359}
1360
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001361/*
1362 * Following LRU functions are allowed to be used without PCG_LOCK.
1363 * Operations are called by routine of global LRU independently from memcg.
1364 * What we have to take care of here is validness of pc->mem_cgroup.
1365 *
1366 * Changes to pc->mem_cgroup happens when
1367 * 1. charge
1368 * 2. moving account
1369 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1370 * It is added to LRU before charge.
1371 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1372 * When moving account, the page is not on LRU. It's isolated.
1373 */
1374
Johannes Weiner925b7672012-01-12 17:18:15 -08001375/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001376 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001377 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001378 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001379 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001380struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001381{
1382 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001383 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001384 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001385 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001386
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001387 if (mem_cgroup_disabled()) {
1388 lruvec = &zone->lruvec;
1389 goto out;
1390 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001391
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001392 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001393 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001394
1395 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001396 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001397 * an uncharged page off lru does nothing to secure
1398 * its former mem_cgroup from sudden removal.
1399 *
1400 * Our caller holds lru_lock, and PageCgroupUsed is updated
1401 * under page_cgroup lock: between them, they make all uses
1402 * of pc->mem_cgroup safe.
1403 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001404 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001405 pc->mem_cgroup = memcg = root_mem_cgroup;
1406
Johannes Weiner925b7672012-01-12 17:18:15 -08001407 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001408 lruvec = &mz->lruvec;
1409out:
1410 /*
1411 * Since a node can be onlined after the mem_cgroup was created,
1412 * we have to be prepared to initialize lruvec->zone here;
1413 * and if offlined then reonlined, we need to reinitialize it.
1414 */
1415 if (unlikely(lruvec->zone != zone))
1416 lruvec->zone = zone;
1417 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001418}
1419
1420/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001421 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1422 * @lruvec: mem_cgroup per zone lru vector
1423 * @lru: index of lru list the page is sitting on
1424 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001425 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001426 * This function must be called when a page is added to or removed from an
1427 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001428 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001429void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1430 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001431{
1432 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001433 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001434
1435 if (mem_cgroup_disabled())
1436 return;
1437
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001438 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1439 lru_size = mz->lru_size + lru;
1440 *lru_size += nr_pages;
1441 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001442}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001443
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001444/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001445 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001446 * hierarchy subtree
1447 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001448bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1449 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001450{
Johannes Weiner91c637342012-05-29 15:06:24 -07001451 if (root_memcg == memcg)
1452 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001453 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001454 return false;
Li Zefanb47f77b2013-09-23 16:55:43 +08001455 return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001456}
1457
1458static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1459 struct mem_cgroup *memcg)
1460{
1461 bool ret;
1462
Johannes Weiner91c637342012-05-29 15:06:24 -07001463 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001464 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001465 rcu_read_unlock();
1466 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001467}
1468
David Rientjesffbdccf2013-07-03 15:01:23 -07001469bool task_in_mem_cgroup(struct task_struct *task,
1470 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001471{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001472 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001473 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001474 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001475
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001476 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001477 if (p) {
1478 curr = try_get_mem_cgroup_from_mm(p->mm);
1479 task_unlock(p);
1480 } else {
1481 /*
1482 * All threads may have already detached their mm's, but the oom
1483 * killer still needs to detect if they have already been oom
1484 * killed to prevent needlessly killing additional tasks.
1485 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001486 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001487 curr = mem_cgroup_from_task(task);
1488 if (curr)
1489 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001490 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001491 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001492 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001493 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001494 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001495 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001496 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001497 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1498 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001499 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001500 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001501 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001502 return ret;
1503}
1504
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001505int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001506{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001507 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001508 unsigned long inactive;
1509 unsigned long active;
1510 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001511
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001512 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1513 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001514
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001515 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1516 if (gb)
1517 inactive_ratio = int_sqrt(10 * gb);
1518 else
1519 inactive_ratio = 1;
1520
Johannes Weiner9b272972011-11-02 13:38:23 -07001521 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001522}
1523
Balbir Singh6d61ef42009-01-07 18:08:06 -08001524#define mem_cgroup_from_res_counter(counter, member) \
1525 container_of(counter, struct mem_cgroup, member)
1526
Johannes Weiner19942822011-02-01 15:52:43 -08001527/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001528 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001529 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001530 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001531 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001532 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001533 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001534static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001535{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001536 unsigned long long margin;
1537
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001538 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001539 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001540 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001541 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001542}
1543
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001544int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001545{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001546 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001547 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001548 return vm_swappiness;
1549
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001550 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001551}
1552
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001553/*
1554 * memcg->moving_account is used for checking possibility that some thread is
1555 * calling move_account(). When a thread on CPU-A starts moving pages under
1556 * a memcg, other threads should check memcg->moving_account under
1557 * rcu_read_lock(), like this:
1558 *
1559 * CPU-A CPU-B
1560 * rcu_read_lock()
1561 * memcg->moving_account+1 if (memcg->mocing_account)
1562 * take heavy locks.
1563 * synchronize_rcu() update something.
1564 * rcu_read_unlock()
1565 * start move here.
1566 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001567
1568/* for quick checking without looking up memcg */
1569atomic_t memcg_moving __read_mostly;
1570
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001571static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001572{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001573 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001574 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001575 synchronize_rcu();
1576}
1577
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001578static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001579{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001580 /*
1581 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1582 * We check NULL in callee rather than caller.
1583 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001584 if (memcg) {
1585 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001586 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001587 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001588}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001589
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001590/*
1591 * 2 routines for checking "mem" is under move_account() or not.
1592 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001593 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1594 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001595 * pc->mem_cgroup may be overwritten.
1596 *
1597 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1598 * under hierarchy of moving cgroups. This is for
1599 * waiting at hith-memory prressure caused by "move".
1600 */
1601
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001602static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001603{
1604 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001605 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001606}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001607
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001608static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001609{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001610 struct mem_cgroup *from;
1611 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001612 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001613 /*
1614 * Unlike task_move routines, we access mc.to, mc.from not under
1615 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1616 */
1617 spin_lock(&mc.lock);
1618 from = mc.from;
1619 to = mc.to;
1620 if (!from)
1621 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001622
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001623 ret = mem_cgroup_same_or_subtree(memcg, from)
1624 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001625unlock:
1626 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001627 return ret;
1628}
1629
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001630static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001631{
1632 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001633 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001634 DEFINE_WAIT(wait);
1635 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1636 /* moving charge context might have finished. */
1637 if (mc.moving_task)
1638 schedule();
1639 finish_wait(&mc.waitq, &wait);
1640 return true;
1641 }
1642 }
1643 return false;
1644}
1645
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001646/*
1647 * Take this lock when
1648 * - a code tries to modify page's memcg while it's USED.
1649 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001650 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001651 */
1652static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1653 unsigned long *flags)
1654{
1655 spin_lock_irqsave(&memcg->move_lock, *flags);
1656}
1657
1658static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1659 unsigned long *flags)
1660{
1661 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1662}
1663
Sha Zhengju58cf1882013-02-22 16:32:05 -08001664#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001665/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001666 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001667 * @memcg: The memory cgroup that went over limit
1668 * @p: Task that is going to be killed
1669 *
1670 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1671 * enabled
1672 */
1673void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1674{
Tejun Heoe61734c2014-02-12 09:29:50 -05001675 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001676 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001677 struct mem_cgroup *iter;
1678 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001679
Sha Zhengju58cf1882013-02-22 16:32:05 -08001680 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001681 return;
1682
Michal Hocko08088cb2014-02-25 15:01:44 -08001683 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001684 rcu_read_lock();
1685
Tejun Heoe61734c2014-02-12 09:29:50 -05001686 pr_info("Task in ");
1687 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1688 pr_info(" killed as a result of limit of ");
1689 pr_cont_cgroup_path(memcg->css.cgroup);
1690 pr_info("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001691
Balbir Singhe2224322009-04-02 16:57:39 -07001692 rcu_read_unlock();
1693
Andrew Mortond0451972013-02-22 16:32:06 -08001694 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001695 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1696 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1697 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001698 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001699 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1700 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1701 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001702 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001703 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1704 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1705 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001706
1707 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001708 pr_info("Memory cgroup stats for ");
1709 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001710 pr_cont(":");
1711
1712 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1713 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1714 continue;
1715 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1716 K(mem_cgroup_read_stat(iter, i)));
1717 }
1718
1719 for (i = 0; i < NR_LRU_LISTS; i++)
1720 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1721 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1722
1723 pr_cont("\n");
1724 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001725 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001726}
1727
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001728/*
1729 * This function returns the number of memcg under hierarchy tree. Returns
1730 * 1(self count) if no children.
1731 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001732static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001733{
1734 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001735 struct mem_cgroup *iter;
1736
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001737 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001738 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001739 return num;
1740}
1741
Balbir Singh6d61ef42009-01-07 18:08:06 -08001742/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001743 * Return the memory (and swap, if configured) limit for a memcg.
1744 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001745static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001746{
1747 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001748
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001749 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001750
David Rientjesa63d83f2010-08-09 17:19:46 -07001751 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001752 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001753 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001754 if (mem_cgroup_swappiness(memcg)) {
1755 u64 memsw;
1756
1757 limit += total_swap_pages << PAGE_SHIFT;
1758 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1759
1760 /*
1761 * If memsw is finite and limits the amount of swap space
1762 * available to this memcg, return that limit.
1763 */
1764 limit = min(limit, memsw);
1765 }
1766
1767 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001768}
1769
David Rientjes19965462012-12-11 16:00:26 -08001770static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1771 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001772{
1773 struct mem_cgroup *iter;
1774 unsigned long chosen_points = 0;
1775 unsigned long totalpages;
1776 unsigned int points = 0;
1777 struct task_struct *chosen = NULL;
1778
David Rientjes876aafb2012-07-31 16:43:48 -07001779 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001780 * If current has a pending SIGKILL or is exiting, then automatically
1781 * select it. The goal is to allow it to allocate so that it may
1782 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001783 */
David Rientjes465adcf2013-04-29 15:08:45 -07001784 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001785 set_thread_flag(TIF_MEMDIE);
1786 return;
1787 }
1788
1789 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001790 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1791 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001792 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001793 struct task_struct *task;
1794
Tejun Heo72ec7022013-08-08 20:11:26 -04001795 css_task_iter_start(&iter->css, &it);
1796 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001797 switch (oom_scan_process_thread(task, totalpages, NULL,
1798 false)) {
1799 case OOM_SCAN_SELECT:
1800 if (chosen)
1801 put_task_struct(chosen);
1802 chosen = task;
1803 chosen_points = ULONG_MAX;
1804 get_task_struct(chosen);
1805 /* fall through */
1806 case OOM_SCAN_CONTINUE:
1807 continue;
1808 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001809 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001810 mem_cgroup_iter_break(memcg, iter);
1811 if (chosen)
1812 put_task_struct(chosen);
1813 return;
1814 case OOM_SCAN_OK:
1815 break;
1816 };
1817 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001818 if (!points || points < chosen_points)
1819 continue;
1820 /* Prefer thread group leaders for display purposes */
1821 if (points == chosen_points &&
1822 thread_group_leader(chosen))
1823 continue;
1824
1825 if (chosen)
1826 put_task_struct(chosen);
1827 chosen = task;
1828 chosen_points = points;
1829 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001830 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001831 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001832 }
1833
1834 if (!chosen)
1835 return;
1836 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001837 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1838 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001839}
1840
Johannes Weiner56600482012-01-12 17:17:59 -08001841static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1842 gfp_t gfp_mask,
1843 unsigned long flags)
1844{
1845 unsigned long total = 0;
1846 bool noswap = false;
1847 int loop;
1848
1849 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1850 noswap = true;
1851 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1852 noswap = true;
1853
1854 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1855 if (loop)
1856 drain_all_stock_async(memcg);
1857 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1858 /*
1859 * Allow limit shrinkers, which are triggered directly
1860 * by userspace, to catch signals and stop reclaim
1861 * after minimal progress, regardless of the margin.
1862 */
1863 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1864 break;
1865 if (mem_cgroup_margin(memcg))
1866 break;
1867 /*
1868 * If nothing was reclaimed after two attempts, there
1869 * may be no reclaimable pages in this hierarchy.
1870 */
1871 if (loop && !total)
1872 break;
1873 }
1874 return total;
1875}
1876
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001877/**
1878 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001879 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001880 * @nid: the node ID to be checked.
1881 * @noswap : specify true here if the user wants flle only information.
1882 *
1883 * This function returns whether the specified memcg contains any
1884 * reclaimable pages on a node. Returns true if there are any reclaimable
1885 * pages in the node.
1886 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001887static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001888 int nid, bool noswap)
1889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001890 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001891 return true;
1892 if (noswap || !total_swap_pages)
1893 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001894 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001895 return true;
1896 return false;
1897
1898}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001899#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001900
1901/*
1902 * Always updating the nodemask is not very good - even if we have an empty
1903 * list or the wrong list here, we can start from some node and traverse all
1904 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1905 *
1906 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001907static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001908{
1909 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001910 /*
1911 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1912 * pagein/pageout changes since the last update.
1913 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001914 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001915 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001916 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001917 return;
1918
Ying Han889976d2011-05-26 16:25:33 -07001919 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001920 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001921
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001922 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001923
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001924 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1925 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001926 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001927
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001928 atomic_set(&memcg->numainfo_events, 0);
1929 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001930}
1931
1932/*
1933 * Selecting a node where we start reclaim from. Because what we need is just
1934 * reducing usage counter, start from anywhere is O,K. Considering
1935 * memory reclaim from current node, there are pros. and cons.
1936 *
1937 * Freeing memory from current node means freeing memory from a node which
1938 * we'll use or we've used. So, it may make LRU bad. And if several threads
1939 * hit limits, it will see a contention on a node. But freeing from remote
1940 * node means more costs for memory reclaim because of memory latency.
1941 *
1942 * Now, we use round-robin. Better algorithm is welcomed.
1943 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001944int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001945{
1946 int node;
1947
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001948 mem_cgroup_may_update_nodemask(memcg);
1949 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001950
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001951 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001952 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001953 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001954 /*
1955 * We call this when we hit limit, not when pages are added to LRU.
1956 * No LRU may hold pages because all pages are UNEVICTABLE or
1957 * memcg is too small and all pages are not on LRU. In that case,
1958 * we use curret node.
1959 */
1960 if (unlikely(node == MAX_NUMNODES))
1961 node = numa_node_id();
1962
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001963 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001964 return node;
1965}
1966
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001967/*
1968 * Check all nodes whether it contains reclaimable pages or not.
1969 * For quick scan, we make use of scan_nodes. This will allow us to skip
1970 * unused nodes. But scan_nodes is lazily updated and may not cotain
1971 * enough new information. We need to do double check.
1972 */
1973static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1974{
1975 int nid;
1976
1977 /*
1978 * quick check...making use of scan_node.
1979 * We can skip unused nodes.
1980 */
1981 if (!nodes_empty(memcg->scan_nodes)) {
1982 for (nid = first_node(memcg->scan_nodes);
1983 nid < MAX_NUMNODES;
1984 nid = next_node(nid, memcg->scan_nodes)) {
1985
1986 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1987 return true;
1988 }
1989 }
1990 /*
1991 * Check rest of nodes.
1992 */
1993 for_each_node_state(nid, N_MEMORY) {
1994 if (node_isset(nid, memcg->scan_nodes))
1995 continue;
1996 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1997 return true;
1998 }
1999 return false;
2000}
2001
Ying Han889976d2011-05-26 16:25:33 -07002002#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002003int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002004{
2005 return 0;
2006}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002007
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002008static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
2009{
2010 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
2011}
Ying Han889976d2011-05-26 16:25:33 -07002012#endif
2013
Andrew Morton0608f432013-09-24 15:27:41 -07002014static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2015 struct zone *zone,
2016 gfp_t gfp_mask,
2017 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002018{
Andrew Morton0608f432013-09-24 15:27:41 -07002019 struct mem_cgroup *victim = NULL;
2020 int total = 0;
2021 int loop = 0;
2022 unsigned long excess;
2023 unsigned long nr_scanned;
2024 struct mem_cgroup_reclaim_cookie reclaim = {
2025 .zone = zone,
2026 .priority = 0,
2027 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002028
Andrew Morton0608f432013-09-24 15:27:41 -07002029 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002030
Andrew Morton0608f432013-09-24 15:27:41 -07002031 while (1) {
2032 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
2033 if (!victim) {
2034 loop++;
2035 if (loop >= 2) {
2036 /*
2037 * If we have not been able to reclaim
2038 * anything, it might because there are
2039 * no reclaimable pages under this hierarchy
2040 */
2041 if (!total)
2042 break;
2043 /*
2044 * We want to do more targeted reclaim.
2045 * excess >> 2 is not to excessive so as to
2046 * reclaim too much, nor too less that we keep
2047 * coming back to reclaim from this cgroup
2048 */
2049 if (total >= (excess >> 2) ||
2050 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
2051 break;
2052 }
2053 continue;
2054 }
2055 if (!mem_cgroup_reclaimable(victim, false))
2056 continue;
2057 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2058 zone, &nr_scanned);
2059 *total_scanned += nr_scanned;
2060 if (!res_counter_soft_limit_excess(&root_memcg->res))
2061 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002062 }
Andrew Morton0608f432013-09-24 15:27:41 -07002063 mem_cgroup_iter_break(root_memcg, victim);
2064 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002065}
2066
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002067#ifdef CONFIG_LOCKDEP
2068static struct lockdep_map memcg_oom_lock_dep_map = {
2069 .name = "memcg_oom_lock",
2070};
2071#endif
2072
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002073static DEFINE_SPINLOCK(memcg_oom_lock);
2074
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002075/*
2076 * Check OOM-Killer is already running under our hierarchy.
2077 * If someone is running, return false.
2078 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002079static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002080{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002081 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002082
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002083 spin_lock(&memcg_oom_lock);
2084
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002085 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002086 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002087 /*
2088 * this subtree of our hierarchy is already locked
2089 * so we cannot give a lock.
2090 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002091 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002092 mem_cgroup_iter_break(memcg, iter);
2093 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002094 } else
2095 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002096 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002097
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002098 if (failed) {
2099 /*
2100 * OK, we failed to lock the whole subtree so we have
2101 * to clean up what we set up to the failing subtree
2102 */
2103 for_each_mem_cgroup_tree(iter, memcg) {
2104 if (iter == failed) {
2105 mem_cgroup_iter_break(memcg, iter);
2106 break;
2107 }
2108 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002109 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002110 } else
2111 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002112
2113 spin_unlock(&memcg_oom_lock);
2114
2115 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002116}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002117
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002118static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002119{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002120 struct mem_cgroup *iter;
2121
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002122 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002123 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002124 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002125 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002126 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002127}
2128
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002129static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002130{
2131 struct mem_cgroup *iter;
2132
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002133 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002134 atomic_inc(&iter->under_oom);
2135}
2136
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002137static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002138{
2139 struct mem_cgroup *iter;
2140
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002141 /*
2142 * When a new child is created while the hierarchy is under oom,
2143 * mem_cgroup_oom_lock() may not be called. We have to use
2144 * atomic_add_unless() here.
2145 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002146 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002147 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002148}
2149
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002150static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2151
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002152struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002153 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002154 wait_queue_t wait;
2155};
2156
2157static int memcg_oom_wake_function(wait_queue_t *wait,
2158 unsigned mode, int sync, void *arg)
2159{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002160 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2161 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002162 struct oom_wait_info *oom_wait_info;
2163
2164 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002165 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002166
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002167 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002168 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002169 * Then we can use css_is_ancestor without taking care of RCU.
2170 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002171 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2172 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002173 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002174 return autoremove_wake_function(wait, mode, sync, arg);
2175}
2176
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002177static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002178{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002179 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002180 /* for filtering, pass "memcg" as argument. */
2181 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002182}
2183
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002184static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002185{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002186 if (memcg && atomic_read(&memcg->under_oom))
2187 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002188}
2189
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002190static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002191{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002192 if (!current->memcg_oom.may_oom)
2193 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002194 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002195 * We are in the middle of the charge context here, so we
2196 * don't want to block when potentially sitting on a callstack
2197 * that holds all kinds of filesystem and mm locks.
2198 *
2199 * Also, the caller may handle a failed allocation gracefully
2200 * (like optional page cache readahead) and so an OOM killer
2201 * invocation might not even be necessary.
2202 *
2203 * That's why we don't do anything here except remember the
2204 * OOM context and then deal with it at the end of the page
2205 * fault when the stack is unwound, the locks are released,
2206 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002207 */
Johannes Weiner49426422013-10-16 13:46:59 -07002208 css_get(&memcg->css);
2209 current->memcg_oom.memcg = memcg;
2210 current->memcg_oom.gfp_mask = mask;
2211 current->memcg_oom.order = order;
2212}
2213
2214/**
2215 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2216 * @handle: actually kill/wait or just clean up the OOM state
2217 *
2218 * This has to be called at the end of a page fault if the memcg OOM
2219 * handler was enabled.
2220 *
2221 * Memcg supports userspace OOM handling where failed allocations must
2222 * sleep on a waitqueue until the userspace task resolves the
2223 * situation. Sleeping directly in the charge context with all kinds
2224 * of locks held is not a good idea, instead we remember an OOM state
2225 * in the task and mem_cgroup_oom_synchronize() has to be called at
2226 * the end of the page fault to complete the OOM handling.
2227 *
2228 * Returns %true if an ongoing memcg OOM situation was detected and
2229 * completed, %false otherwise.
2230 */
2231bool mem_cgroup_oom_synchronize(bool handle)
2232{
2233 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2234 struct oom_wait_info owait;
2235 bool locked;
2236
2237 /* OOM is global, do not handle */
2238 if (!memcg)
2239 return false;
2240
2241 if (!handle)
2242 goto cleanup;
2243
2244 owait.memcg = memcg;
2245 owait.wait.flags = 0;
2246 owait.wait.func = memcg_oom_wake_function;
2247 owait.wait.private = current;
2248 INIT_LIST_HEAD(&owait.wait.task_list);
2249
2250 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002251 mem_cgroup_mark_under_oom(memcg);
2252
2253 locked = mem_cgroup_oom_trylock(memcg);
2254
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002255 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002256 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002257
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002258 if (locked && !memcg->oom_kill_disable) {
2259 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002260 finish_wait(&memcg_oom_waitq, &owait.wait);
2261 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2262 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002263 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002264 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002265 mem_cgroup_unmark_under_oom(memcg);
2266 finish_wait(&memcg_oom_waitq, &owait.wait);
2267 }
2268
2269 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002270 mem_cgroup_oom_unlock(memcg);
2271 /*
2272 * There is no guarantee that an OOM-lock contender
2273 * sees the wakeups triggered by the OOM kill
2274 * uncharges. Wake any sleepers explicitely.
2275 */
2276 memcg_oom_recover(memcg);
2277 }
Johannes Weiner49426422013-10-16 13:46:59 -07002278cleanup:
2279 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002280 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002281 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002282}
2283
Balbir Singhd69b0422009-06-17 16:26:34 -07002284/*
2285 * Currently used to update mapped file statistics, but the routine can be
2286 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002287 *
2288 * Notes: Race condition
2289 *
2290 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2291 * it tends to be costly. But considering some conditions, we doesn't need
2292 * to do so _always_.
2293 *
2294 * Considering "charge", lock_page_cgroup() is not required because all
2295 * file-stat operations happen after a page is attached to radix-tree. There
2296 * are no race with "charge".
2297 *
2298 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2299 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2300 * if there are race with "uncharge". Statistics itself is properly handled
2301 * by flags.
2302 *
2303 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002304 * small, we check mm->moving_account and detect there are possibility of race
2305 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002306 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002307
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002308void __mem_cgroup_begin_update_page_stat(struct page *page,
2309 bool *locked, unsigned long *flags)
2310{
2311 struct mem_cgroup *memcg;
2312 struct page_cgroup *pc;
2313
2314 pc = lookup_page_cgroup(page);
2315again:
2316 memcg = pc->mem_cgroup;
2317 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2318 return;
2319 /*
2320 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002321 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002322 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002323 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002324 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002325 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002326 return;
2327
2328 move_lock_mem_cgroup(memcg, flags);
2329 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2330 move_unlock_mem_cgroup(memcg, flags);
2331 goto again;
2332 }
2333 *locked = true;
2334}
2335
2336void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2337{
2338 struct page_cgroup *pc = lookup_page_cgroup(page);
2339
2340 /*
2341 * It's guaranteed that pc->mem_cgroup never changes while
2342 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002343 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002344 */
2345 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2346}
2347
Greg Thelen2a7106f2011-01-13 15:47:37 -08002348void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002349 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002350{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002351 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002352 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002353 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002354
Johannes Weinercfa44942012-01-12 17:18:38 -08002355 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002356 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002357
Sha Zhengju658b72c2013-09-12 15:13:52 -07002358 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002359 memcg = pc->mem_cgroup;
2360 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002361 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002362
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002363 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002364}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002365
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002366/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002367 * size of first charge trial. "32" comes from vmscan.c's magic value.
2368 * TODO: maybe necessary to use big numbers in big irons.
2369 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002370#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002371struct memcg_stock_pcp {
2372 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002373 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002374 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002375 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002376#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002377};
2378static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002379static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002380
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002381/**
2382 * consume_stock: Try to consume stocked charge on this cpu.
2383 * @memcg: memcg to consume from.
2384 * @nr_pages: how many pages to charge.
2385 *
2386 * The charges will only happen if @memcg matches the current cpu's memcg
2387 * stock, and at least @nr_pages are available in that stock. Failure to
2388 * service an allocation will refill the stock.
2389 *
2390 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002391 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002392static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002393{
2394 struct memcg_stock_pcp *stock;
2395 bool ret = true;
2396
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002397 if (nr_pages > CHARGE_BATCH)
2398 return false;
2399
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002400 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002401 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2402 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002403 else /* need to call res_counter_charge */
2404 ret = false;
2405 put_cpu_var(memcg_stock);
2406 return ret;
2407}
2408
2409/*
2410 * Returns stocks cached in percpu to res_counter and reset cached information.
2411 */
2412static void drain_stock(struct memcg_stock_pcp *stock)
2413{
2414 struct mem_cgroup *old = stock->cached;
2415
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002416 if (stock->nr_pages) {
2417 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2418
2419 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002420 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002421 res_counter_uncharge(&old->memsw, bytes);
2422 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002423 }
2424 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002425}
2426
2427/*
2428 * This must be called under preempt disabled or must be called by
2429 * a thread which is pinned to local cpu.
2430 */
2431static void drain_local_stock(struct work_struct *dummy)
2432{
2433 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2434 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002435 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002436}
2437
Michal Hockoe4777492013-02-22 16:35:40 -08002438static void __init memcg_stock_init(void)
2439{
2440 int cpu;
2441
2442 for_each_possible_cpu(cpu) {
2443 struct memcg_stock_pcp *stock =
2444 &per_cpu(memcg_stock, cpu);
2445 INIT_WORK(&stock->work, drain_local_stock);
2446 }
2447}
2448
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002449/*
2450 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002451 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002452 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002453static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002454{
2455 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2456
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002457 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002458 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002459 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002460 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002461 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002462 put_cpu_var(memcg_stock);
2463}
2464
2465/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002466 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002467 * of the hierarchy under it. sync flag says whether we should block
2468 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002469 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002470static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002471{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002472 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002473
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002474 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002475 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002476 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002477 for_each_online_cpu(cpu) {
2478 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002479 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002480
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002481 memcg = stock->cached;
2482 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002483 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002484 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002485 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002486 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2487 if (cpu == curcpu)
2488 drain_local_stock(&stock->work);
2489 else
2490 schedule_work_on(cpu, &stock->work);
2491 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002492 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002493 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002494
2495 if (!sync)
2496 goto out;
2497
2498 for_each_online_cpu(cpu) {
2499 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002500 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002501 flush_work(&stock->work);
2502 }
2503out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002504 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002505}
2506
2507/*
2508 * Tries to drain stocked charges in other cpus. This function is asynchronous
2509 * and just put a work per cpu for draining localy on each cpu. Caller can
2510 * expects some charges will be back to res_counter later but cannot wait for
2511 * it.
2512 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002513static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002514{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002515 /*
2516 * If someone calls draining, avoid adding more kworker runs.
2517 */
2518 if (!mutex_trylock(&percpu_charge_mutex))
2519 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002520 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002521 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002522}
2523
2524/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002525static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002526{
2527 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002528 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002529 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002530 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002531}
2532
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002533/*
2534 * This function drains percpu counter value from DEAD cpu and
2535 * move it to local cpu. Note that this function can be preempted.
2536 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002537static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002538{
2539 int i;
2540
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002541 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002542 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002543 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002544
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002545 per_cpu(memcg->stat->count[i], cpu) = 0;
2546 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002547 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002548 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002549 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002550
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002551 per_cpu(memcg->stat->events[i], cpu) = 0;
2552 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002553 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002554 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002555}
2556
Paul Gortmaker0db06282013-06-19 14:53:51 -04002557static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002558 unsigned long action,
2559 void *hcpu)
2560{
2561 int cpu = (unsigned long)hcpu;
2562 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002563 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002564
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002565 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002566 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002567
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002568 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002569 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002570
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002571 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002572 mem_cgroup_drain_pcp_counter(iter, cpu);
2573
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002574 stock = &per_cpu(memcg_stock, cpu);
2575 drain_stock(stock);
2576 return NOTIFY_OK;
2577}
2578
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002579
2580/* See __mem_cgroup_try_charge() for details */
2581enum {
2582 CHARGE_OK, /* success */
2583 CHARGE_RETRY, /* need to retry but retry is not bad */
2584 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2585 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002586};
2587
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002588static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002589 unsigned int nr_pages, unsigned int min_pages,
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002590 bool invoke_oom)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002591{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002592 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002593 struct mem_cgroup *mem_over_limit;
2594 struct res_counter *fail_res;
2595 unsigned long flags = 0;
2596 int ret;
2597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002598 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002599
2600 if (likely(!ret)) {
2601 if (!do_swap_account)
2602 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002603 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002604 if (likely(!ret))
2605 return CHARGE_OK;
2606
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002607 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002608 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2609 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2610 } else
2611 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002612 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002613 * Never reclaim on behalf of optional batching, retry with a
2614 * single page instead.
2615 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002616 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002617 return CHARGE_RETRY;
2618
2619 if (!(gfp_mask & __GFP_WAIT))
2620 return CHARGE_WOULDBLOCK;
2621
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002622 if (gfp_mask & __GFP_NORETRY)
2623 return CHARGE_NOMEM;
2624
Johannes Weiner56600482012-01-12 17:17:59 -08002625 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002626 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002627 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002628 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002629 * Even though the limit is exceeded at this point, reclaim
2630 * may have been able to free some pages. Retry the charge
2631 * before killing the task.
2632 *
2633 * Only for regular pages, though: huge pages are rather
2634 * unlikely to succeed so close to the limit, and we fall back
2635 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002636 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002637 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002638 return CHARGE_RETRY;
2639
2640 /*
2641 * At task move, charge accounts can be doubly counted. So, it's
2642 * better to wait until the end of task_move if something is going on.
2643 */
2644 if (mem_cgroup_wait_acct_move(mem_over_limit))
2645 return CHARGE_RETRY;
2646
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002647 if (invoke_oom)
2648 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002649
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002650 return CHARGE_NOMEM;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002651}
2652
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002653/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002654 * __mem_cgroup_try_charge() does
2655 * 1. detect memcg to be charged against from passed *mm and *ptr,
2656 * 2. update res_counter
2657 * 3. call memory reclaim if necessary.
2658 *
2659 * In some special case, if the task is fatal, fatal_signal_pending() or
2660 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2661 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2662 * as possible without any hazards. 2: all pages should have a valid
2663 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2664 * pointer, that is treated as a charge to root_mem_cgroup.
2665 *
2666 * So __mem_cgroup_try_charge() will return
2667 * 0 ... on success, filling *ptr with a valid memcg pointer.
2668 * -ENOMEM ... charge failure because of resource limits.
2669 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2670 *
2671 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2672 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002673 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002674static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002675 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002676 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002677 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002678 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002679{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002680 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002681 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002682 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002683 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002684
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002685 /*
2686 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2687 * in system level. So, allow to go ahead dying process in addition to
2688 * MEMDIE process.
2689 */
2690 if (unlikely(test_thread_flag(TIF_MEMDIE)
2691 || fatal_signal_pending(current)))
2692 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002693
Johannes Weiner49426422013-10-16 13:46:59 -07002694 if (unlikely(task_in_memcg_oom(current)))
Johannes Weiner1f14c1a2013-12-12 17:12:35 -08002695 goto nomem;
Johannes Weiner49426422013-10-16 13:46:59 -07002696
Johannes Weinera0d8b002013-12-12 17:12:20 -08002697 if (gfp_mask & __GFP_NOFAIL)
2698 oom = false;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002699again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002700 if (*ptr) { /* css should be a valid one */
2701 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002702 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002703 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002704 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002705 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002706 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002707 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002708 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002709
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002710 rcu_read_lock();
2711 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002712 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002713 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002715 * race with swapoff. Then, we have small risk of mis-accouning.
2716 * But such kind of mis-account by race always happens because
2717 * we don't have cgroup_mutex(). It's overkill and we allo that
2718 * small race, here.
2719 * (*) swapoff at el will charge against mm-struct not against
2720 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002721 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002723 if (!memcg)
2724 memcg = root_mem_cgroup;
2725 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002726 rcu_read_unlock();
2727 goto done;
2728 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002729 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002730 /*
2731 * It seems dagerous to access memcg without css_get().
2732 * But considering how consume_stok works, it's not
2733 * necessary. If consume_stock success, some charges
2734 * from this memcg are cached on this cpu. So, we
2735 * don't need to call css_get()/css_tryget() before
2736 * calling consume_stock().
2737 */
2738 rcu_read_unlock();
2739 goto done;
2740 }
2741 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002742 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002743 rcu_read_unlock();
2744 goto again;
2745 }
2746 rcu_read_unlock();
2747 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002748
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002749 do {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002750 bool invoke_oom = oom && !nr_oom_retries;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002751
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002752 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002753 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002754 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002755 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002756 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002757
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002758 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
2759 nr_pages, invoke_oom);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002760 switch (ret) {
2761 case CHARGE_OK:
2762 break;
2763 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002764 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002765 css_put(&memcg->css);
2766 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002767 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002768 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002769 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002770 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002771 case CHARGE_NOMEM: /* OOM routine works */
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002772 if (!oom || invoke_oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002773 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002774 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002775 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002776 nr_oom_retries--;
2777 break;
Balbir Singh66e17072008-02-07 00:13:56 -08002778 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002779 } while (ret != CHARGE_OK);
2780
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002781 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002782 refill_stock(memcg, batch - nr_pages);
2783 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002784done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002785 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002786 return 0;
2787nomem:
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002788 if (!(gfp_mask & __GFP_NOFAIL)) {
2789 *ptr = NULL;
2790 return -ENOMEM;
2791 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002792bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002793 *ptr = root_mem_cgroup;
2794 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002795}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002796
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002797/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002798 * Somemtimes we have to undo a charge we got by try_charge().
2799 * This function is for that and do uncharge, put css's refcnt.
2800 * gotten by try_charge().
2801 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002802static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002803 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002804{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002805 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002806 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002807
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002808 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002809 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002810 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002811 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002812}
2813
2814/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002815 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2816 * This is useful when moving usage to parent cgroup.
2817 */
2818static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2819 unsigned int nr_pages)
2820{
2821 unsigned long bytes = nr_pages * PAGE_SIZE;
2822
2823 if (mem_cgroup_is_root(memcg))
2824 return;
2825
2826 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2827 if (do_swap_account)
2828 res_counter_uncharge_until(&memcg->memsw,
2829 memcg->memsw.parent, bytes);
2830}
2831
2832/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002833 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002834 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2835 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2836 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002837 */
2838static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2839{
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002840 /* ID 0 is unused ID */
2841 if (!id)
2842 return NULL;
Li Zefan34c00c32013-09-23 16:56:01 +08002843 return mem_cgroup_from_id(id);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002844}
2845
Wu Fengguange42d9d52009-12-16 12:19:59 +01002846struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002847{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002848 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002849 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002850 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002851 swp_entry_t ent;
2852
Sasha Levin309381fea2014-01-23 15:52:54 -08002853 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002854
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002855 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002856 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002857 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002858 memcg = pc->mem_cgroup;
2859 if (memcg && !css_tryget(&memcg->css))
2860 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002861 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002862 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002863 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002864 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002865 memcg = mem_cgroup_lookup(id);
2866 if (memcg && !css_tryget(&memcg->css))
2867 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002868 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002869 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002870 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002871 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002872}
2873
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002874static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002875 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002876 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002877 enum charge_type ctype,
2878 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002879{
Johannes Weinerce587e62012-04-24 20:22:33 +02002880 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002881 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002882 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002883 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002884 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002885
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002886 lock_page_cgroup(pc);
Sasha Levin309381fea2014-01-23 15:52:54 -08002887 VM_BUG_ON_PAGE(PageCgroupUsed(pc), page);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002888 /*
2889 * we don't need page_cgroup_lock about tail pages, becase they are not
2890 * accessed by any other context at this point.
2891 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002892
2893 /*
2894 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2895 * may already be on some other mem_cgroup's LRU. Take care of it.
2896 */
2897 if (lrucare) {
2898 zone = page_zone(page);
2899 spin_lock_irq(&zone->lru_lock);
2900 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002901 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002902 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002903 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002904 was_on_lru = true;
2905 }
2906 }
2907
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002908 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002909 /*
2910 * We access a page_cgroup asynchronously without lock_page_cgroup().
2911 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2912 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2913 * before USED bit, we need memory barrier here.
2914 * See mem_cgroup_add_lru_list(), etc.
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002915 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002916 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002917 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002918
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002919 if (lrucare) {
2920 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002921 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Sasha Levin309381fea2014-01-23 15:52:54 -08002922 VM_BUG_ON_PAGE(PageLRU(page), page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002923 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002924 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002925 }
2926 spin_unlock_irq(&zone->lru_lock);
2927 }
2928
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002929 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002930 anon = true;
2931 else
2932 anon = false;
2933
David Rientjesb070e652013-05-07 16:18:09 -07002934 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002935 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002936
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002937 /*
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002938 * "charge_statistics" updated event counter. Then, check it.
2939 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2940 * if they exceeds softlimit.
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002941 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002942 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002943}
2944
Glauber Costa7cf27982012-12-18 14:22:55 -08002945static DEFINE_MUTEX(set_limit_mutex);
2946
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002947#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08002948static DEFINE_MUTEX(activate_kmem_mutex);
2949
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002950static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2951{
2952 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
Vladimir Davydov6de64be2014-01-23 15:53:08 -08002953 memcg_kmem_is_active(memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002954}
2955
Glauber Costa1f458cb2012-12-18 14:22:50 -08002956/*
2957 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2958 * in the memcg_cache_params struct.
2959 */
2960static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2961{
2962 struct kmem_cache *cachep;
2963
2964 VM_BUG_ON(p->is_root_cache);
2965 cachep = p->root_cache;
Qiang Huang7a67d7a2013-11-12 15:08:24 -08002966 return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg));
Glauber Costa1f458cb2012-12-18 14:22:50 -08002967}
2968
Glauber Costa749c5412012-12-18 14:23:01 -08002969#ifdef CONFIG_SLABINFO
Tejun Heo2da8ca82013-12-05 12:28:04 -05002970static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v)
Glauber Costa749c5412012-12-18 14:23:01 -08002971{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002972 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Glauber Costa749c5412012-12-18 14:23:01 -08002973 struct memcg_cache_params *params;
2974
2975 if (!memcg_can_account_kmem(memcg))
2976 return -EIO;
2977
2978 print_slabinfo_header(m);
2979
2980 mutex_lock(&memcg->slab_caches_mutex);
2981 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2982 cache_show(memcg_params_to_cache(params), m);
2983 mutex_unlock(&memcg->slab_caches_mutex);
2984
2985 return 0;
2986}
2987#endif
2988
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002989static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2990{
2991 struct res_counter *fail_res;
2992 struct mem_cgroup *_memcg;
2993 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002994
2995 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2996 if (ret)
2997 return ret;
2998
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002999 _memcg = memcg;
3000 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
Qiang Huangb9921ec2013-11-12 15:07:22 -08003001 &_memcg, oom_gfp_allowed(gfp));
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003002
3003 if (ret == -EINTR) {
3004 /*
3005 * __mem_cgroup_try_charge() chosed to bypass to root due to
3006 * OOM kill or fatal signal. Since our only options are to
3007 * either fail the allocation or charge it to this cgroup, do
3008 * it as a temporary condition. But we can't fail. From a
3009 * kmem/slab perspective, the cache has already been selected,
3010 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3011 * our minds.
3012 *
3013 * This condition will only trigger if the task entered
3014 * memcg_charge_kmem in a sane state, but was OOM-killed during
3015 * __mem_cgroup_try_charge() above. Tasks that were already
3016 * dying when the allocation triggers should have been already
3017 * directed to the root cgroup in memcontrol.h
3018 */
3019 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3020 if (do_swap_account)
3021 res_counter_charge_nofail(&memcg->memsw, size,
3022 &fail_res);
3023 ret = 0;
3024 } else if (ret)
3025 res_counter_uncharge(&memcg->kmem, size);
3026
3027 return ret;
3028}
3029
3030static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3031{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003032 res_counter_uncharge(&memcg->res, size);
3033 if (do_swap_account)
3034 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003035
3036 /* Not down to 0 */
3037 if (res_counter_uncharge(&memcg->kmem, size))
3038 return;
3039
Li Zefan10d5ebf2013-07-08 16:00:33 -07003040 /*
3041 * Releases a reference taken in kmem_cgroup_css_offline in case
3042 * this last uncharge is racing with the offlining code or it is
3043 * outliving the memcg existence.
3044 *
3045 * The memory barrier imposed by test&clear is paired with the
3046 * explicit one in memcg_kmem_mark_dead().
3047 */
Glauber Costa7de37682012-12-18 14:22:07 -08003048 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003049 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003050}
3051
Glauber Costa2633d7a2012-12-18 14:22:34 -08003052/*
3053 * helper for acessing a memcg's index. It will be used as an index in the
3054 * child cache array in kmem_cache, and also to derive its name. This function
3055 * will return -1 when this is not a kmem-limited memcg.
3056 */
3057int memcg_cache_id(struct mem_cgroup *memcg)
3058{
3059 return memcg ? memcg->kmemcg_id : -1;
3060}
3061
Glauber Costa55007d82012-12-18 14:22:38 -08003062static size_t memcg_caches_array_size(int num_groups)
3063{
3064 ssize_t size;
3065 if (num_groups <= 0)
3066 return 0;
3067
3068 size = 2 * num_groups;
3069 if (size < MEMCG_CACHES_MIN_SIZE)
3070 size = MEMCG_CACHES_MIN_SIZE;
3071 else if (size > MEMCG_CACHES_MAX_SIZE)
3072 size = MEMCG_CACHES_MAX_SIZE;
3073
3074 return size;
3075}
3076
3077/*
3078 * We should update the current array size iff all caches updates succeed. This
3079 * can only be done from the slab side. The slab mutex needs to be held when
3080 * calling this.
3081 */
3082void memcg_update_array_size(int num)
3083{
3084 if (num > memcg_limited_groups_array_size)
3085 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3086}
3087
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003088static void kmem_cache_destroy_work_func(struct work_struct *w);
3089
Glauber Costa55007d82012-12-18 14:22:38 -08003090int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3091{
3092 struct memcg_cache_params *cur_params = s->memcg_params;
3093
Qiang Huangf35c3a82013-11-12 15:08:22 -08003094 VM_BUG_ON(!is_root_cache(s));
Glauber Costa55007d82012-12-18 14:22:38 -08003095
3096 if (num_groups > memcg_limited_groups_array_size) {
3097 int i;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003098 struct memcg_cache_params *new_params;
Glauber Costa55007d82012-12-18 14:22:38 -08003099 ssize_t size = memcg_caches_array_size(num_groups);
3100
3101 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003102 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003103
Vladimir Davydovf8570262014-01-23 15:53:06 -08003104 new_params = kzalloc(size, GFP_KERNEL);
3105 if (!new_params)
Glauber Costa55007d82012-12-18 14:22:38 -08003106 return -ENOMEM;
Glauber Costa55007d82012-12-18 14:22:38 -08003107
Vladimir Davydovf8570262014-01-23 15:53:06 -08003108 new_params->is_root_cache = true;
Glauber Costa55007d82012-12-18 14:22:38 -08003109
3110 /*
3111 * There is the chance it will be bigger than
3112 * memcg_limited_groups_array_size, if we failed an allocation
3113 * in a cache, in which case all caches updated before it, will
3114 * have a bigger array.
3115 *
3116 * But if that is the case, the data after
3117 * memcg_limited_groups_array_size is certainly unused
3118 */
3119 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3120 if (!cur_params->memcg_caches[i])
3121 continue;
Vladimir Davydovf8570262014-01-23 15:53:06 -08003122 new_params->memcg_caches[i] =
Glauber Costa55007d82012-12-18 14:22:38 -08003123 cur_params->memcg_caches[i];
3124 }
3125
3126 /*
3127 * Ideally, we would wait until all caches succeed, and only
3128 * then free the old one. But this is not worth the extra
3129 * pointer per-cache we'd have to have for this.
3130 *
3131 * It is not a big deal if some caches are left with a size
3132 * bigger than the others. And all updates will reset this
3133 * anyway.
3134 */
Vladimir Davydovf8570262014-01-23 15:53:06 -08003135 rcu_assign_pointer(s->memcg_params, new_params);
3136 if (cur_params)
3137 kfree_rcu(cur_params, rcu_head);
Glauber Costa55007d82012-12-18 14:22:38 -08003138 }
3139 return 0;
3140}
3141
Vladimir Davydov363a0442014-01-23 15:52:56 -08003142int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s,
3143 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003144{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003145 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003146
3147 if (!memcg_kmem_enabled())
3148 return 0;
3149
Andrey Vagin90c7a792013-09-11 14:22:18 -07003150 if (!memcg) {
3151 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003152 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003153 } else
3154 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003155
Glauber Costa2633d7a2012-12-18 14:22:34 -08003156 s->memcg_params = kzalloc(size, GFP_KERNEL);
3157 if (!s->memcg_params)
3158 return -ENOMEM;
3159
Glauber Costa943a4512012-12-18 14:23:03 -08003160 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003161 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003162 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003163 INIT_WORK(&s->memcg_params->destroy,
3164 kmem_cache_destroy_work_func);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003165 } else
3166 s->memcg_params->is_root_cache = true;
3167
Glauber Costa2633d7a2012-12-18 14:22:34 -08003168 return 0;
3169}
3170
Vladimir Davydov363a0442014-01-23 15:52:56 -08003171void memcg_free_cache_params(struct kmem_cache *s)
3172{
3173 kfree(s->memcg_params);
3174}
3175
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003176void memcg_register_cache(struct kmem_cache *s)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003177{
Glauber Costad7f25f82012-12-18 14:22:40 -08003178 struct kmem_cache *root;
3179 struct mem_cgroup *memcg;
3180 int id;
3181
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003182 if (is_root_cache(s))
Glauber Costad7f25f82012-12-18 14:22:40 -08003183 return;
3184
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003185 /*
3186 * Holding the slab_mutex assures nobody will touch the memcg_caches
3187 * array while we are modifying it.
3188 */
3189 lockdep_assert_held(&slab_mutex);
3190
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003191 root = s->memcg_params->root_cache;
3192 memcg = s->memcg_params->memcg;
3193 id = memcg_cache_id(memcg);
3194
3195 css_get(&memcg->css);
3196
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003197
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003198 /*
Vladimir Davydov959c8962014-01-23 15:52:59 -08003199 * Since readers won't lock (see cache_from_memcg_idx()), we need a
3200 * barrier here to ensure nobody will see the kmem_cache partially
3201 * initialized.
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003202 */
Vladimir Davydov959c8962014-01-23 15:52:59 -08003203 smp_wmb();
3204
Vladimir Davydov96403da2014-01-23 15:53:01 -08003205 /*
3206 * Initialize the pointer to this cache in its parent's memcg_params
3207 * before adding it to the memcg_slab_caches list, otherwise we can
3208 * fail to convert memcg_params_to_cache() while traversing the list.
3209 */
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003210 VM_BUG_ON(root->memcg_params->memcg_caches[id]);
Vladimir Davydov959c8962014-01-23 15:52:59 -08003211 root->memcg_params->memcg_caches[id] = s;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003212
3213 mutex_lock(&memcg->slab_caches_mutex);
3214 list_add(&s->memcg_params->list, &memcg->memcg_slab_caches);
3215 mutex_unlock(&memcg->slab_caches_mutex);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003216}
3217
3218void memcg_unregister_cache(struct kmem_cache *s)
3219{
3220 struct kmem_cache *root;
3221 struct mem_cgroup *memcg;
3222 int id;
3223
3224 if (is_root_cache(s))
3225 return;
Glauber Costad7f25f82012-12-18 14:22:40 -08003226
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003227 /*
3228 * Holding the slab_mutex assures nobody will touch the memcg_caches
3229 * array while we are modifying it.
3230 */
3231 lockdep_assert_held(&slab_mutex);
3232
Glauber Costad7f25f82012-12-18 14:22:40 -08003233 root = s->memcg_params->root_cache;
Vladimir Davydov96403da2014-01-23 15:53:01 -08003234 memcg = s->memcg_params->memcg;
3235 id = memcg_cache_id(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08003236
3237 mutex_lock(&memcg->slab_caches_mutex);
3238 list_del(&s->memcg_params->list);
3239 mutex_unlock(&memcg->slab_caches_mutex);
3240
Vladimir Davydov96403da2014-01-23 15:53:01 -08003241 /*
3242 * Clear the pointer to this cache in its parent's memcg_params only
3243 * after removing it from the memcg_slab_caches list, otherwise we can
3244 * fail to convert memcg_params_to_cache() while traversing the list.
3245 */
Vladimir Davydov2edefe12014-01-23 15:53:02 -08003246 VM_BUG_ON(!root->memcg_params->memcg_caches[id]);
Vladimir Davydov96403da2014-01-23 15:53:01 -08003247 root->memcg_params->memcg_caches[id] = NULL;
3248
Li Zefan20f05312013-07-08 16:00:31 -07003249 css_put(&memcg->css);
Glauber Costa2633d7a2012-12-18 14:22:34 -08003250}
3251
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003252/*
3253 * During the creation a new cache, we need to disable our accounting mechanism
3254 * altogether. This is true even if we are not creating, but rather just
3255 * enqueing new caches to be created.
3256 *
3257 * This is because that process will trigger allocations; some visible, like
3258 * explicit kmallocs to auxiliary data structures, name strings and internal
3259 * cache structures; some well concealed, like INIT_WORK() that can allocate
3260 * objects during debug.
3261 *
3262 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3263 * to it. This may not be a bounded recursion: since the first cache creation
3264 * failed to complete (waiting on the allocation), we'll just try to create the
3265 * cache again, failing at the same point.
3266 *
3267 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3268 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3269 * inside the following two functions.
3270 */
3271static inline void memcg_stop_kmem_account(void)
3272{
3273 VM_BUG_ON(!current->mm);
3274 current->memcg_kmem_skip_account++;
3275}
3276
3277static inline void memcg_resume_kmem_account(void)
3278{
3279 VM_BUG_ON(!current->mm);
3280 current->memcg_kmem_skip_account--;
3281}
3282
Glauber Costa1f458cb2012-12-18 14:22:50 -08003283static void kmem_cache_destroy_work_func(struct work_struct *w)
3284{
3285 struct kmem_cache *cachep;
3286 struct memcg_cache_params *p;
3287
3288 p = container_of(w, struct memcg_cache_params, destroy);
3289
3290 cachep = memcg_params_to_cache(p);
3291
Glauber Costa22933152012-12-18 14:22:59 -08003292 /*
3293 * If we get down to 0 after shrink, we could delete right away.
3294 * However, memcg_release_pages() already puts us back in the workqueue
3295 * in that case. If we proceed deleting, we'll get a dangling
3296 * reference, and removing the object from the workqueue in that case
3297 * is unnecessary complication. We are not a fast path.
3298 *
3299 * Note that this case is fundamentally different from racing with
3300 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3301 * kmem_cache_shrink, not only we would be reinserting a dead cache
3302 * into the queue, but doing so from inside the worker racing to
3303 * destroy it.
3304 *
3305 * So if we aren't down to zero, we'll just schedule a worker and try
3306 * again
3307 */
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003308 if (atomic_read(&cachep->memcg_params->nr_pages) != 0)
Glauber Costa22933152012-12-18 14:22:59 -08003309 kmem_cache_shrink(cachep);
Vladimir Davydov0d8a4a32014-01-23 15:53:39 -08003310 else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003311 kmem_cache_destroy(cachep);
3312}
3313
3314void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3315{
3316 if (!cachep->memcg_params->dead)
3317 return;
3318
3319 /*
Glauber Costa22933152012-12-18 14:22:59 -08003320 * There are many ways in which we can get here.
3321 *
3322 * We can get to a memory-pressure situation while the delayed work is
3323 * still pending to run. The vmscan shrinkers can then release all
3324 * cache memory and get us to destruction. If this is the case, we'll
3325 * be executed twice, which is a bug (the second time will execute over
3326 * bogus data). In this case, cancelling the work should be fine.
3327 *
3328 * But we can also get here from the worker itself, if
3329 * kmem_cache_shrink is enough to shake all the remaining objects and
3330 * get the page count to 0. In this case, we'll deadlock if we try to
3331 * cancel the work (the worker runs with an internal lock held, which
3332 * is the same lock we would hold for cancel_work_sync().)
3333 *
3334 * Since we can't possibly know who got us here, just refrain from
3335 * running if there is already work pending
3336 */
3337 if (work_pending(&cachep->memcg_params->destroy))
3338 return;
3339 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003340 * We have to defer the actual destroying to a workqueue, because
3341 * we might currently be in a context that cannot sleep.
3342 */
3343 schedule_work(&cachep->memcg_params->destroy);
3344}
3345
Vladimir Davydov842e2872014-01-23 15:53:03 -08003346static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3347 struct kmem_cache *s)
Michal Hockod9c10dd2013-03-28 08:48:14 +01003348{
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003349 struct kmem_cache *new = NULL;
Tejun Heoe61734c2014-02-12 09:29:50 -05003350 static char *tmp_path = NULL, *tmp_name = NULL;
Vladimir Davydov842e2872014-01-23 15:53:03 -08003351 static DEFINE_MUTEX(mutex); /* protects tmp_name */
Michal Hockod9c10dd2013-03-28 08:48:14 +01003352
Vladimir Davydov842e2872014-01-23 15:53:03 -08003353 BUG_ON(!memcg_can_account_kmem(memcg));
Michal Hockod9c10dd2013-03-28 08:48:14 +01003354
Vladimir Davydov842e2872014-01-23 15:53:03 -08003355 mutex_lock(&mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003356 /*
3357 * kmem_cache_create_memcg duplicates the given name and
3358 * cgroup_name for this name requires RCU context.
3359 * This static temporary buffer is used to prevent from
3360 * pointless shortliving allocation.
3361 */
Tejun Heoe61734c2014-02-12 09:29:50 -05003362 if (!tmp_path || !tmp_name) {
3363 if (!tmp_path)
3364 tmp_path = kmalloc(PATH_MAX, GFP_KERNEL);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003365 if (!tmp_name)
Tejun Heoe61734c2014-02-12 09:29:50 -05003366 tmp_name = kmalloc(NAME_MAX + 1, GFP_KERNEL);
3367 if (!tmp_path || !tmp_name)
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003368 goto out;
Michal Hockod9c10dd2013-03-28 08:48:14 +01003369 }
3370
Tejun Heoe61734c2014-02-12 09:29:50 -05003371 cgroup_name(memcg->css.cgroup, tmp_name, NAME_MAX + 1);
3372 snprintf(tmp_path, PATH_MAX, "%s(%d:%s)", s->name,
3373 memcg_cache_id(memcg), tmp_name);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003374
Tejun Heoe61734c2014-02-12 09:29:50 -05003375 new = kmem_cache_create_memcg(memcg, tmp_path, s->object_size, s->align,
Michal Hockod9c10dd2013-03-28 08:48:14 +01003376 (s->flags & ~SLAB_PANIC), s->ctor, s);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003377 if (new)
3378 new->allocflags |= __GFP_KMEMCG;
Vladimir Davydov842e2872014-01-23 15:53:03 -08003379 else
3380 new = s;
Vladimir Davydov7c094fd2014-01-30 15:46:14 -08003381out:
Vladimir Davydov842e2872014-01-23 15:53:03 -08003382 mutex_unlock(&mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003383 return new;
3384}
3385
Glauber Costa7cf27982012-12-18 14:22:55 -08003386void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3387{
3388 struct kmem_cache *c;
3389 int i;
3390
3391 if (!s->memcg_params)
3392 return;
3393 if (!s->memcg_params->is_root_cache)
3394 return;
3395
3396 /*
3397 * If the cache is being destroyed, we trust that there is no one else
3398 * requesting objects from it. Even if there are, the sanity checks in
3399 * kmem_cache_destroy should caught this ill-case.
3400 *
3401 * Still, we don't want anyone else freeing memcg_caches under our
3402 * noses, which can happen if a new memcg comes to life. As usual,
Vladimir Davydovd6441632014-01-23 15:53:09 -08003403 * we'll take the activate_kmem_mutex to protect ourselves against
3404 * this.
Glauber Costa7cf27982012-12-18 14:22:55 -08003405 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08003406 mutex_lock(&activate_kmem_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003407 for_each_memcg_cache_index(i) {
3408 c = cache_from_memcg_idx(s, i);
Glauber Costa7cf27982012-12-18 14:22:55 -08003409 if (!c)
3410 continue;
3411
3412 /*
3413 * We will now manually delete the caches, so to avoid races
3414 * we need to cancel all pending destruction workers and
3415 * proceed with destruction ourselves.
3416 *
3417 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3418 * and that could spawn the workers again: it is likely that
3419 * the cache still have active pages until this very moment.
3420 * This would lead us back to mem_cgroup_destroy_cache.
3421 *
3422 * But that will not execute at all if the "dead" flag is not
3423 * set, so flip it down to guarantee we are in control.
3424 */
3425 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003426 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003427 kmem_cache_destroy(c);
3428 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08003429 mutex_unlock(&activate_kmem_mutex);
Glauber Costa7cf27982012-12-18 14:22:55 -08003430}
3431
Glauber Costad7f25f82012-12-18 14:22:40 -08003432struct create_work {
3433 struct mem_cgroup *memcg;
3434 struct kmem_cache *cachep;
3435 struct work_struct work;
3436};
3437
Glauber Costa1f458cb2012-12-18 14:22:50 -08003438static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3439{
3440 struct kmem_cache *cachep;
3441 struct memcg_cache_params *params;
3442
3443 if (!memcg_kmem_is_active(memcg))
3444 return;
3445
3446 mutex_lock(&memcg->slab_caches_mutex);
3447 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3448 cachep = memcg_params_to_cache(params);
3449 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003450 schedule_work(&cachep->memcg_params->destroy);
3451 }
3452 mutex_unlock(&memcg->slab_caches_mutex);
3453}
3454
Glauber Costad7f25f82012-12-18 14:22:40 -08003455static void memcg_create_cache_work_func(struct work_struct *w)
3456{
3457 struct create_work *cw;
3458
3459 cw = container_of(w, struct create_work, work);
3460 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08003461 css_put(&cw->memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003462 kfree(cw);
3463}
3464
3465/*
3466 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003467 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003468static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3469 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003470{
3471 struct create_work *cw;
3472
3473 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003474 if (cw == NULL) {
3475 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003476 return;
3477 }
3478
3479 cw->memcg = memcg;
3480 cw->cachep = cachep;
3481
3482 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3483 schedule_work(&cw->work);
3484}
3485
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003486static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3487 struct kmem_cache *cachep)
3488{
3489 /*
3490 * We need to stop accounting when we kmalloc, because if the
3491 * corresponding kmalloc cache is not yet created, the first allocation
3492 * in __memcg_create_cache_enqueue will recurse.
3493 *
3494 * However, it is better to enclose the whole function. Depending on
3495 * the debugging options enabled, INIT_WORK(), for instance, can
3496 * trigger an allocation. This too, will make us recurse. Because at
3497 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3498 * the safest choice is to do it like this, wrapping the whole function.
3499 */
3500 memcg_stop_kmem_account();
3501 __memcg_create_cache_enqueue(memcg, cachep);
3502 memcg_resume_kmem_account();
3503}
Glauber Costad7f25f82012-12-18 14:22:40 -08003504/*
3505 * Return the kmem_cache we're supposed to use for a slab allocation.
3506 * We try to use the current memcg's version of the cache.
3507 *
3508 * If the cache does not exist yet, if we are the first user of it,
3509 * we either create it immediately, if possible, or create it asynchronously
3510 * in a workqueue.
3511 * In the latter case, we will let the current allocation go through with
3512 * the original cache.
3513 *
3514 * Can't be called in interrupt context or from kernel threads.
3515 * This function needs to be called with rcu_read_lock() held.
3516 */
3517struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3518 gfp_t gfp)
3519{
3520 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08003521 struct kmem_cache *memcg_cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003522
3523 VM_BUG_ON(!cachep->memcg_params);
3524 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3525
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003526 if (!current->mm || current->memcg_kmem_skip_account)
3527 return cachep;
3528
Glauber Costad7f25f82012-12-18 14:22:40 -08003529 rcu_read_lock();
3530 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003531
3532 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003533 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003534
Vladimir Davydov959c8962014-01-23 15:52:59 -08003535 memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg));
3536 if (likely(memcg_cachep)) {
3537 cachep = memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07003538 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003539 }
3540
Li Zefanca0dde92013-04-29 15:08:57 -07003541 /* The corresponding put will be done in the workqueue. */
3542 if (!css_tryget(&memcg->css))
3543 goto out;
3544 rcu_read_unlock();
3545
3546 /*
3547 * If we are in a safe context (can wait, and not in interrupt
3548 * context), we could be be predictable and return right away.
3549 * This would guarantee that the allocation being performed
3550 * already belongs in the new cache.
3551 *
3552 * However, there are some clashes that can arrive from locking.
3553 * For instance, because we acquire the slab_mutex while doing
3554 * kmem_cache_dup, this means no further allocation could happen
3555 * with the slab_mutex held.
3556 *
3557 * Also, because cache creation issue get_online_cpus(), this
3558 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3559 * that ends up reversed during cpu hotplug. (cpuset allocates
3560 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3561 * better to defer everything.
3562 */
3563 memcg_create_cache_enqueue(memcg, cachep);
3564 return cachep;
3565out:
3566 rcu_read_unlock();
3567 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003568}
3569EXPORT_SYMBOL(__memcg_kmem_get_cache);
3570
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003571/*
3572 * We need to verify if the allocation against current->mm->owner's memcg is
3573 * possible for the given order. But the page is not allocated yet, so we'll
3574 * need a further commit step to do the final arrangements.
3575 *
3576 * It is possible for the task to switch cgroups in this mean time, so at
3577 * commit time, we can't rely on task conversion any longer. We'll then use
3578 * the handle argument to return to the caller which cgroup we should commit
3579 * against. We could also return the memcg directly and avoid the pointer
3580 * passing, but a boolean return value gives better semantics considering
3581 * the compiled-out case as well.
3582 *
3583 * Returning true means the allocation is possible.
3584 */
3585bool
3586__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3587{
3588 struct mem_cgroup *memcg;
3589 int ret;
3590
3591 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003592
3593 /*
3594 * Disabling accounting is only relevant for some specific memcg
3595 * internal allocations. Therefore we would initially not have such
3596 * check here, since direct calls to the page allocator that are marked
3597 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3598 * concerned with cache allocations, and by having this test at
3599 * memcg_kmem_get_cache, we are already able to relay the allocation to
3600 * the root cache and bypass the memcg cache altogether.
3601 *
3602 * There is one exception, though: the SLUB allocator does not create
3603 * large order caches, but rather service large kmallocs directly from
3604 * the page allocator. Therefore, the following sequence when backed by
3605 * the SLUB allocator:
3606 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003607 * memcg_stop_kmem_account();
3608 * kmalloc(<large_number>)
3609 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003610 *
3611 * would effectively ignore the fact that we should skip accounting,
3612 * since it will drive us directly to this function without passing
3613 * through the cache selector memcg_kmem_get_cache. Such large
3614 * allocations are extremely rare but can happen, for instance, for the
3615 * cache arrays. We bring this test here.
3616 */
3617 if (!current->mm || current->memcg_kmem_skip_account)
3618 return true;
3619
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003620 memcg = try_get_mem_cgroup_from_mm(current->mm);
3621
3622 /*
3623 * very rare case described in mem_cgroup_from_task. Unfortunately there
3624 * isn't much we can do without complicating this too much, and it would
3625 * be gfp-dependent anyway. Just let it go
3626 */
3627 if (unlikely(!memcg))
3628 return true;
3629
3630 if (!memcg_can_account_kmem(memcg)) {
3631 css_put(&memcg->css);
3632 return true;
3633 }
3634
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003635 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3636 if (!ret)
3637 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003638
3639 css_put(&memcg->css);
3640 return (ret == 0);
3641}
3642
3643void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3644 int order)
3645{
3646 struct page_cgroup *pc;
3647
3648 VM_BUG_ON(mem_cgroup_is_root(memcg));
3649
3650 /* The page allocation failed. Revert */
3651 if (!page) {
3652 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003653 return;
3654 }
3655
3656 pc = lookup_page_cgroup(page);
3657 lock_page_cgroup(pc);
3658 pc->mem_cgroup = memcg;
3659 SetPageCgroupUsed(pc);
3660 unlock_page_cgroup(pc);
3661}
3662
3663void __memcg_kmem_uncharge_pages(struct page *page, int order)
3664{
3665 struct mem_cgroup *memcg = NULL;
3666 struct page_cgroup *pc;
3667
3668
3669 pc = lookup_page_cgroup(page);
3670 /*
3671 * Fast unlocked return. Theoretically might have changed, have to
3672 * check again after locking.
3673 */
3674 if (!PageCgroupUsed(pc))
3675 return;
3676
3677 lock_page_cgroup(pc);
3678 if (PageCgroupUsed(pc)) {
3679 memcg = pc->mem_cgroup;
3680 ClearPageCgroupUsed(pc);
3681 }
3682 unlock_page_cgroup(pc);
3683
3684 /*
3685 * We trust that only if there is a memcg associated with the page, it
3686 * is a valid allocation
3687 */
3688 if (!memcg)
3689 return;
3690
Sasha Levin309381fea2014-01-23 15:52:54 -08003691 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003692 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003693}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003694#else
3695static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3696{
3697}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003698#endif /* CONFIG_MEMCG_KMEM */
3699
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003700#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3701
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003702#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003703/*
3704 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003705 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3706 * charge/uncharge will be never happen and move_account() is done under
3707 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003708 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003709void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003710{
3711 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003712 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003713 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003714 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003715
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003716 if (mem_cgroup_disabled())
3717 return;
David Rientjesb070e652013-05-07 16:18:09 -07003718
3719 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003720 for (i = 1; i < HPAGE_PMD_NR; i++) {
3721 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003722 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003723 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003724 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3725 }
David Rientjesb070e652013-05-07 16:18:09 -07003726 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3727 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003728}
Hugh Dickins12d27102012-01-12 17:19:52 -08003729#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003730
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003731/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003732 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003733 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003734 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003735 * @pc: page_cgroup of the page.
3736 * @from: mem_cgroup which the page is moved from.
3737 * @to: mem_cgroup which the page is moved to. @from != @to.
3738 *
3739 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003740 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003741 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003742 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003743 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3744 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003745 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003746static int mem_cgroup_move_account(struct page *page,
3747 unsigned int nr_pages,
3748 struct page_cgroup *pc,
3749 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003750 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003751{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003752 unsigned long flags;
3753 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003754 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003755
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003756 VM_BUG_ON(from == to);
Sasha Levin309381fea2014-01-23 15:52:54 -08003757 VM_BUG_ON_PAGE(PageLRU(page), page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003758 /*
3759 * The page is isolated from LRU. So, collapse function
3760 * will not handle this page. But page splitting can happen.
3761 * Do this check under compound_page_lock(). The caller should
3762 * hold it.
3763 */
3764 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003765 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003766 goto out;
3767
3768 lock_page_cgroup(pc);
3769
3770 ret = -EINVAL;
3771 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3772 goto unlock;
3773
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003774 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003775
Johannes Weiner59d1d252014-04-07 15:37:40 -07003776 if (!anon && page_mapped(page)) {
3777 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3778 nr_pages);
3779 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3780 nr_pages);
3781 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003782
Johannes Weiner59d1d252014-04-07 15:37:40 -07003783 if (PageWriteback(page)) {
3784 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3785 nr_pages);
3786 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3787 nr_pages);
3788 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003789
David Rientjesb070e652013-05-07 16:18:09 -07003790 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003791
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003792 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003793 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003794 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003795 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003796 ret = 0;
3797unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003798 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003799 /*
3800 * check events
3801 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003802 memcg_check_events(to, page);
3803 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003804out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003805 return ret;
3806}
3807
Michal Hocko2ef37d32012-10-26 13:37:30 +02003808/**
3809 * mem_cgroup_move_parent - moves page to the parent group
3810 * @page: the page to move
3811 * @pc: page_cgroup of the page
3812 * @child: page's cgroup
3813 *
3814 * move charges to its parent or the root cgroup if the group has no
3815 * parent (aka use_hierarchy==0).
3816 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3817 * mem_cgroup_move_account fails) the failure is always temporary and
3818 * it signals a race with a page removal/uncharge or migration. In the
3819 * first case the page is on the way out and it will vanish from the LRU
3820 * on the next attempt and the call should be retried later.
3821 * Isolation from the LRU fails only if page has been isolated from
3822 * the LRU since we looked at it and that usually means either global
3823 * reclaim or migration going on. The page will either get back to the
3824 * LRU or vanish.
3825 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3826 * (!PageCgroupUsed) or moved to a different group. The page will
3827 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003828 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003829static int mem_cgroup_move_parent(struct page *page,
3830 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003831 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003832{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003833 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003834 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003835 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003836 int ret;
3837
Michal Hockod8423012012-10-26 13:37:29 +02003838 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003839
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003840 ret = -EBUSY;
3841 if (!get_page_unless_zero(page))
3842 goto out;
3843 if (isolate_lru_page(page))
3844 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003845
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003846 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003847
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003848 parent = parent_mem_cgroup(child);
3849 /*
3850 * If no parent, move charges to root cgroup.
3851 */
3852 if (!parent)
3853 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003854
Michal Hocko2ef37d32012-10-26 13:37:30 +02003855 if (nr_pages > 1) {
Sasha Levin309381fea2014-01-23 15:52:54 -08003856 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003857 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003858 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003859
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003860 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003861 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003862 if (!ret)
3863 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003864
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003865 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003866 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003867 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003868put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003869 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003870out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003871 return ret;
3872}
3873
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003874int mem_cgroup_newpage_charge(struct page *page,
3875 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003876{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003877 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003878 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003879 bool oom = true;
3880 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003881
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003882 if (mem_cgroup_disabled())
3883 return 0;
3884
3885 VM_BUG_ON_PAGE(page_mapped(page), page);
3886 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page);
3887 VM_BUG_ON(!mm);
3888
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003889 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003890 nr_pages <<= compound_order(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08003891 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Johannes Weiner8493ae42011-02-01 15:52:44 -08003892 /*
3893 * Never OOM-kill a process for a huge page. The
3894 * fault handler will fall back to regular pages.
3895 */
3896 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003897 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003898
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003899 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003900 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003901 return ret;
Johannes Weiner1bec6b32014-04-07 15:37:41 -07003902 __mem_cgroup_commit_charge(memcg, page, nr_pages,
3903 MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003904 return 0;
3905}
3906
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003907/*
3908 * While swap-in, try_charge -> commit or cancel, the page is locked.
3909 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003910 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003911 * "commit()" or removed by "cancel()"
3912 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003913static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3914 struct page *page,
3915 gfp_t mask,
3916 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003917{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003918 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003919 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003920 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003921
Johannes Weiner90deb782012-07-31 16:45:47 -07003922 pc = lookup_page_cgroup(page);
3923 /*
3924 * Every swap fault against a single page tries to charge the
3925 * page, bail as early as possible. shmem_unuse() encounters
3926 * already charged pages, too. The USED bit is protected by
3927 * the page lock, which serializes swap cache removal, which
3928 * in turn serializes uncharging.
3929 */
3930 if (PageCgroupUsed(pc))
3931 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003932 if (!do_swap_account)
3933 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003934 memcg = try_get_mem_cgroup_from_page(page);
3935 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003936 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08003937 *memcgp = memcg;
3938 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003939 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003940 if (ret == -EINTR)
3941 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003942 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003943charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003944 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
3945 if (ret == -EINTR)
3946 ret = 0;
3947 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003948}
3949
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003950int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
3951 gfp_t gfp_mask, struct mem_cgroup **memcgp)
3952{
3953 *memcgp = NULL;
3954 if (mem_cgroup_disabled())
3955 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003956 /*
3957 * A racing thread's fault, or swapoff, may have already
3958 * updated the pte, and even removed page from swap cache: in
3959 * those cases unuse_pte()'s pte_same() test will fail; but
3960 * there's also a KSM case which does need to charge the page.
3961 */
3962 if (!PageSwapCache(page)) {
3963 int ret;
3964
3965 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
3966 if (ret == -EINTR)
3967 ret = 0;
3968 return ret;
3969 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003970 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
3971}
3972
Johannes Weiner827a03d2012-07-31 16:45:36 -07003973void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
3974{
3975 if (mem_cgroup_disabled())
3976 return;
3977 if (!memcg)
3978 return;
3979 __mem_cgroup_cancel_charge(memcg, 1);
3980}
3981
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003982static void
Johannes Weiner72835c82012-01-12 17:18:32 -08003983__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003984 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003985{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003986 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003987 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08003988 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003989 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07003990
Johannes Weinerce587e62012-04-24 20:22:33 +02003991 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003992 /*
3993 * Now swap is on-memory. This means this page may be
3994 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003995 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
3996 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
3997 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003998 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003999 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004000 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004001 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004002 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004003}
4004
Johannes Weiner72835c82012-01-12 17:18:32 -08004005void mem_cgroup_commit_charge_swapin(struct page *page,
4006 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004007{
Johannes Weiner72835c82012-01-12 17:18:32 -08004008 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004009 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004010}
4011
Johannes Weiner827a03d2012-07-31 16:45:36 -07004012int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4013 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004014{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004015 struct mem_cgroup *memcg = NULL;
4016 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4017 int ret;
4018
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004019 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004020 return 0;
4021 if (PageCompound(page))
4022 return 0;
4023
Johannes Weiner1bec6b32014-04-07 15:37:41 -07004024 if (!PageSwapCache(page)) {
Johannes Weiner284f39a2014-04-07 15:37:41 -07004025 /*
4026 * Page cache insertions can happen without an actual
4027 * task context, e.g. during disk probing on boot.
4028 */
4029 if (!mm)
4030 memcg = root_mem_cgroup;
Johannes Weiner1bec6b32014-04-07 15:37:41 -07004031 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &memcg, true);
4032 if (ret != -ENOMEM)
4033 __mem_cgroup_commit_charge(memcg, page, 1, type, false);
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);
Sasha Levin309381fea2014-01-23 15:52:54 -08004113 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004114 }
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
Li Zefan40503772013-07-08 16:00:34 -07004178 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004179 */
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);
Li Zefan40503772013-07-08 16:00:34 -07004183 css_get(&memcg->css);
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;
Sasha Levin309381fea2014-01-23 15:52:54 -08004205 VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), 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{
Sasha Levin309381fea2014-01-23 15:52:54 -08004225 VM_BUG_ON_PAGE(page_mapped(page), page);
4226 VM_BUG_ON_PAGE(page->mapping, page);
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,
Li Zefan40503772013-07-08 16:00:34 -07004295 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004296 */
4297 if (do_swap_account && swapout && memcg)
Li Zefan34c00c32013-09-23 16:56:01 +08004298 swap_cgroup_record(ent, mem_cgroup_id(memcg));
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);
Li Zefan40503772013-07-08 16:00:34 -07004326 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004327 }
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
Li Zefan34c00c32013-09-23 16:56:01 +08004350 old_id = mem_cgroup_id(from);
4351 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004352
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
Li Zefan40503772013-07-08 16:00:34 -07004360 * improvement. But we cannot postpone css_get(to) because if
4361 * the process that has been moved to @to does swap-in, the
4362 * refcount of @to might be decreased to 0.
4363 *
4364 * We are in attach() phase, so the cgroup is guaranteed to be
4365 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004366 */
Li Zefan40503772013-07-08 16:00:34 -07004367 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004368 return 0;
4369 }
4370 return -EINVAL;
4371}
4372#else
4373static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004374 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004375{
4376 return -EINVAL;
4377}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004378#endif
4379
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004380/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004381 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4382 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004383 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004384void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4385 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004386{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004387 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004388 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004389 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004390 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004391
Johannes Weiner72835c82012-01-12 17:18:32 -08004392 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004393
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004394 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004395 return;
Balbir Singh40779602008-04-04 14:29:59 -07004396
Mel Gormanb32967f2012-11-19 12:35:47 +00004397 if (PageTransHuge(page))
4398 nr_pages <<= compound_order(page);
4399
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004400 pc = lookup_page_cgroup(page);
4401 lock_page_cgroup(pc);
4402 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004403 memcg = pc->mem_cgroup;
4404 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004405 /*
4406 * At migrating an anonymous page, its mapcount goes down
4407 * to 0 and uncharge() will be called. But, even if it's fully
4408 * unmapped, migration may fail and this page has to be
4409 * charged again. We set MIGRATION flag here and delay uncharge
4410 * until end_migration() is called
4411 *
4412 * Corner Case Thinking
4413 * A)
4414 * When the old page was mapped as Anon and it's unmap-and-freed
4415 * while migration was ongoing.
4416 * If unmap finds the old page, uncharge() of it will be delayed
4417 * until end_migration(). If unmap finds a new page, it's
4418 * uncharged when it make mapcount to be 1->0. If unmap code
4419 * finds swap_migration_entry, the new page will not be mapped
4420 * and end_migration() will find it(mapcount==0).
4421 *
4422 * B)
4423 * When the old page was mapped but migraion fails, the kernel
4424 * remaps it. A charge for it is kept by MIGRATION flag even
4425 * if mapcount goes down to 0. We can do remap successfully
4426 * without charging it again.
4427 *
4428 * C)
4429 * The "old" page is under lock_page() until the end of
4430 * migration, so, the old page itself will not be swapped-out.
4431 * If the new page is swapped out before end_migraton, our
4432 * hook to usual swap-out path will catch the event.
4433 */
4434 if (PageAnon(page))
4435 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004436 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004437 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004438 /*
4439 * If the page is not charged at this point,
4440 * we return here.
4441 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004442 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004443 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004444
Johannes Weiner72835c82012-01-12 17:18:32 -08004445 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004446 /*
4447 * We charge new page before it's used/mapped. So, even if unlock_page()
4448 * is called before end_migration, we can catch all events on this new
4449 * page. In the case new page is migrated but not remapped, new page's
4450 * mapcount will be finally 0 and we call uncharge in end_migration().
4451 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004452 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004453 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004454 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004455 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004456 /*
4457 * The page is committed to the memcg, but it's not actually
4458 * charged to the res_counter since we plan on replacing the
4459 * old one and only one page is going to be left afterwards.
4460 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004461 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004462}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004463
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004464/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004465void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004466 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004467{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004468 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004469 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004470 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004471
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004472 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004473 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004474
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004475 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004476 used = oldpage;
4477 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004478 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004479 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004480 unused = oldpage;
4481 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004482 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004483 __mem_cgroup_uncharge_common(unused,
4484 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4485 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4486 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004487 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004488 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004489 * We disallowed uncharge of pages under migration because mapcount
4490 * of the page goes down to zero, temporarly.
4491 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004492 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004493 pc = lookup_page_cgroup(oldpage);
4494 lock_page_cgroup(pc);
4495 ClearPageCgroupMigration(pc);
4496 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004497
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004498 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004499 * If a page is a file cache, radix-tree replacement is very atomic
4500 * and we can skip this check. When it was an Anon page, its mapcount
4501 * goes down to 0. But because we added MIGRATION flage, it's not
4502 * uncharged yet. There are several case but page->mapcount check
4503 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4504 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004505 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004506 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004507 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004508}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004509
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004510/*
4511 * At replace page cache, newpage is not under any memcg but it's on
4512 * LRU. So, this function doesn't touch res_counter but handles LRU
4513 * in correct way. Both pages are locked so we cannot race with uncharge.
4514 */
4515void mem_cgroup_replace_page_cache(struct page *oldpage,
4516 struct page *newpage)
4517{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004518 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004519 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004520 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004521
4522 if (mem_cgroup_disabled())
4523 return;
4524
4525 pc = lookup_page_cgroup(oldpage);
4526 /* fix accounting on old pages */
4527 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004528 if (PageCgroupUsed(pc)) {
4529 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004530 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004531 ClearPageCgroupUsed(pc);
4532 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004533 unlock_page_cgroup(pc);
4534
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004535 /*
4536 * When called from shmem_replace_page(), in some cases the
4537 * oldpage has already been charged, and in some cases not.
4538 */
4539 if (!memcg)
4540 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004541 /*
4542 * Even if newpage->mapping was NULL before starting replacement,
4543 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4544 * LRU while we overwrite pc->mem_cgroup.
4545 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004546 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004547}
4548
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004549#ifdef CONFIG_DEBUG_VM
4550static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4551{
4552 struct page_cgroup *pc;
4553
4554 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004555 /*
4556 * Can be NULL while feeding pages into the page allocator for
4557 * the first time, i.e. during boot or memory hotplug;
4558 * or when mem_cgroup_disabled().
4559 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004560 if (likely(pc) && PageCgroupUsed(pc))
4561 return pc;
4562 return NULL;
4563}
4564
4565bool mem_cgroup_bad_page_check(struct page *page)
4566{
4567 if (mem_cgroup_disabled())
4568 return false;
4569
4570 return lookup_page_cgroup_used(page) != NULL;
4571}
4572
4573void mem_cgroup_print_bad_page(struct page *page)
4574{
4575 struct page_cgroup *pc;
4576
4577 pc = lookup_page_cgroup_used(page);
4578 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004579 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4580 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004581 }
4582}
4583#endif
4584
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004585static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004586 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004587{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004588 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004589 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004590 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004591 int children = mem_cgroup_count_children(memcg);
4592 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004593 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004594
4595 /*
4596 * For keeping hierarchical_reclaim simple, how long we should retry
4597 * is depends on callers. We set our retry-count to be function
4598 * of # of children which we should visit in this loop.
4599 */
4600 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4601
4602 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004603
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004604 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004605 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004606 if (signal_pending(current)) {
4607 ret = -EINTR;
4608 break;
4609 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004610 /*
4611 * Rather than hide all in some function, I do this in
4612 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004613 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004614 */
4615 mutex_lock(&set_limit_mutex);
4616 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4617 if (memswlimit < val) {
4618 ret = -EINVAL;
4619 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004620 break;
4621 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004622
4623 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4624 if (memlimit < val)
4625 enlarge = 1;
4626
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004627 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004628 if (!ret) {
4629 if (memswlimit == val)
4630 memcg->memsw_is_minimum = true;
4631 else
4632 memcg->memsw_is_minimum = false;
4633 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004634 mutex_unlock(&set_limit_mutex);
4635
4636 if (!ret)
4637 break;
4638
Johannes Weiner56600482012-01-12 17:17:59 -08004639 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4640 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004641 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4642 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004643 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004644 retry_count--;
4645 else
4646 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004647 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004648 if (!ret && enlarge)
4649 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004650
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004651 return ret;
4652}
4653
Li Zefan338c8432009-06-17 16:27:15 -07004654static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4655 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004656{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004657 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004658 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004659 int children = mem_cgroup_count_children(memcg);
4660 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004661 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004662
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004663 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004664 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004665 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004666 while (retry_count) {
4667 if (signal_pending(current)) {
4668 ret = -EINTR;
4669 break;
4670 }
4671 /*
4672 * Rather than hide all in some function, I do this in
4673 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004674 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004675 */
4676 mutex_lock(&set_limit_mutex);
4677 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4678 if (memlimit > val) {
4679 ret = -EINVAL;
4680 mutex_unlock(&set_limit_mutex);
4681 break;
4682 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004683 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4684 if (memswlimit < val)
4685 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004686 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004687 if (!ret) {
4688 if (memlimit == val)
4689 memcg->memsw_is_minimum = true;
4690 else
4691 memcg->memsw_is_minimum = false;
4692 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004693 mutex_unlock(&set_limit_mutex);
4694
4695 if (!ret)
4696 break;
4697
Johannes Weiner56600482012-01-12 17:17:59 -08004698 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4699 MEM_CGROUP_RECLAIM_NOSWAP |
4700 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004701 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004702 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004703 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004704 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004705 else
4706 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004707 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004708 if (!ret && enlarge)
4709 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004710 return ret;
4711}
4712
Andrew Morton0608f432013-09-24 15:27:41 -07004713unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
4714 gfp_t gfp_mask,
4715 unsigned long *total_scanned)
4716{
4717 unsigned long nr_reclaimed = 0;
4718 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4719 unsigned long reclaimed;
4720 int loop = 0;
4721 struct mem_cgroup_tree_per_zone *mctz;
4722 unsigned long long excess;
4723 unsigned long nr_scanned;
4724
4725 if (order > 0)
4726 return 0;
4727
4728 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
4729 /*
4730 * This loop can run a while, specially if mem_cgroup's continuously
4731 * keep exceeding their soft limit and putting the system under
4732 * pressure
4733 */
4734 do {
4735 if (next_mz)
4736 mz = next_mz;
4737 else
4738 mz = mem_cgroup_largest_soft_limit_node(mctz);
4739 if (!mz)
4740 break;
4741
4742 nr_scanned = 0;
4743 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
4744 gfp_mask, &nr_scanned);
4745 nr_reclaimed += reclaimed;
4746 *total_scanned += nr_scanned;
4747 spin_lock(&mctz->lock);
4748
4749 /*
4750 * If we failed to reclaim anything from this memory cgroup
4751 * it is time to move on to the next cgroup
4752 */
4753 next_mz = NULL;
4754 if (!reclaimed) {
4755 do {
4756 /*
4757 * Loop until we find yet another one.
4758 *
4759 * By the time we get the soft_limit lock
4760 * again, someone might have aded the
4761 * group back on the RB tree. Iterate to
4762 * make sure we get a different mem.
4763 * mem_cgroup_largest_soft_limit_node returns
4764 * NULL if no other cgroup is present on
4765 * the tree
4766 */
4767 next_mz =
4768 __mem_cgroup_largest_soft_limit_node(mctz);
4769 if (next_mz == mz)
4770 css_put(&next_mz->memcg->css);
4771 else /* next_mz == NULL or other memcg */
4772 break;
4773 } while (1);
4774 }
4775 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4776 excess = res_counter_soft_limit_excess(&mz->memcg->res);
4777 /*
4778 * One school of thought says that we should not add
4779 * back the node to the tree if reclaim returns 0.
4780 * But our reclaim could return 0, simply because due
4781 * to priority we are exposing a smaller subset of
4782 * memory to reclaim from. Consider this as a longer
4783 * term TODO.
4784 */
4785 /* If excess == 0, no tree ops */
4786 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
4787 spin_unlock(&mctz->lock);
4788 css_put(&mz->memcg->css);
4789 loop++;
4790 /*
4791 * Could not reclaim anything and there are no more
4792 * mem cgroups to try or we seem to be looping without
4793 * reclaiming anything.
4794 */
4795 if (!nr_reclaimed &&
4796 (next_mz == NULL ||
4797 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4798 break;
4799 } while (!nr_reclaimed);
4800 if (next_mz)
4801 css_put(&next_mz->memcg->css);
4802 return nr_reclaimed;
4803}
4804
Michal Hocko2ef37d32012-10-26 13:37:30 +02004805/**
4806 * mem_cgroup_force_empty_list - clears LRU of a group
4807 * @memcg: group to clear
4808 * @node: NUMA node
4809 * @zid: zone id
4810 * @lru: lru to to clear
4811 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004812 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004813 * reclaim the pages page themselves - pages are moved to the parent (or root)
4814 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004815 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004816static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004817 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004818{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004819 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004820 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004821 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004822 struct page *busy;
4823 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004824
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004825 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004826 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4827 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004828
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004829 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004830 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004831 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004832 struct page *page;
4833
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004834 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004835 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004836 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004837 break;
4838 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004839 page = list_entry(list->prev, struct page, lru);
4840 if (busy == page) {
4841 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004842 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004843 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004844 continue;
4845 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004846 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004847
Johannes Weiner925b7672012-01-12 17:18:15 -08004848 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004849
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004850 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004851 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004852 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004853 cond_resched();
4854 } else
4855 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004856 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004857}
4858
4859/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004860 * make mem_cgroup's charge to be 0 if there is no task by moving
4861 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004862 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004863 *
4864 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004865 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004866static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004867{
Michal Hockoc26251f2012-10-26 13:37:28 +02004868 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004869 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004870
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004871 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004872 /* This is for making all *used* pages to be on LRU. */
4873 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004874 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004875 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004876 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004877 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004878 enum lru_list lru;
4879 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004880 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004881 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004882 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004883 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004884 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004885 mem_cgroup_end_move(memcg);
4886 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004887 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004888
Michal Hocko2ef37d32012-10-26 13:37:30 +02004889 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004890 * Kernel memory may not necessarily be trackable to a specific
4891 * process. So they are not migrated, and therefore we can't
4892 * expect their value to drop to 0 here.
4893 * Having res filled up with kmem only is enough.
4894 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004895 * This is a safety check because mem_cgroup_force_empty_list
4896 * could have raced with mem_cgroup_replace_page_cache callers
4897 * so the lru seemed empty but the page could have been added
4898 * right after the check. RES_USAGE should be safe as we always
4899 * charge before adding to the LRU.
4900 */
Glauber Costabea207c2012-12-18 14:22:11 -08004901 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4902 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4903 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004904}
4905
Glauber Costab5f99b52013-02-22 16:34:53 -08004906static inline bool memcg_has_children(struct mem_cgroup *memcg)
4907{
Johannes Weiner696ac172013-10-31 16:34:15 -07004908 lockdep_assert_held(&memcg_create_mutex);
4909 /*
4910 * The lock does not prevent addition or deletion to the list
4911 * of children, but it prevents a new child from being
4912 * initialized based on this parent in css_online(), so it's
4913 * enough to decide whether hierarchically inherited
4914 * attributes can still be changed or not.
4915 */
4916 return memcg->use_hierarchy &&
4917 !list_empty(&memcg->css.cgroup->children);
Glauber Costab5f99b52013-02-22 16:34:53 -08004918}
4919
4920/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004921 * Reclaims as many pages from the given memcg as possible and moves
4922 * the rest to the parent.
4923 *
4924 * Caller is responsible for holding css reference for memcg.
4925 */
4926static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4927{
4928 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4929 struct cgroup *cgrp = memcg->css.cgroup;
4930
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004931 /* returns EBUSY if there is a task or if we come here twice. */
Tejun Heo07bc3562014-02-13 06:58:39 -05004932 if (cgroup_has_tasks(cgrp) || !list_empty(&cgrp->children))
Michal Hockoc26251f2012-10-26 13:37:28 +02004933 return -EBUSY;
4934
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004935 /* we call try-to-free pages for make this cgroup empty */
4936 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004937 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004938 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004939 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004940
Michal Hockoc26251f2012-10-26 13:37:28 +02004941 if (signal_pending(current))
4942 return -EINTR;
4943
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004944 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07004945 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004946 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004947 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004948 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02004949 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004950 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004951
4952 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004953 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02004954 mem_cgroup_reparent_charges(memcg);
4955
4956 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004957}
4958
Tejun Heo182446d2013-08-08 20:11:24 -04004959static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
4960 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004961{
Tejun Heo182446d2013-08-08 20:11:24 -04004962 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02004963
Michal Hockod8423012012-10-26 13:37:29 +02004964 if (mem_cgroup_is_root(memcg))
4965 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07004966 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004967}
4968
Tejun Heo182446d2013-08-08 20:11:24 -04004969static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
4970 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004971{
Tejun Heo182446d2013-08-08 20:11:24 -04004972 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004973}
4974
Tejun Heo182446d2013-08-08 20:11:24 -04004975static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
4976 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004977{
4978 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04004979 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04004980 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08004981
Glauber Costa09998212013-02-22 16:34:55 -08004982 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07004983
4984 if (memcg->use_hierarchy == val)
4985 goto out;
4986
Balbir Singh18f59ea2009-01-07 18:08:07 -08004987 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02004988 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08004989 * in the child subtrees. If it is unset, then the change can
4990 * occur, provided the current cgroup has no children.
4991 *
4992 * For the root cgroup, parent_mem is NULL, we allow value to be
4993 * set if there are no children.
4994 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004995 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08004996 (val == 1 || val == 0)) {
Johannes Weiner696ac172013-10-31 16:34:15 -07004997 if (list_empty(&memcg->css.cgroup->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004998 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004999 else
5000 retval = -EBUSY;
5001 } else
5002 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005003
5004out:
Glauber Costa09998212013-02-22 16:34:55 -08005005 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005006
5007 return retval;
5008}
5009
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005010
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005011static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005012 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005013{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005014 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005015 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005016
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005017 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005018 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005019 val += mem_cgroup_read_stat(iter, idx);
5020
5021 if (val < 0) /* race ? */
5022 val = 0;
5023 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005024}
5025
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005026static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005027{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005028 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005029
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005030 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005031 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005032 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005033 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005034 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005035 }
5036
David Rientjesb070e652013-05-07 16:18:09 -07005037 /*
5038 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5039 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5040 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005041 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5042 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005043
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005044 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005045 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005046
5047 return val << PAGE_SHIFT;
5048}
5049
Tejun Heo791badb2013-12-05 12:28:02 -05005050static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
5051 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005052{
Tejun Heo182446d2013-08-08 20:11:24 -04005053 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005054 u64 val;
Tejun Heo791badb2013-12-05 12:28:02 -05005055 int name;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005056 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005057
5058 type = MEMFILE_TYPE(cft->private);
5059 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005060
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005061 switch (type) {
5062 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005063 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005064 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005065 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005066 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005067 break;
5068 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005069 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005070 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005071 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005073 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005074 case _KMEM:
5075 val = res_counter_read_u64(&memcg->kmem, name);
5076 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005077 default:
5078 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005079 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005080
Tejun Heo791badb2013-12-05 12:28:02 -05005081 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005082}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005083
Glauber Costa510fc4e2012-12-18 14:21:47 -08005084#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08005085/* should be called with activate_kmem_mutex held */
5086static int __memcg_activate_kmem(struct mem_cgroup *memcg,
5087 unsigned long long limit)
5088{
5089 int err = 0;
5090 int memcg_id;
5091
5092 if (memcg_kmem_is_active(memcg))
5093 return 0;
5094
5095 /*
5096 * We are going to allocate memory for data shared by all memory
5097 * cgroups so let's stop accounting here.
5098 */
5099 memcg_stop_kmem_account();
5100
Glauber Costa510fc4e2012-12-18 14:21:47 -08005101 /*
5102 * For simplicity, we won't allow this to be disabled. It also can't
5103 * be changed if the cgroup has children already, or if tasks had
5104 * already joined.
5105 *
5106 * If tasks join before we set the limit, a person looking at
5107 * kmem.usage_in_bytes will have no way to determine when it took
5108 * place, which makes the value quite meaningless.
5109 *
5110 * After it first became limited, changes in the value of the limit are
5111 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005112 */
Glauber Costa09998212013-02-22 16:34:55 -08005113 mutex_lock(&memcg_create_mutex);
Tejun Heo07bc3562014-02-13 06:58:39 -05005114 if (cgroup_has_tasks(memcg->css.cgroup) || memcg_has_children(memcg))
Vladimir Davydovd6441632014-01-23 15:53:09 -08005115 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08005116 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005117 if (err)
5118 goto out;
5119
5120 memcg_id = ida_simple_get(&kmem_limited_groups,
5121 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
5122 if (memcg_id < 0) {
5123 err = memcg_id;
5124 goto out;
5125 }
5126
5127 /*
5128 * Make sure we have enough space for this cgroup in each root cache's
5129 * memcg_params.
5130 */
5131 err = memcg_update_all_caches(memcg_id + 1);
5132 if (err)
5133 goto out_rmid;
5134
5135 memcg->kmemcg_id = memcg_id;
5136 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
5137 mutex_init(&memcg->slab_caches_mutex);
5138
5139 /*
5140 * We couldn't have accounted to this cgroup, because it hasn't got the
5141 * active bit set yet, so this should succeed.
5142 */
5143 err = res_counter_set_limit(&memcg->kmem, limit);
5144 VM_BUG_ON(err);
5145
5146 static_key_slow_inc(&memcg_kmem_enabled_key);
5147 /*
5148 * Setting the active bit after enabling static branching will
5149 * guarantee no one starts accounting before all call sites are
5150 * patched.
5151 */
5152 memcg_kmem_set_active(memcg);
5153out:
5154 memcg_resume_kmem_account();
5155 return err;
5156
5157out_rmid:
5158 ida_simple_remove(&kmem_limited_groups, memcg_id);
5159 goto out;
5160}
5161
5162static int memcg_activate_kmem(struct mem_cgroup *memcg,
5163 unsigned long long limit)
5164{
5165 int ret;
5166
5167 mutex_lock(&activate_kmem_mutex);
5168 ret = __memcg_activate_kmem(memcg, limit);
5169 mutex_unlock(&activate_kmem_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005170 return ret;
5171}
5172
Vladimir Davydovd6441632014-01-23 15:53:09 -08005173static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5174 unsigned long long val)
5175{
5176 int ret;
5177
5178 if (!memcg_kmem_is_active(memcg))
5179 ret = memcg_activate_kmem(memcg, val);
5180 else
5181 ret = res_counter_set_limit(&memcg->kmem, val);
5182 return ret;
5183}
5184
Glauber Costa55007d82012-12-18 14:22:38 -08005185static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005186{
Glauber Costa55007d82012-12-18 14:22:38 -08005187 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005188 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005189
Glauber Costa510fc4e2012-12-18 14:21:47 -08005190 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005191 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08005192
Vladimir Davydovd6441632014-01-23 15:53:09 -08005193 mutex_lock(&activate_kmem_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08005194 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08005195 * If the parent cgroup is not kmem-active now, it cannot be activated
5196 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08005197 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08005198 if (memcg_kmem_is_active(parent))
5199 ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX);
5200 mutex_unlock(&activate_kmem_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005201 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005202}
Vladimir Davydovd6441632014-01-23 15:53:09 -08005203#else
5204static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
5205 unsigned long long val)
5206{
5207 return -EINVAL;
5208}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005209#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005210
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005211/*
5212 * The user of this function is...
5213 * RES_LIMIT.
5214 */
Tejun Heo182446d2013-08-08 20:11:24 -04005215static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Tejun Heo4d3bb512014-03-19 10:23:54 -04005216 char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005217{
Tejun Heo182446d2013-08-08 20:11:24 -04005218 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005219 enum res_type type;
5220 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005221 unsigned long long val;
5222 int ret;
5223
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005224 type = MEMFILE_TYPE(cft->private);
5225 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005226
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005227 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005228 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005229 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5230 ret = -EINVAL;
5231 break;
5232 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005233 /* This function does all necessary parse...reuse it */
5234 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005235 if (ret)
5236 break;
5237 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005238 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005239 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005240 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005241 else if (type == _KMEM)
Vladimir Davydovd6441632014-01-23 15:53:09 -08005242 ret = memcg_update_kmem_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005243 else
5244 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005245 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005246 case RES_SOFT_LIMIT:
5247 ret = res_counter_memparse_write_strategy(buffer, &val);
5248 if (ret)
5249 break;
5250 /*
5251 * For memsw, soft limits are hard to implement in terms
5252 * of semantics, for now, we support soft limits for
5253 * control without swap
5254 */
5255 if (type == _MEM)
5256 ret = res_counter_set_soft_limit(&memcg->res, val);
5257 else
5258 ret = -EINVAL;
5259 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005260 default:
5261 ret = -EINVAL; /* should be BUG() ? */
5262 break;
5263 }
5264 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005265}
5266
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005267static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5268 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5269{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005270 unsigned long long min_limit, min_memsw_limit, tmp;
5271
5272 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5273 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005274 if (!memcg->use_hierarchy)
5275 goto out;
5276
Tejun Heo63876982013-08-08 20:11:23 -04005277 while (css_parent(&memcg->css)) {
5278 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005279 if (!memcg->use_hierarchy)
5280 break;
5281 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5282 min_limit = min(min_limit, tmp);
5283 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5284 min_memsw_limit = min(min_memsw_limit, tmp);
5285 }
5286out:
5287 *mem_limit = min_limit;
5288 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005289}
5290
Tejun Heo182446d2013-08-08 20:11:24 -04005291static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005292{
Tejun Heo182446d2013-08-08 20:11:24 -04005293 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005294 int name;
5295 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005296
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005297 type = MEMFILE_TYPE(event);
5298 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005299
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005300 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005301 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005302 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005303 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005304 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005305 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005306 else if (type == _KMEM)
5307 res_counter_reset_max(&memcg->kmem);
5308 else
5309 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005310 break;
5311 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005312 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005313 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005314 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005315 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005316 else if (type == _KMEM)
5317 res_counter_reset_failcnt(&memcg->kmem);
5318 else
5319 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005320 break;
5321 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005322
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005323 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005324}
5325
Tejun Heo182446d2013-08-08 20:11:24 -04005326static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005327 struct cftype *cft)
5328{
Tejun Heo182446d2013-08-08 20:11:24 -04005329 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005330}
5331
Daisuke Nishimura02491442010-03-10 15:22:17 -08005332#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005333static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005334 struct cftype *cft, u64 val)
5335{
Tejun Heo182446d2013-08-08 20:11:24 -04005336 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005337
5338 if (val >= (1 << NR_MOVE_TYPE))
5339 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005340
Glauber Costaee5e8472013-02-22 16:34:50 -08005341 /*
5342 * No kind of locking is needed in here, because ->can_attach() will
5343 * check this value once in the beginning of the process, and then carry
5344 * on with stale data. This means that changes to this value will only
5345 * affect task migrations starting after the change.
5346 */
5347 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005348 return 0;
5349}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005350#else
Tejun Heo182446d2013-08-08 20:11:24 -04005351static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005352 struct cftype *cft, u64 val)
5353{
5354 return -ENOSYS;
5355}
5356#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005357
Ying Han406eb0c2011-05-26 16:25:37 -07005358#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05005359static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07005360{
Greg Thelen25485de2013-11-12 15:07:40 -08005361 struct numa_stat {
5362 const char *name;
5363 unsigned int lru_mask;
5364 };
5365
5366 static const struct numa_stat stats[] = {
5367 { "total", LRU_ALL },
5368 { "file", LRU_ALL_FILE },
5369 { "anon", LRU_ALL_ANON },
5370 { "unevictable", BIT(LRU_UNEVICTABLE) },
5371 };
5372 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07005373 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08005374 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05005375 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07005376
Greg Thelen25485de2013-11-12 15:07:40 -08005377 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5378 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
5379 seq_printf(m, "%s=%lu", stat->name, nr);
5380 for_each_node_state(nid, N_MEMORY) {
5381 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5382 stat->lru_mask);
5383 seq_printf(m, " N%d=%lu", nid, nr);
5384 }
5385 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005386 }
Ying Han406eb0c2011-05-26 16:25:37 -07005387
Ying Han071aee12013-11-12 15:07:41 -08005388 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5389 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07005390
Ying Han071aee12013-11-12 15:07:41 -08005391 nr = 0;
5392 for_each_mem_cgroup_tree(iter, memcg)
5393 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
5394 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
5395 for_each_node_state(nid, N_MEMORY) {
5396 nr = 0;
5397 for_each_mem_cgroup_tree(iter, memcg)
5398 nr += mem_cgroup_node_nr_lru_pages(
5399 iter, nid, stat->lru_mask);
5400 seq_printf(m, " N%d=%lu", nid, nr);
5401 }
5402 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005403 }
Ying Han406eb0c2011-05-26 16:25:37 -07005404
Ying Han406eb0c2011-05-26 16:25:37 -07005405 return 0;
5406}
5407#endif /* CONFIG_NUMA */
5408
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005409static inline void mem_cgroup_lru_names_not_uptodate(void)
5410{
5411 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5412}
5413
Tejun Heo2da8ca82013-12-05 12:28:04 -05005414static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005415{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005416 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005417 struct mem_cgroup *mi;
5418 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005419
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005420 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005421 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005422 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005423 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5424 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005425 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005426
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005427 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5428 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5429 mem_cgroup_read_events(memcg, i));
5430
5431 for (i = 0; i < NR_LRU_LISTS; i++)
5432 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5433 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5434
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005435 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005436 {
5437 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005438 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005439 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005440 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005441 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5442 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005443 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005444
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005445 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5446 long long val = 0;
5447
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005448 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005449 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005450 for_each_mem_cgroup_tree(mi, memcg)
5451 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5452 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5453 }
5454
5455 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5456 unsigned long long val = 0;
5457
5458 for_each_mem_cgroup_tree(mi, memcg)
5459 val += mem_cgroup_read_events(mi, i);
5460 seq_printf(m, "total_%s %llu\n",
5461 mem_cgroup_events_names[i], val);
5462 }
5463
5464 for (i = 0; i < NR_LRU_LISTS; i++) {
5465 unsigned long long val = 0;
5466
5467 for_each_mem_cgroup_tree(mi, memcg)
5468 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5469 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005470 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005471
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005472#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005473 {
5474 int nid, zid;
5475 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005476 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005477 unsigned long recent_rotated[2] = {0, 0};
5478 unsigned long recent_scanned[2] = {0, 0};
5479
5480 for_each_online_node(nid)
5481 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005482 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005483 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005484
Hugh Dickins89abfab2012-05-29 15:06:53 -07005485 recent_rotated[0] += rstat->recent_rotated[0];
5486 recent_rotated[1] += rstat->recent_rotated[1];
5487 recent_scanned[0] += rstat->recent_scanned[0];
5488 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005489 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005490 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5491 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5492 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5493 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005494 }
5495#endif
5496
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005497 return 0;
5498}
5499
Tejun Heo182446d2013-08-08 20:11:24 -04005500static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5501 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005502{
Tejun Heo182446d2013-08-08 20:11:24 -04005503 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005504
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005505 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005506}
5507
Tejun Heo182446d2013-08-08 20:11:24 -04005508static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5509 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005510{
Tejun Heo182446d2013-08-08 20:11:24 -04005511 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005512 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005513
Tejun Heo63876982013-08-08 20:11:23 -04005514 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005515 return -EINVAL;
5516
Glauber Costa09998212013-02-22 16:34:55 -08005517 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005518
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005519 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005520 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005521 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005522 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005523 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005524
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005525 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005526
Glauber Costa09998212013-02-22 16:34:55 -08005527 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005528
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005529 return 0;
5530}
5531
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005532static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5533{
5534 struct mem_cgroup_threshold_ary *t;
5535 u64 usage;
5536 int i;
5537
5538 rcu_read_lock();
5539 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005540 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005541 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005542 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005543
5544 if (!t)
5545 goto unlock;
5546
5547 usage = mem_cgroup_usage(memcg, swap);
5548
5549 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005550 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005551 * If it's not true, a threshold was crossed after last
5552 * call of __mem_cgroup_threshold().
5553 */
Phil Carmody5407a562010-05-26 14:42:42 -07005554 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005555
5556 /*
5557 * Iterate backward over array of thresholds starting from
5558 * current_threshold and check if a threshold is crossed.
5559 * If none of thresholds below usage is crossed, we read
5560 * only one element of the array here.
5561 */
5562 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5563 eventfd_signal(t->entries[i].eventfd, 1);
5564
5565 /* i = current_threshold + 1 */
5566 i++;
5567
5568 /*
5569 * Iterate forward over array of thresholds starting from
5570 * current_threshold+1 and check if a threshold is crossed.
5571 * If none of thresholds above usage is crossed, we read
5572 * only one element of the array here.
5573 */
5574 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5575 eventfd_signal(t->entries[i].eventfd, 1);
5576
5577 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005578 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005579unlock:
5580 rcu_read_unlock();
5581}
5582
5583static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5584{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005585 while (memcg) {
5586 __mem_cgroup_threshold(memcg, false);
5587 if (do_swap_account)
5588 __mem_cgroup_threshold(memcg, true);
5589
5590 memcg = parent_mem_cgroup(memcg);
5591 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005592}
5593
5594static int compare_thresholds(const void *a, const void *b)
5595{
5596 const struct mem_cgroup_threshold *_a = a;
5597 const struct mem_cgroup_threshold *_b = b;
5598
Greg Thelen2bff24a2013-09-11 14:23:08 -07005599 if (_a->threshold > _b->threshold)
5600 return 1;
5601
5602 if (_a->threshold < _b->threshold)
5603 return -1;
5604
5605 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005606}
5607
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005608static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005609{
5610 struct mem_cgroup_eventfd_list *ev;
5611
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005612 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005613 eventfd_signal(ev->eventfd, 1);
5614 return 0;
5615}
5616
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005617static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005618{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005619 struct mem_cgroup *iter;
5620
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005621 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005622 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005623}
5624
Tejun Heo59b6f872013-11-22 18:20:43 -05005625static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005626 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005627{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005628 struct mem_cgroup_thresholds *thresholds;
5629 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005630 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005631 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005632
5633 ret = res_counter_memparse_write_strategy(args, &threshold);
5634 if (ret)
5635 return ret;
5636
5637 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005638
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005639 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005640 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005641 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005642 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005643 else
5644 BUG();
5645
5646 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5647
5648 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005649 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005650 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5651
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005652 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005653
5654 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005655 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005656 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005657 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005658 ret = -ENOMEM;
5659 goto unlock;
5660 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005661 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005662
5663 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005664 if (thresholds->primary) {
5665 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005666 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005667 }
5668
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005669 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005670 new->entries[size - 1].eventfd = eventfd;
5671 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005672
5673 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005674 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005675 compare_thresholds, NULL);
5676
5677 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005678 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005679 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005680 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005681 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005682 * new->current_threshold will not be used until
5683 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005684 * it here.
5685 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005686 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005687 } else
5688 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005689 }
5690
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005691 /* Free old spare buffer and save old primary buffer as spare */
5692 kfree(thresholds->spare);
5693 thresholds->spare = thresholds->primary;
5694
5695 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005696
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005697 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005698 synchronize_rcu();
5699
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005700unlock:
5701 mutex_unlock(&memcg->thresholds_lock);
5702
5703 return ret;
5704}
5705
Tejun Heo59b6f872013-11-22 18:20:43 -05005706static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005707 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005708{
Tejun Heo59b6f872013-11-22 18:20:43 -05005709 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005710}
5711
Tejun Heo59b6f872013-11-22 18:20:43 -05005712static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005713 struct eventfd_ctx *eventfd, const char *args)
5714{
Tejun Heo59b6f872013-11-22 18:20:43 -05005715 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005716}
5717
Tejun Heo59b6f872013-11-22 18:20:43 -05005718static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005719 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005720{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005721 struct mem_cgroup_thresholds *thresholds;
5722 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005723 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005724 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005725
5726 mutex_lock(&memcg->thresholds_lock);
5727 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005728 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005729 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005730 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005731 else
5732 BUG();
5733
Anton Vorontsov371528c2012-02-24 05:14:46 +04005734 if (!thresholds->primary)
5735 goto unlock;
5736
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005737 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5738
5739 /* Check if a threshold crossed before removing */
5740 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5741
5742 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005743 size = 0;
5744 for (i = 0; i < thresholds->primary->size; i++) {
5745 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005746 size++;
5747 }
5748
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005749 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005750
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005751 /* Set thresholds array to NULL if we don't have thresholds */
5752 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005753 kfree(new);
5754 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005755 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005756 }
5757
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005758 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005759
5760 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005761 new->current_threshold = -1;
5762 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5763 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005764 continue;
5765
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005766 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005767 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005768 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005769 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005770 * until rcu_assign_pointer(), so it's safe to increment
5771 * it here.
5772 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005773 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005774 }
5775 j++;
5776 }
5777
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005778swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005779 /* Swap primary and spare array */
5780 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005781 /* If all events are unregistered, free the spare array */
5782 if (!new) {
5783 kfree(thresholds->spare);
5784 thresholds->spare = NULL;
5785 }
5786
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005787 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005788
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005789 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005790 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005791unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005792 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005793}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005794
Tejun Heo59b6f872013-11-22 18:20:43 -05005795static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005796 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005797{
Tejun Heo59b6f872013-11-22 18:20:43 -05005798 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05005799}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005800
Tejun Heo59b6f872013-11-22 18:20:43 -05005801static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005802 struct eventfd_ctx *eventfd)
5803{
Tejun Heo59b6f872013-11-22 18:20:43 -05005804 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05005805}
5806
Tejun Heo59b6f872013-11-22 18:20:43 -05005807static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005808 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005809{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005810 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005811
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005812 event = kmalloc(sizeof(*event), GFP_KERNEL);
5813 if (!event)
5814 return -ENOMEM;
5815
Michal Hocko1af8efe2011-07-26 16:08:24 -07005816 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005817
5818 event->eventfd = eventfd;
5819 list_add(&event->list, &memcg->oom_notify);
5820
5821 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005822 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005823 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005824 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005825
5826 return 0;
5827}
5828
Tejun Heo59b6f872013-11-22 18:20:43 -05005829static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05005830 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005831{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005832 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005833
Michal Hocko1af8efe2011-07-26 16:08:24 -07005834 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005835
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005836 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005837 if (ev->eventfd == eventfd) {
5838 list_del(&ev->list);
5839 kfree(ev);
5840 }
5841 }
5842
Michal Hocko1af8efe2011-07-26 16:08:24 -07005843 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005844}
5845
Tejun Heo2da8ca82013-12-05 12:28:04 -05005846static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005847{
Tejun Heo2da8ca82013-12-05 12:28:04 -05005848 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005849
Tejun Heo791badb2013-12-05 12:28:02 -05005850 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
5851 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005852 return 0;
5853}
5854
Tejun Heo182446d2013-08-08 20:11:24 -04005855static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005856 struct cftype *cft, u64 val)
5857{
Tejun Heo182446d2013-08-08 20:11:24 -04005858 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005859 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005860
5861 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005862 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005863 return -EINVAL;
5864
Glauber Costa09998212013-02-22 16:34:55 -08005865 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005866 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005867 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005868 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005869 return -EINVAL;
5870 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005871 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005872 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005873 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005874 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005875 return 0;
5876}
5877
Andrew Mortonc255a452012-07-31 16:43:02 -07005878#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005879static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005880{
Glauber Costa55007d82012-12-18 14:22:38 -08005881 int ret;
5882
Glauber Costa2633d7a2012-12-18 14:22:34 -08005883 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005884 ret = memcg_propagate_kmem(memcg);
5885 if (ret)
5886 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005887
Glauber Costa1d62e432012-04-09 19:36:33 -03005888 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005889}
Glauber Costae5671df2011-12-11 21:47:01 +00005890
Li Zefan10d5ebf2013-07-08 16:00:33 -07005891static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005892{
Glauber Costa1d62e432012-04-09 19:36:33 -03005893 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005894}
5895
5896static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5897{
5898 if (!memcg_kmem_is_active(memcg))
5899 return;
5900
5901 /*
5902 * kmem charges can outlive the cgroup. In the case of slab
5903 * pages, for instance, a page contain objects from various
5904 * processes. As we prevent from taking a reference for every
5905 * such allocation we have to be careful when doing uncharge
5906 * (see memcg_uncharge_kmem) and here during offlining.
5907 *
5908 * The idea is that that only the _last_ uncharge which sees
5909 * the dead memcg will drop the last reference. An additional
5910 * reference is taken here before the group is marked dead
5911 * which is then paired with css_put during uncharge resp. here.
5912 *
5913 * Although this might sound strange as this path is called from
5914 * css_offline() when the referencemight have dropped down to 0
5915 * and shouldn't be incremented anymore (css_tryget would fail)
5916 * we do not have other options because of the kmem allocations
5917 * lifetime.
5918 */
5919 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005920
5921 memcg_kmem_mark_dead(memcg);
5922
5923 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5924 return;
5925
Glauber Costa7de37682012-12-18 14:22:07 -08005926 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005927 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005928}
Glauber Costae5671df2011-12-11 21:47:01 +00005929#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005930static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005931{
5932 return 0;
5933}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005934
Li Zefan10d5ebf2013-07-08 16:00:33 -07005935static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5936{
5937}
5938
5939static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005940{
5941}
Glauber Costae5671df2011-12-11 21:47:01 +00005942#endif
5943
Tejun Heo79bd9812013-11-22 18:20:42 -05005944/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05005945 * DO NOT USE IN NEW FILES.
5946 *
5947 * "cgroup.event_control" implementation.
5948 *
5949 * This is way over-engineered. It tries to support fully configurable
5950 * events for each user. Such level of flexibility is completely
5951 * unnecessary especially in the light of the planned unified hierarchy.
5952 *
5953 * Please deprecate this and replace with something simpler if at all
5954 * possible.
5955 */
5956
5957/*
Tejun Heo79bd9812013-11-22 18:20:42 -05005958 * Unregister event and free resources.
5959 *
5960 * Gets called from workqueue.
5961 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05005962static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05005963{
Tejun Heo3bc942f2013-11-22 18:20:44 -05005964 struct mem_cgroup_event *event =
5965 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05005966 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005967
5968 remove_wait_queue(event->wqh, &event->wait);
5969
Tejun Heo59b6f872013-11-22 18:20:43 -05005970 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05005971
5972 /* Notify userspace the event is going away. */
5973 eventfd_signal(event->eventfd, 1);
5974
5975 eventfd_ctx_put(event->eventfd);
5976 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05005977 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05005978}
5979
5980/*
5981 * Gets called on POLLHUP on eventfd when user closes it.
5982 *
5983 * Called with wqh->lock held and interrupts disabled.
5984 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05005985static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
5986 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05005987{
Tejun Heo3bc942f2013-11-22 18:20:44 -05005988 struct mem_cgroup_event *event =
5989 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05005990 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005991 unsigned long flags = (unsigned long)key;
5992
5993 if (flags & POLLHUP) {
5994 /*
5995 * If the event has been detached at cgroup removal, we
5996 * can simply return knowing the other side will cleanup
5997 * for us.
5998 *
5999 * We can't race against event freeing since the other
6000 * side will require wqh->lock via remove_wait_queue(),
6001 * which we hold.
6002 */
Tejun Heofba94802013-11-22 18:20:43 -05006003 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006004 if (!list_empty(&event->list)) {
6005 list_del_init(&event->list);
6006 /*
6007 * We are in atomic context, but cgroup_event_remove()
6008 * may sleep, so we have to call it in workqueue.
6009 */
6010 schedule_work(&event->remove);
6011 }
Tejun Heofba94802013-11-22 18:20:43 -05006012 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006013 }
6014
6015 return 0;
6016}
6017
Tejun Heo3bc942f2013-11-22 18:20:44 -05006018static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05006019 wait_queue_head_t *wqh, poll_table *pt)
6020{
Tejun Heo3bc942f2013-11-22 18:20:44 -05006021 struct mem_cgroup_event *event =
6022 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05006023
6024 event->wqh = wqh;
6025 add_wait_queue(wqh, &event->wait);
6026}
6027
6028/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05006029 * DO NOT USE IN NEW FILES.
6030 *
Tejun Heo79bd9812013-11-22 18:20:42 -05006031 * Parse input and register new cgroup event handler.
6032 *
6033 * Input must be in format '<event_fd> <control_fd> <args>'.
6034 * Interpretation of args is defined by control file implementation.
6035 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05006036static int memcg_write_event_control(struct cgroup_subsys_state *css,
Tejun Heo4d3bb512014-03-19 10:23:54 -04006037 struct cftype *cft, char *buffer)
Tejun Heo79bd9812013-11-22 18:20:42 -05006038{
Tejun Heofba94802013-11-22 18:20:43 -05006039 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006040 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05006041 struct cgroup_subsys_state *cfile_css;
6042 unsigned int efd, cfd;
6043 struct fd efile;
6044 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05006045 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05006046 char *endp;
6047 int ret;
6048
6049 efd = simple_strtoul(buffer, &endp, 10);
6050 if (*endp != ' ')
6051 return -EINVAL;
6052 buffer = endp + 1;
6053
6054 cfd = simple_strtoul(buffer, &endp, 10);
6055 if ((*endp != ' ') && (*endp != '\0'))
6056 return -EINVAL;
6057 buffer = endp + 1;
6058
6059 event = kzalloc(sizeof(*event), GFP_KERNEL);
6060 if (!event)
6061 return -ENOMEM;
6062
Tejun Heo59b6f872013-11-22 18:20:43 -05006063 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05006064 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006065 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
6066 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
6067 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05006068
6069 efile = fdget(efd);
6070 if (!efile.file) {
6071 ret = -EBADF;
6072 goto out_kfree;
6073 }
6074
6075 event->eventfd = eventfd_ctx_fileget(efile.file);
6076 if (IS_ERR(event->eventfd)) {
6077 ret = PTR_ERR(event->eventfd);
6078 goto out_put_efile;
6079 }
6080
6081 cfile = fdget(cfd);
6082 if (!cfile.file) {
6083 ret = -EBADF;
6084 goto out_put_eventfd;
6085 }
6086
6087 /* the process need read permission on control file */
6088 /* AV: shouldn't we check that it's been opened for read instead? */
6089 ret = inode_permission(file_inode(cfile.file), MAY_READ);
6090 if (ret < 0)
6091 goto out_put_cfile;
6092
Tejun Heo79bd9812013-11-22 18:20:42 -05006093 /*
Tejun Heofba94802013-11-22 18:20:43 -05006094 * Determine the event callbacks and set them in @event. This used
6095 * to be done via struct cftype but cgroup core no longer knows
6096 * about these events. The following is crude but the whole thing
6097 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05006098 *
6099 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05006100 */
6101 name = cfile.file->f_dentry->d_name.name;
6102
6103 if (!strcmp(name, "memory.usage_in_bytes")) {
6104 event->register_event = mem_cgroup_usage_register_event;
6105 event->unregister_event = mem_cgroup_usage_unregister_event;
6106 } else if (!strcmp(name, "memory.oom_control")) {
6107 event->register_event = mem_cgroup_oom_register_event;
6108 event->unregister_event = mem_cgroup_oom_unregister_event;
6109 } else if (!strcmp(name, "memory.pressure_level")) {
6110 event->register_event = vmpressure_register_event;
6111 event->unregister_event = vmpressure_unregister_event;
6112 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05006113 event->register_event = memsw_cgroup_usage_register_event;
6114 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05006115 } else {
6116 ret = -EINVAL;
6117 goto out_put_cfile;
6118 }
6119
6120 /*
Tejun Heob5557c42013-11-22 18:20:42 -05006121 * Verify @cfile should belong to @css. Also, remaining events are
6122 * automatically removed on cgroup destruction but the removal is
6123 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05006124 */
Tejun Heo5a17f542014-02-11 11:52:47 -05006125 cfile_css = css_tryget_from_dir(cfile.file->f_dentry->d_parent,
6126 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05006127 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05006128 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05006129 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05006130 if (cfile_css != css) {
6131 css_put(cfile_css);
6132 goto out_put_cfile;
6133 }
Tejun Heo79bd9812013-11-22 18:20:42 -05006134
Tejun Heo59b6f872013-11-22 18:20:43 -05006135 ret = event->register_event(memcg, event->eventfd, buffer);
Tejun Heo79bd9812013-11-22 18:20:42 -05006136 if (ret)
6137 goto out_put_css;
6138
6139 efile.file->f_op->poll(efile.file, &event->pt);
6140
Tejun Heofba94802013-11-22 18:20:43 -05006141 spin_lock(&memcg->event_list_lock);
6142 list_add(&event->list, &memcg->event_list);
6143 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05006144
6145 fdput(cfile);
6146 fdput(efile);
6147
6148 return 0;
6149
6150out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05006151 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05006152out_put_cfile:
6153 fdput(cfile);
6154out_put_eventfd:
6155 eventfd_ctx_put(event->eventfd);
6156out_put_efile:
6157 fdput(efile);
6158out_kfree:
6159 kfree(event);
6160
6161 return ret;
6162}
6163
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006164static struct cftype mem_cgroup_files[] = {
6165 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006166 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006167 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006168 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006169 },
6170 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006171 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006172 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006173 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006174 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07006175 },
6176 {
Balbir Singh0eea1032008-02-07 00:13:57 -08006177 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006178 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07006179 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006180 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006181 },
6182 {
Balbir Singh296c81d2009-09-23 15:56:36 -07006183 .name = "soft_limit_in_bytes",
6184 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
6185 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006186 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07006187 },
6188 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006189 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006190 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07006191 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006192 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006193 },
Balbir Singh8697d332008-02-07 00:13:59 -08006194 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006195 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006196 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006197 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08006198 {
6199 .name = "force_empty",
6200 .trigger = mem_cgroup_force_empty_write,
6201 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08006202 {
6203 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07006204 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08006205 .write_u64 = mem_cgroup_hierarchy_write,
6206 .read_u64 = mem_cgroup_hierarchy_read,
6207 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006208 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05006209 .name = "cgroup.event_control", /* XXX: for compat */
6210 .write_string = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05006211 .flags = CFTYPE_NO_PREFIX,
6212 .mode = S_IWUGO,
6213 },
6214 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006215 .name = "swappiness",
6216 .read_u64 = mem_cgroup_swappiness_read,
6217 .write_u64 = mem_cgroup_swappiness_write,
6218 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006219 {
6220 .name = "move_charge_at_immigrate",
6221 .read_u64 = mem_cgroup_move_charge_read,
6222 .write_u64 = mem_cgroup_move_charge_write,
6223 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006224 {
6225 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006226 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006227 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006228 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6229 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006230 {
6231 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006232 },
Ying Han406eb0c2011-05-26 16:25:37 -07006233#ifdef CONFIG_NUMA
6234 {
6235 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006236 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006237 },
6238#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006239#ifdef CONFIG_MEMCG_KMEM
6240 {
6241 .name = "kmem.limit_in_bytes",
6242 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6243 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006244 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006245 },
6246 {
6247 .name = "kmem.usage_in_bytes",
6248 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006249 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006250 },
6251 {
6252 .name = "kmem.failcnt",
6253 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6254 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006255 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006256 },
6257 {
6258 .name = "kmem.max_usage_in_bytes",
6259 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6260 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006261 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08006262 },
Glauber Costa749c5412012-12-18 14:23:01 -08006263#ifdef CONFIG_SLABINFO
6264 {
6265 .name = "kmem.slabinfo",
Tejun Heo2da8ca82013-12-05 12:28:04 -05006266 .seq_show = mem_cgroup_slabinfo_read,
Glauber Costa749c5412012-12-18 14:23:01 -08006267 },
6268#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006269#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006270 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006271};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006272
Michal Hocko2d110852013-02-22 16:34:43 -08006273#ifdef CONFIG_MEMCG_SWAP
6274static struct cftype memsw_cgroup_files[] = {
6275 {
6276 .name = "memsw.usage_in_bytes",
6277 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05006278 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006279 },
6280 {
6281 .name = "memsw.max_usage_in_bytes",
6282 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6283 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006284 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006285 },
6286 {
6287 .name = "memsw.limit_in_bytes",
6288 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6289 .write_string = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05006290 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006291 },
6292 {
6293 .name = "memsw.failcnt",
6294 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6295 .trigger = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05006296 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08006297 },
6298 { }, /* terminate */
6299};
6300#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006301static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006302{
6303 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006304 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006305 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006306 /*
6307 * This routine is called against possible nodes.
6308 * But it's BUG to call kmalloc() against offline node.
6309 *
6310 * TODO: this routine can waste much memory for nodes which will
6311 * never be onlined. It's better to use memory hotplug callback
6312 * function.
6313 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006314 if (!node_state(node, N_NORMAL_MEMORY))
6315 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006316 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006317 if (!pn)
6318 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006319
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006320 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6321 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006322 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006323 mz->usage_in_excess = 0;
6324 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006325 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006326 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006327 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006328 return 0;
6329}
6330
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006331static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006332{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006333 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006334}
6335
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006336static struct mem_cgroup *mem_cgroup_alloc(void)
6337{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006338 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006339 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006340
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006341 size = sizeof(struct mem_cgroup);
6342 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006343
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006344 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006345 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006346 return NULL;
6347
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006348 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6349 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006350 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006351 spin_lock_init(&memcg->pcp_counter_lock);
6352 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006353
6354out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006355 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006356 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006357}
6358
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006359/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006360 * At destroying mem_cgroup, references from swap_cgroup can remain.
6361 * (scanning all at force_empty is too costly...)
6362 *
6363 * Instead of clearing all references at force_empty, we remember
6364 * the number of reference from swap_cgroup and free mem_cgroup when
6365 * it goes down to 0.
6366 *
6367 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006368 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006369
6370static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006371{
Glauber Costac8b2a362012-12-18 14:22:13 -08006372 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07006373
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006374 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006375
6376 for_each_node(node)
6377 free_mem_cgroup_per_zone_info(memcg, node);
6378
6379 free_percpu(memcg->stat);
6380
Glauber Costa3f134612012-05-29 15:07:11 -07006381 /*
6382 * We need to make sure that (at least for now), the jump label
6383 * destruction code runs outside of the cgroup lock. This is because
6384 * get_online_cpus(), which is called from the static_branch update,
6385 * can't be called inside the cgroup_lock. cpusets are the ones
6386 * enforcing this dependency, so if they ever change, we might as well.
6387 *
6388 * schedule_work() will guarantee this happens. Be careful if you need
6389 * to move this code around, and make sure it is outside
6390 * the cgroup_lock.
6391 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006392 disarm_static_keys(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08006393 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006394}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006395
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006396/*
6397 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6398 */
Glauber Costae1aab162011-12-11 21:47:03 +00006399struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006400{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006401 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006402 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006403 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006404}
Glauber Costae1aab162011-12-11 21:47:03 +00006405EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006406
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006407static void __init mem_cgroup_soft_limit_tree_init(void)
6408{
6409 struct mem_cgroup_tree_per_node *rtpn;
6410 struct mem_cgroup_tree_per_zone *rtpz;
6411 int tmp, node, zone;
6412
6413 for_each_node(node) {
6414 tmp = node;
6415 if (!node_state(node, N_NORMAL_MEMORY))
6416 tmp = -1;
6417 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
6418 BUG_ON(!rtpn);
6419
6420 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6421
6422 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6423 rtpz = &rtpn->rb_tree_per_zone[zone];
6424 rtpz->rb_root = RB_ROOT;
6425 spin_lock_init(&rtpz->lock);
6426 }
6427 }
6428}
6429
Li Zefan0eb253e2009-01-15 13:51:25 -08006430static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006431mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006432{
Glauber Costad142e3e2013-02-22 16:34:52 -08006433 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006434 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006435 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006436
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006437 memcg = mem_cgroup_alloc();
6438 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006439 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006440
Bob Liu3ed28fa2012-01-12 17:19:04 -08006441 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006442 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006443 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006444
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006445 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006446 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006447 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006448 res_counter_init(&memcg->res, NULL);
6449 res_counter_init(&memcg->memsw, NULL);
6450 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006451 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006452
Glauber Costad142e3e2013-02-22 16:34:52 -08006453 memcg->last_scanned_node = MAX_NUMNODES;
6454 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006455 memcg->move_charge_at_immigrate = 0;
6456 mutex_init(&memcg->thresholds_lock);
6457 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006458 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05006459 INIT_LIST_HEAD(&memcg->event_list);
6460 spin_lock_init(&memcg->event_list_lock);
Glauber Costad142e3e2013-02-22 16:34:52 -08006461
6462 return &memcg->css;
6463
6464free_out:
6465 __mem_cgroup_free(memcg);
6466 return ERR_PTR(error);
6467}
6468
6469static int
Tejun Heoeb954192013-08-08 20:11:23 -04006470mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006471{
Tejun Heoeb954192013-08-08 20:11:23 -04006472 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6473 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006474
Li Zefan4219b2d2013-09-23 16:56:29 +08006475 if (css->cgroup->id > MEM_CGROUP_ID_MAX)
6476 return -ENOSPC;
6477
Tejun Heo63876982013-08-08 20:11:23 -04006478 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006479 return 0;
6480
Glauber Costa09998212013-02-22 16:34:55 -08006481 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006482
6483 memcg->use_hierarchy = parent->use_hierarchy;
6484 memcg->oom_kill_disable = parent->oom_kill_disable;
6485 memcg->swappiness = mem_cgroup_swappiness(parent);
6486
6487 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006488 res_counter_init(&memcg->res, &parent->res);
6489 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006490 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006491
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006492 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006493 * No need to take a reference to the parent because cgroup
6494 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006495 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006496 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006497 res_counter_init(&memcg->res, NULL);
6498 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006499 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006500 /*
6501 * Deeper hierachy with use_hierarchy == false doesn't make
6502 * much sense so let cgroup subsystem know about this
6503 * unfortunate state in our controller.
6504 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006505 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05006506 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006507 }
Glauber Costa09998212013-02-22 16:34:55 -08006508 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08006509
Tejun Heo073219e2014-02-08 10:36:58 -05006510 return memcg_init_kmem(memcg, &memory_cgrp_subsys);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006511}
6512
Michal Hocko5f578162013-04-29 15:07:17 -07006513/*
6514 * Announce all parents that a group from their hierarchy is gone.
6515 */
6516static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6517{
6518 struct mem_cgroup *parent = memcg;
6519
6520 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006521 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006522
6523 /*
6524 * if the root memcg is not hierarchical we have to check it
6525 * explicitely.
6526 */
6527 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006528 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006529}
6530
Tejun Heoeb954192013-08-08 20:11:23 -04006531static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006532{
Tejun Heoeb954192013-08-08 20:11:23 -04006533 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05006534 struct mem_cgroup_event *event, *tmp;
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08006535 struct cgroup_subsys_state *iter;
Tejun Heo79bd9812013-11-22 18:20:42 -05006536
6537 /*
6538 * Unregister events and notify userspace.
6539 * Notify userspace about cgroup removing only after rmdir of cgroup
6540 * directory to avoid race between userspace and kernelspace.
6541 */
Tejun Heofba94802013-11-22 18:20:43 -05006542 spin_lock(&memcg->event_list_lock);
6543 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05006544 list_del_init(&event->list);
6545 schedule_work(&event->remove);
6546 }
Tejun Heofba94802013-11-22 18:20:43 -05006547 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006548
Li Zefan10d5ebf2013-07-08 16:00:33 -07006549 kmem_cgroup_css_offline(memcg);
6550
Michal Hocko5f578162013-04-29 15:07:17 -07006551 mem_cgroup_invalidate_reclaim_iterators(memcg);
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08006552
6553 /*
6554 * This requires that offlining is serialized. Right now that is
6555 * guaranteed because css_killed_work_fn() holds the cgroup_mutex.
6556 */
6557 css_for_each_descendant_post(iter, css)
6558 mem_cgroup_reparent_charges(mem_cgroup_from_css(iter));
6559
Glauber Costa1f458cb2012-12-18 14:22:50 -08006560 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006561 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006562}
6563
Tejun Heoeb954192013-08-08 20:11:23 -04006564static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006565{
Tejun Heoeb954192013-08-08 20:11:23 -04006566 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner96f1c582013-12-12 17:12:34 -08006567 /*
6568 * XXX: css_offline() would be where we should reparent all
6569 * memory to prepare the cgroup for destruction. However,
6570 * memcg does not do css_tryget() and res_counter charging
6571 * under the same RCU lock region, which means that charging
6572 * could race with offlining. Offlining only happens to
6573 * cgroups with no tasks in them but charges can show up
6574 * without any tasks from the swapin path when the target
6575 * memcg is looked up from the swapout record and not from the
6576 * current task as it usually is. A race like this can leak
6577 * charges and put pages with stale cgroup pointers into
6578 * circulation:
6579 *
6580 * #0 #1
6581 * lookup_swap_cgroup_id()
6582 * rcu_read_lock()
6583 * mem_cgroup_lookup()
6584 * css_tryget()
6585 * rcu_read_unlock()
6586 * disable css_tryget()
6587 * call_rcu()
6588 * offline_css()
6589 * reparent_charges()
6590 * res_counter_charge()
6591 * css_put()
6592 * css_free()
6593 * pc->mem_cgroup = dead memcg
6594 * add page to lru
6595 *
6596 * The bulk of the charges are still moved in offline_css() to
6597 * avoid pinning a lot of pages in case a long-term reference
6598 * like a swapout record is deferring the css_free() to long
6599 * after offlining. But this makes sure we catch any charges
6600 * made after offlining:
6601 */
6602 mem_cgroup_reparent_charges(memcg);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006603
Li Zefan10d5ebf2013-07-08 16:00:33 -07006604 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006605 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006606}
6607
Daisuke Nishimura02491442010-03-10 15:22:17 -08006608#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006609/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006610#define PRECHARGE_COUNT_AT_ONCE 256
6611static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006612{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006613 int ret = 0;
6614 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006615 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006616
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006617 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006618 mc.precharge += count;
6619 /* we don't need css_get for root */
6620 return ret;
6621 }
6622 /* try to charge at once */
6623 if (count > 1) {
6624 struct res_counter *dummy;
6625 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006626 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006627 * by cgroup_lock_live_cgroup() that it is not removed and we
6628 * are still under the same cgroup_mutex. So we can postpone
6629 * css_get().
6630 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006631 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006632 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006633 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006634 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006635 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006636 goto one_by_one;
6637 }
6638 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006639 return ret;
6640 }
6641one_by_one:
6642 /* fall back to one by one charge */
6643 while (count--) {
6644 if (signal_pending(current)) {
6645 ret = -EINTR;
6646 break;
6647 }
6648 if (!batch_count--) {
6649 batch_count = PRECHARGE_COUNT_AT_ONCE;
6650 cond_resched();
6651 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006652 ret = __mem_cgroup_try_charge(NULL,
6653 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006654 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006655 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006656 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006657 mc.precharge++;
6658 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006659 return ret;
6660}
6661
6662/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006663 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006664 * @vma: the vma the pte to be checked belongs
6665 * @addr: the address corresponding to the pte to be checked
6666 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006667 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006668 *
6669 * Returns
6670 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6671 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6672 * move charge. if @target is not NULL, the page is stored in target->page
6673 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006674 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6675 * target for charge migration. if @target is not NULL, the entry is stored
6676 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006677 *
6678 * Called with pte lock held.
6679 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006680union mc_target {
6681 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006682 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006683};
6684
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006685enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006686 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006687 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006688 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006689};
6690
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006691static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6692 unsigned long addr, pte_t ptent)
6693{
6694 struct page *page = vm_normal_page(vma, addr, ptent);
6695
6696 if (!page || !page_mapped(page))
6697 return NULL;
6698 if (PageAnon(page)) {
6699 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006700 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006701 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006702 } else if (!move_file())
6703 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006704 return NULL;
6705 if (!get_page_unless_zero(page))
6706 return NULL;
6707
6708 return page;
6709}
6710
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006711#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006712static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6713 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6714{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006715 struct page *page = NULL;
6716 swp_entry_t ent = pte_to_swp_entry(ptent);
6717
6718 if (!move_anon() || non_swap_entry(ent))
6719 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006720 /*
6721 * Because lookup_swap_cache() updates some statistics counter,
6722 * we call find_get_page() with swapper_space directly.
6723 */
Shaohua Li33806f02013-02-22 16:34:37 -08006724 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006725 if (do_swap_account)
6726 entry->val = ent.val;
6727
6728 return page;
6729}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006730#else
6731static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6732 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6733{
6734 return NULL;
6735}
6736#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006737
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006738static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6739 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6740{
6741 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006742 struct address_space *mapping;
6743 pgoff_t pgoff;
6744
6745 if (!vma->vm_file) /* anonymous vma */
6746 return NULL;
6747 if (!move_file())
6748 return NULL;
6749
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006750 mapping = vma->vm_file->f_mapping;
6751 if (pte_none(ptent))
6752 pgoff = linear_page_index(vma, addr);
6753 else /* pte_file(ptent) is true */
6754 pgoff = pte_to_pgoff(ptent);
6755
6756 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006757 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006758
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006759#ifdef CONFIG_SWAP
6760 /* shmem/tmpfs may report page out on swap: account for that too. */
6761 if (radix_tree_exceptional_entry(page)) {
6762 swp_entry_t swap = radix_to_swp_entry(page);
6763 if (do_swap_account)
6764 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006765 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006766 }
6767#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006768 return page;
6769}
6770
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006771static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006772 unsigned long addr, pte_t ptent, union mc_target *target)
6773{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006774 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006775 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006776 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006777 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006778
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006779 if (pte_present(ptent))
6780 page = mc_handle_present_pte(vma, addr, ptent);
6781 else if (is_swap_pte(ptent))
6782 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006783 else if (pte_none(ptent) || pte_file(ptent))
6784 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006785
6786 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006787 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006788 if (page) {
6789 pc = lookup_page_cgroup(page);
6790 /*
6791 * Do only loose check w/o page_cgroup lock.
6792 * mem_cgroup_move_account() checks the pc is valid or not under
6793 * the lock.
6794 */
6795 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6796 ret = MC_TARGET_PAGE;
6797 if (target)
6798 target->page = page;
6799 }
6800 if (!ret || !target)
6801 put_page(page);
6802 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006803 /* There is a swap entry and a page doesn't exist or isn't charged */
6804 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08006805 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006806 ret = MC_TARGET_SWAP;
6807 if (target)
6808 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006809 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006810 return ret;
6811}
6812
Naoya Horiguchi12724852012-03-21 16:34:28 -07006813#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6814/*
6815 * We don't consider swapping or file mapped pages because THP does not
6816 * support them for now.
6817 * Caller should make sure that pmd_trans_huge(pmd) is true.
6818 */
6819static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6820 unsigned long addr, pmd_t pmd, union mc_target *target)
6821{
6822 struct page *page = NULL;
6823 struct page_cgroup *pc;
6824 enum mc_target_type ret = MC_TARGET_NONE;
6825
6826 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08006827 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006828 if (!move_anon())
6829 return ret;
6830 pc = lookup_page_cgroup(page);
6831 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6832 ret = MC_TARGET_PAGE;
6833 if (target) {
6834 get_page(page);
6835 target->page = page;
6836 }
6837 }
6838 return ret;
6839}
6840#else
6841static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6842 unsigned long addr, pmd_t pmd, union mc_target *target)
6843{
6844 return MC_TARGET_NONE;
6845}
6846#endif
6847
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006848static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6849 unsigned long addr, unsigned long end,
6850 struct mm_walk *walk)
6851{
6852 struct vm_area_struct *vma = walk->private;
6853 pte_t *pte;
6854 spinlock_t *ptl;
6855
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006856 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006857 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6858 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006859 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006860 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006861 }
Dave Hansen03319322011-03-22 16:32:56 -07006862
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006863 if (pmd_trans_unstable(pmd))
6864 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006865 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6866 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006867 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006868 mc.precharge++; /* increment precharge temporarily */
6869 pte_unmap_unlock(pte - 1, ptl);
6870 cond_resched();
6871
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006872 return 0;
6873}
6874
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006875static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6876{
6877 unsigned long precharge;
6878 struct vm_area_struct *vma;
6879
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006880 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006881 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6882 struct mm_walk mem_cgroup_count_precharge_walk = {
6883 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6884 .mm = mm,
6885 .private = vma,
6886 };
6887 if (is_vm_hugetlb_page(vma))
6888 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006889 walk_page_range(vma->vm_start, vma->vm_end,
6890 &mem_cgroup_count_precharge_walk);
6891 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006892 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006893
6894 precharge = mc.precharge;
6895 mc.precharge = 0;
6896
6897 return precharge;
6898}
6899
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006900static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6901{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006902 unsigned long precharge = mem_cgroup_count_precharge(mm);
6903
6904 VM_BUG_ON(mc.moving_task);
6905 mc.moving_task = current;
6906 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006907}
6908
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006909/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6910static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006911{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006912 struct mem_cgroup *from = mc.from;
6913 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006914 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006915
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006916 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006917 if (mc.precharge) {
6918 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6919 mc.precharge = 0;
6920 }
6921 /*
6922 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6923 * we must uncharge here.
6924 */
6925 if (mc.moved_charge) {
6926 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6927 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006928 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006929 /* we must fixup refcnts and charges */
6930 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006931 /* uncharge swap account from the old cgroup */
6932 if (!mem_cgroup_is_root(mc.from))
6933 res_counter_uncharge(&mc.from->memsw,
6934 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006935
6936 for (i = 0; i < mc.moved_swap; i++)
6937 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006938
6939 if (!mem_cgroup_is_root(mc.to)) {
6940 /*
6941 * we charged both to->res and to->memsw, so we should
6942 * uncharge to->res.
6943 */
6944 res_counter_uncharge(&mc.to->res,
6945 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006946 }
Li Zefan40503772013-07-08 16:00:34 -07006947 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006948 mc.moved_swap = 0;
6949 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006950 memcg_oom_recover(from);
6951 memcg_oom_recover(to);
6952 wake_up_all(&mc.waitq);
6953}
6954
6955static void mem_cgroup_clear_mc(void)
6956{
6957 struct mem_cgroup *from = mc.from;
6958
6959 /*
6960 * we must clear moving_task before waking up waiters at the end of
6961 * task migration.
6962 */
6963 mc.moving_task = NULL;
6964 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006965 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006966 mc.from = NULL;
6967 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006968 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006969 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006970}
6971
Tejun Heoeb954192013-08-08 20:11:23 -04006972static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006973 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006974{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006975 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006976 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04006977 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08006978 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006979
Glauber Costaee5e8472013-02-22 16:34:50 -08006980 /*
6981 * We are now commited to this value whatever it is. Changes in this
6982 * tunable will only affect upcoming migrations, not the current one.
6983 * So we need to save it, and keep it going.
6984 */
6985 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6986 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006987 struct mm_struct *mm;
6988 struct mem_cgroup *from = mem_cgroup_from_task(p);
6989
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006990 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006991
6992 mm = get_task_mm(p);
6993 if (!mm)
6994 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006995 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006996 if (mm->owner == p) {
6997 VM_BUG_ON(mc.from);
6998 VM_BUG_ON(mc.to);
6999 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007000 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007001 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07007002 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007003 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007004 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07007005 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08007006 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07007007 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007008 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007009
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007010 ret = mem_cgroup_precharge_mc(mm);
7011 if (ret)
7012 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007013 }
7014 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007015 }
7016 return ret;
7017}
7018
Tejun Heoeb954192013-08-08 20:11:23 -04007019static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007020 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007021{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007022 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007023}
7024
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007025static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
7026 unsigned long addr, unsigned long end,
7027 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007028{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007029 int ret = 0;
7030 struct vm_area_struct *vma = walk->private;
7031 pte_t *pte;
7032 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07007033 enum mc_target_type target_type;
7034 union mc_target target;
7035 struct page *page;
7036 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007037
Naoya Horiguchi12724852012-03-21 16:34:28 -07007038 /*
7039 * We don't take compound_lock() here but no race with splitting thp
7040 * happens because:
7041 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
7042 * under splitting, which means there's no concurrent thp split,
7043 * - if another thread runs into split_huge_page() just after we
7044 * entered this if-block, the thread must wait for page table lock
7045 * to be unlocked in __split_huge_page_splitting(), where the main
7046 * part of thp split is not executed yet.
7047 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007048 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07007049 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007050 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07007051 return 0;
7052 }
7053 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
7054 if (target_type == MC_TARGET_PAGE) {
7055 page = target.page;
7056 if (!isolate_lru_page(page)) {
7057 pc = lookup_page_cgroup(page);
7058 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07007059 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07007060 mc.precharge -= HPAGE_PMD_NR;
7061 mc.moved_charge += HPAGE_PMD_NR;
7062 }
7063 putback_lru_page(page);
7064 }
7065 put_page(page);
7066 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08007067 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07007068 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07007069 }
7070
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07007071 if (pmd_trans_unstable(pmd))
7072 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007073retry:
7074 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
7075 for (; addr != end; addr += PAGE_SIZE) {
7076 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08007077 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007078
7079 if (!mc.precharge)
7080 break;
7081
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007082 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007083 case MC_TARGET_PAGE:
7084 page = target.page;
7085 if (isolate_lru_page(page))
7086 goto put;
7087 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07007088 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07007089 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007090 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007091 /* we uncharge from mc.from later. */
7092 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007093 }
7094 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07007095put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007096 put_page(page);
7097 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08007098 case MC_TARGET_SWAP:
7099 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07007100 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08007101 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08007102 /* we fixup refcnts and charges later. */
7103 mc.moved_swap++;
7104 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08007105 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007106 default:
7107 break;
7108 }
7109 }
7110 pte_unmap_unlock(pte - 1, ptl);
7111 cond_resched();
7112
7113 if (addr != end) {
7114 /*
7115 * We have consumed all precharges we got in can_attach().
7116 * We try charge one by one, but don't do any additional
7117 * charges to mc.to if we have failed in charge once in attach()
7118 * phase.
7119 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08007120 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007121 if (!ret)
7122 goto retry;
7123 }
7124
7125 return ret;
7126}
7127
7128static void mem_cgroup_move_charge(struct mm_struct *mm)
7129{
7130 struct vm_area_struct *vma;
7131
7132 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007133retry:
7134 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
7135 /*
7136 * Someone who are holding the mmap_sem might be waiting in
7137 * waitq. So we cancel all extra charges, wake up all waiters,
7138 * and retry. Because we cancel precharges, we might not be able
7139 * to move enough charges, but moving charge is a best-effort
7140 * feature anyway, so it wouldn't be a big problem.
7141 */
7142 __mem_cgroup_clear_mc();
7143 cond_resched();
7144 goto retry;
7145 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007146 for (vma = mm->mmap; vma; vma = vma->vm_next) {
7147 int ret;
7148 struct mm_walk mem_cgroup_move_charge_walk = {
7149 .pmd_entry = mem_cgroup_move_charge_pte_range,
7150 .mm = mm,
7151 .private = vma,
7152 };
7153 if (is_vm_hugetlb_page(vma))
7154 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08007155 ret = walk_page_range(vma->vm_start, vma->vm_end,
7156 &mem_cgroup_move_charge_walk);
7157 if (ret)
7158 /*
7159 * means we have consumed all precharges and failed in
7160 * doing additional charge. Just abandon here.
7161 */
7162 break;
7163 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007164 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007165}
7166
Tejun Heoeb954192013-08-08 20:11:23 -04007167static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007168 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08007169{
Tejun Heo2f7ee562011-12-12 18:12:21 -08007170 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007171 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007172
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007173 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007174 if (mc.to)
7175 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08007176 mmput(mm);
7177 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07007178 if (mc.to)
7179 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08007180}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007181#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04007182static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007183 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007184{
7185 return 0;
7186}
Tejun Heoeb954192013-08-08 20:11:23 -04007187static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007188 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007189{
7190}
Tejun Heoeb954192013-08-08 20:11:23 -04007191static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08007192 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07007193{
7194}
7195#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08007196
Tejun Heof00baae2013-04-15 13:41:15 -07007197/*
7198 * Cgroup retains root cgroups across [un]mount cycles making it necessary
7199 * to verify sane_behavior flag on each mount attempt.
7200 */
Tejun Heoeb954192013-08-08 20:11:23 -04007201static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07007202{
7203 /*
7204 * use_hierarchy is forced with sane_behavior. cgroup core
7205 * guarantees that @root doesn't have any children, so turning it
7206 * on for the root memcg is enough.
7207 */
Tejun Heoeb954192013-08-08 20:11:23 -04007208 if (cgroup_sane_behavior(root_css->cgroup))
7209 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07007210}
7211
Tejun Heo073219e2014-02-08 10:36:58 -05007212struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08007213 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08007214 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08007215 .css_offline = mem_cgroup_css_offline,
7216 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007217 .can_attach = mem_cgroup_can_attach,
7218 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08007219 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07007220 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07007221 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08007222 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007223};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007224
Andrew Mortonc255a452012-07-31 16:43:02 -07007225#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08007226static int __init enable_swap_account(char *s)
7227{
Michal Hockoa2c89902011-05-24 17:12:50 -07007228 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007229 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07007230 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007231 really_do_swap_account = 0;
7232 return 1;
7233}
Michal Hockoa2c89902011-05-24 17:12:50 -07007234__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007235
Michal Hocko2d110852013-02-22 16:34:43 -08007236static void __init memsw_file_init(void)
7237{
Tejun Heo073219e2014-02-08 10:36:58 -05007238 WARN_ON(cgroup_add_cftypes(&memory_cgrp_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08007239}
Michal Hocko6acc8b02013-02-22 16:34:45 -08007240
7241static void __init enable_swap_cgroup(void)
7242{
7243 if (!mem_cgroup_disabled() && really_do_swap_account) {
7244 do_swap_account = 1;
7245 memsw_file_init();
7246 }
7247}
7248
Michal Hocko2d110852013-02-22 16:34:43 -08007249#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08007250static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08007251{
7252}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007253#endif
Michal Hocko2d110852013-02-22 16:34:43 -08007254
7255/*
Michal Hocko10813122013-02-22 16:35:41 -08007256 * subsys_initcall() for memory controller.
7257 *
7258 * Some parts like hotcpu_notifier() have to be initialized from this context
7259 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7260 * everything that doesn't depend on a specific mem_cgroup structure should
7261 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007262 */
7263static int __init mem_cgroup_init(void)
7264{
7265 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007266 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07007267 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007268 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007269 return 0;
7270}
7271subsys_initcall(mem_cgroup_init);