blob: 23976fd885fd588a7687bf5631f98ea38f2235b4 [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
Fabian Frederickada4ba52014-06-04 16:08:08 -070083static int really_do_swap_account __initdata;
Michal Hockoa42c3902010-11-24 12:57:08 -080084#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
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -0700295 /* css_online() has been completed */
296 int initialized;
297
Li Zefan465939a2013-07-08 16:00:38 -0700298 /*
299 * the counter to account for mem+swap usage.
300 */
301 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700302
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800303 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800304 * the counter to account for kernel memory usage.
305 */
306 struct res_counter kmem;
307 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800308 * Should the accounting and control be hierarchical, per subtree?
309 */
310 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800311 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700312
313 bool oom_lock;
314 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700315 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700316
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700317 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700318 /* OOM-Killer disable */
319 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800320
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)
Vladimir Davydovbd673142014-06-04 16:07:40 -0700360 /* analogous to slab_common's slab_caches list, but per-memcg;
361 * protected by memcg_slab_mutex */
Glauber Costa2633d7a2012-12-18 14:22:34 -0800362 struct list_head memcg_slab_caches;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800363 /* Index in the kmem_cache->memcg_params->memcg_caches array */
364 int kmemcg_id;
365#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800366
367 int last_scanned_node;
368#if MAX_NUMNODES > 1
369 nodemask_t scan_nodes;
370 atomic_t numainfo_events;
371 atomic_t numainfo_updating;
372#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700373
Tejun Heofba94802013-11-22 18:20:43 -0500374 /* List of events which userspace want to receive */
375 struct list_head event_list;
376 spinlock_t event_list_lock;
377
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700378 struct mem_cgroup_per_node *nodeinfo[0];
379 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800380};
381
Glauber Costa510fc4e2012-12-18 14:21:47 -0800382/* internal only representation about the status of kmem accounting. */
383enum {
Vladimir Davydov6de64be2014-01-23 15:53:08 -0800384 KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */
Glauber Costa7de37682012-12-18 14:22:07 -0800385 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800386};
387
Glauber Costa510fc4e2012-12-18 14:21:47 -0800388#ifdef CONFIG_MEMCG_KMEM
389static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
390{
391 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
392}
Glauber Costa7de37682012-12-18 14:22:07 -0800393
394static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
395{
396 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
397}
398
399static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
400{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700401 /*
402 * Our caller must use css_get() first, because memcg_uncharge_kmem()
403 * will call css_put() if it sees the memcg is dead.
404 */
405 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800406 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
407 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
408}
409
410static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
411{
412 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
413 &memcg->kmem_account_flags);
414}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800415#endif
416
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800417/* Stuffs for move charges at task migration. */
418/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800419 * Types of charges to be moved. "move_charge_at_immitgrate" and
420 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800421 */
422enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800423 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700424 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800425 NR_MOVE_TYPE,
426};
427
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800428/* "mc" and its members are protected by cgroup_mutex */
429static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800430 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800431 struct mem_cgroup *from;
432 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800433 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800434 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800435 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800436 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800437 struct task_struct *moving_task; /* a task moving charges */
438 wait_queue_head_t waitq; /* a waitq for other context */
439} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700440 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800441 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
442};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800443
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700444static bool move_anon(void)
445{
Glauber Costaee5e8472013-02-22 16:34:50 -0800446 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700447}
448
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700449static bool move_file(void)
450{
Glauber Costaee5e8472013-02-22 16:34:50 -0800451 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700452}
453
Balbir Singh4e416952009-09-23 15:56:39 -0700454/*
455 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
456 * limit reclaim to prevent infinite loops, if they ever occur.
457 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700458#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700459#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700460
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800461enum charge_type {
462 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700463 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800464 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700465 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700466 NR_CHARGE_TYPE,
467};
468
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800469/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800470enum res_type {
471 _MEM,
472 _MEMSWAP,
473 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800474 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800475};
476
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700477#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
478#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800479#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700480/* Used for OOM nofiier */
481#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800482
Balbir Singh75822b42009-09-23 15:56:38 -0700483/*
Glauber Costa09998212013-02-22 16:34:55 -0800484 * The memcg_create_mutex will be held whenever a new cgroup is created.
485 * As a consequence, any change that needs to protect against new child cgroups
486 * appearing has to hold it as well.
487 */
488static DEFINE_MUTEX(memcg_create_mutex);
489
Wanpeng Lib2145142012-07-31 16:46:01 -0700490struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
491{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400492 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700493}
494
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700495/* Some nice accessors for the vmpressure. */
496struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
497{
498 if (!memcg)
499 memcg = root_mem_cgroup;
500 return &memcg->vmpressure;
501}
502
503struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
504{
505 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
506}
507
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700508static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
509{
510 return (memcg == root_mem_cgroup);
511}
512
Li Zefan4219b2d2013-09-23 16:56:29 +0800513/*
514 * We restrict the id in the range of [1, 65535], so it can fit into
515 * an unsigned short.
516 */
517#define MEM_CGROUP_ID_MAX USHRT_MAX
518
Li Zefan34c00c32013-09-23 16:56:01 +0800519static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
520{
Tejun Heo15a4c832014-05-04 15:09:14 -0400521 return memcg->css.id;
Li Zefan34c00c32013-09-23 16:56:01 +0800522}
523
524static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
525{
526 struct cgroup_subsys_state *css;
527
Tejun Heo7d699dd2014-05-04 15:09:13 -0400528 css = css_from_id(id, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800529 return mem_cgroup_from_css(css);
530}
531
Glauber Costae1aab162011-12-11 21:47:03 +0000532/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700533#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000534
Glauber Costae1aab162011-12-11 21:47:03 +0000535void sock_update_memcg(struct sock *sk)
536{
Glauber Costa376be5f2012-01-20 04:57:14 +0000537 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000538 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700539 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000540
541 BUG_ON(!sk->sk_prot->proto_cgroup);
542
Glauber Costaf3f511e2012-01-05 20:16:39 +0000543 /* Socket cloning can throw us here with sk_cgrp already
544 * filled. It won't however, necessarily happen from
545 * process context. So the test for root memcg given
546 * the current task's memcg won't help us in this case.
547 *
548 * Respecting the original socket's memcg is a better
549 * decision in this case.
550 */
551 if (sk->sk_cgrp) {
552 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700553 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000554 return;
555 }
556
Glauber Costae1aab162011-12-11 21:47:03 +0000557 rcu_read_lock();
558 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700559 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700560 if (!mem_cgroup_is_root(memcg) &&
Tejun Heoec903c02014-05-13 12:11:01 -0400561 memcg_proto_active(cg_proto) &&
562 css_tryget_online(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700563 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000564 }
565 rcu_read_unlock();
566 }
567}
568EXPORT_SYMBOL(sock_update_memcg);
569
570void sock_release_memcg(struct sock *sk)
571{
Glauber Costa376be5f2012-01-20 04:57:14 +0000572 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000573 struct mem_cgroup *memcg;
574 WARN_ON(!sk->sk_cgrp->memcg);
575 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700576 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000577 }
578}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000579
580struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
581{
582 if (!memcg || mem_cgroup_is_root(memcg))
583 return NULL;
584
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700585 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000586}
587EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000588
Glauber Costa3f134612012-05-29 15:07:11 -0700589static void disarm_sock_keys(struct mem_cgroup *memcg)
590{
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700591 if (!memcg_proto_activated(&memcg->tcp_mem))
Glauber Costa3f134612012-05-29 15:07:11 -0700592 return;
593 static_key_slow_dec(&memcg_socket_limit_enabled);
594}
595#else
596static void disarm_sock_keys(struct mem_cgroup *memcg)
597{
598}
599#endif
600
Glauber Costaa8964b92012-12-18 14:22:09 -0800601#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800602/*
603 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800604 * The main reason for not using cgroup id for this:
605 * this works better in sparse environments, where we have a lot of memcgs,
606 * but only a few kmem-limited. Or also, if we have, for instance, 200
607 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
608 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800609 *
610 * The current size of the caches array is stored in
611 * memcg_limited_groups_array_size. It will double each time we have to
612 * increase it.
613 */
614static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800615int memcg_limited_groups_array_size;
616
Glauber Costa55007d82012-12-18 14:22:38 -0800617/*
618 * MIN_SIZE is different than 1, because we would like to avoid going through
619 * the alloc/free process all the time. In a small machine, 4 kmem-limited
620 * cgroups is a reasonable guess. In the future, it could be a parameter or
621 * tunable, but that is strictly not necessary.
622 *
Li Zefanb8627832013-09-23 16:56:47 +0800623 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800624 * this constant directly from cgroup, but it is understandable that this is
625 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800626 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800627 * increase ours as well if it increases.
628 */
629#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800630#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800631
Glauber Costad7f25f82012-12-18 14:22:40 -0800632/*
633 * A lot of the calls to the cache allocation functions are expected to be
634 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
635 * conditional to this static branch, we'll have to allow modules that does
636 * kmem_cache_alloc and the such to see this symbol as well
637 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800638struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800639EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800640
Vladimir Davydovf3bb3042014-10-09 15:28:45 -0700641static void memcg_free_cache_id(int id);
642
Glauber Costaa8964b92012-12-18 14:22:09 -0800643static void disarm_kmem_keys(struct mem_cgroup *memcg)
644{
Glauber Costa55007d82012-12-18 14:22:38 -0800645 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800646 static_key_slow_dec(&memcg_kmem_enabled_key);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -0700647 memcg_free_cache_id(memcg->kmemcg_id);
Glauber Costa55007d82012-12-18 14:22:38 -0800648 }
Glauber Costabea207c2012-12-18 14:22:11 -0800649 /*
650 * This check can't live in kmem destruction function,
651 * since the charges will outlive the cgroup
652 */
653 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800654}
655#else
656static void disarm_kmem_keys(struct mem_cgroup *memcg)
657{
658}
659#endif /* CONFIG_MEMCG_KMEM */
660
661static void disarm_static_keys(struct mem_cgroup *memcg)
662{
663 disarm_sock_keys(memcg);
664 disarm_kmem_keys(memcg);
665}
666
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700667static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800668
Balbir Singhf64c3f52009-09-23 15:56:37 -0700669static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700670mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700671{
Jianyu Zhane2318752014-06-06 14:38:20 -0700672 int nid = zone_to_nid(zone);
673 int zid = zone_idx(zone);
674
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700675 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700676}
677
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700678struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100679{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100681}
682
Balbir Singhf64c3f52009-09-23 15:56:37 -0700683static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700684mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700685{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700686 int nid = page_to_nid(page);
687 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700688
Jianyu Zhane2318752014-06-06 14:38:20 -0700689 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700690}
691
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700692static struct mem_cgroup_tree_per_zone *
693soft_limit_tree_node_zone(int nid, int zid)
694{
695 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
696}
697
698static struct mem_cgroup_tree_per_zone *
699soft_limit_tree_from_page(struct page *page)
700{
701 int nid = page_to_nid(page);
702 int zid = page_zonenum(page);
703
704 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
705}
706
Johannes Weinercf2c8122014-06-06 14:38:21 -0700707static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
708 struct mem_cgroup_tree_per_zone *mctz,
709 unsigned long long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700710{
711 struct rb_node **p = &mctz->rb_root.rb_node;
712 struct rb_node *parent = NULL;
713 struct mem_cgroup_per_zone *mz_node;
714
715 if (mz->on_tree)
716 return;
717
718 mz->usage_in_excess = new_usage_in_excess;
719 if (!mz->usage_in_excess)
720 return;
721 while (*p) {
722 parent = *p;
723 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
724 tree_node);
725 if (mz->usage_in_excess < mz_node->usage_in_excess)
726 p = &(*p)->rb_left;
727 /*
728 * We can't avoid mem cgroups that are over their soft
729 * limit by the same amount
730 */
731 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
732 p = &(*p)->rb_right;
733 }
734 rb_link_node(&mz->tree_node, parent, p);
735 rb_insert_color(&mz->tree_node, &mctz->rb_root);
736 mz->on_tree = true;
737}
738
Johannes Weinercf2c8122014-06-06 14:38:21 -0700739static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
740 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700741{
742 if (!mz->on_tree)
743 return;
744 rb_erase(&mz->tree_node, &mctz->rb_root);
745 mz->on_tree = false;
746}
747
Johannes Weinercf2c8122014-06-06 14:38:21 -0700748static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
749 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700750{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700751 unsigned long flags;
752
753 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700754 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700755 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700756}
757
758
759static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
760{
761 unsigned long long excess;
762 struct mem_cgroup_per_zone *mz;
763 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700764
Jianyu Zhane2318752014-06-06 14:38:20 -0700765 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700766 /*
767 * Necessary to update all ancestors when hierarchy is used.
768 * because their event counter is not touched.
769 */
770 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700771 mz = mem_cgroup_page_zoneinfo(memcg, page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700772 excess = res_counter_soft_limit_excess(&memcg->res);
773 /*
774 * We have to update the tree if mz is on RB-tree or
775 * mem is over its softlimit.
776 */
777 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700778 unsigned long flags;
779
780 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700781 /* if on-tree, remove it */
782 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700783 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700784 /*
785 * Insert again. mz->usage_in_excess will be updated.
786 * If excess is 0, no tree ops.
787 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700788 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700789 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700790 }
791 }
792}
793
794static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
795{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700796 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700797 struct mem_cgroup_per_zone *mz;
798 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700799
Jianyu Zhane2318752014-06-06 14:38:20 -0700800 for_each_node(nid) {
801 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
802 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
803 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700804 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700805 }
806 }
807}
808
809static struct mem_cgroup_per_zone *
810__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
811{
812 struct rb_node *rightmost = NULL;
813 struct mem_cgroup_per_zone *mz;
814
815retry:
816 mz = NULL;
817 rightmost = rb_last(&mctz->rb_root);
818 if (!rightmost)
819 goto done; /* Nothing to reclaim from */
820
821 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
822 /*
823 * Remove the node now but someone else can add it back,
824 * we will to add it back at the end of reclaim to its correct
825 * position in the tree.
826 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700827 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700828 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400829 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700830 goto retry;
831done:
832 return mz;
833}
834
835static struct mem_cgroup_per_zone *
836mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
837{
838 struct mem_cgroup_per_zone *mz;
839
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700840 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700841 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700842 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700843 return mz;
844}
845
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700846/*
847 * Implementation Note: reading percpu statistics for memcg.
848 *
849 * Both of vmstat[] and percpu_counter has threshold and do periodic
850 * synchronization to implement "quick" read. There are trade-off between
851 * reading cost and precision of value. Then, we may have a chance to implement
852 * a periodic synchronizion of counter in memcg's counter.
853 *
854 * But this _read() function is used for user interface now. The user accounts
855 * memory usage by memory cgroup and he _always_ requires exact value because
856 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
857 * have to visit all online cpus and make sum. So, for now, unnecessary
858 * synchronization is not implemented. (just implemented for cpu hotplug)
859 *
860 * If there are kernel internal actions which can make use of some not-exact
861 * value, and reading all cpu value can be performance bottleneck in some
862 * common workload, threashold and synchonization as vmstat[] should be
863 * implemented.
864 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700865static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700866 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800867{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700868 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800869 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800870
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700871 get_online_cpus();
872 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700873 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700874#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700875 spin_lock(&memcg->pcp_counter_lock);
876 val += memcg->nocpu_base.count[idx];
877 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700878#endif
879 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800880 return val;
881}
882
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700883static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700884 enum mem_cgroup_events_index idx)
885{
886 unsigned long val = 0;
887 int cpu;
888
David Rientjes9c567512013-10-16 13:46:43 -0700889 get_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700890 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700891 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700892#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700893 spin_lock(&memcg->pcp_counter_lock);
894 val += memcg->nocpu_base.events[idx];
895 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700896#endif
David Rientjes9c567512013-10-16 13:46:43 -0700897 put_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700898 return val;
899}
900
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700901static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700902 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700903 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800904{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700905 /*
906 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
907 * counted as CACHE even if it's on ANON LRU.
908 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700909 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700910 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700911 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800912 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700913 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700914 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700915
David Rientjesb070e652013-05-07 16:18:09 -0700916 if (PageTransHuge(page))
917 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
918 nr_pages);
919
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800920 /* pagein of a big page is an event. So, ignore page size */
921 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700922 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800923 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700924 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800925 nr_pages = -nr_pages; /* for event */
926 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800927
Johannes Weiner13114712012-05-29 15:07:07 -0700928 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800929}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800930
Jianyu Zhane2318752014-06-06 14:38:20 -0700931unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700932{
933 struct mem_cgroup_per_zone *mz;
934
935 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
936 return mz->lru_size[lru];
937}
938
Jianyu Zhane2318752014-06-06 14:38:20 -0700939static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
940 int nid,
941 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700942{
Jianyu Zhane2318752014-06-06 14:38:20 -0700943 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700944 int zid;
945
Jianyu Zhane2318752014-06-06 14:38:20 -0700946 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700947
Jianyu Zhane2318752014-06-06 14:38:20 -0700948 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
949 struct mem_cgroup_per_zone *mz;
950 enum lru_list lru;
951
952 for_each_lru(lru) {
953 if (!(BIT(lru) & lru_mask))
954 continue;
955 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
956 nr += mz->lru_size[lru];
957 }
958 }
959 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700960}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700961
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700962static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700963 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800964{
Jianyu Zhane2318752014-06-06 14:38:20 -0700965 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700966 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800967
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800968 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700969 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
970 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800971}
972
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800973static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
974 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800975{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700976 unsigned long val, next;
977
Johannes Weiner13114712012-05-29 15:07:07 -0700978 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700979 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700980 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800981 if ((long)next - (long)val < 0) {
982 switch (target) {
983 case MEM_CGROUP_TARGET_THRESH:
984 next = val + THRESHOLDS_EVENTS_TARGET;
985 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700986 case MEM_CGROUP_TARGET_SOFTLIMIT:
987 next = val + SOFTLIMIT_EVENTS_TARGET;
988 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800989 case MEM_CGROUP_TARGET_NUMAINFO:
990 next = val + NUMAINFO_EVENTS_TARGET;
991 break;
992 default:
993 break;
994 }
995 __this_cpu_write(memcg->stat->targets[target], next);
996 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700997 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800998 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800999}
1000
1001/*
1002 * Check events in order.
1003 *
1004 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001005static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001006{
1007 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001008 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1009 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001010 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001011 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001012
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001013 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1014 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001015#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001016 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1017 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001018#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001019 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001020 if (unlikely(do_softlimit))
1021 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001022#if MAX_NUMNODES > 1
1023 if (unlikely(do_numainfo))
1024 atomic_inc(&memcg->numainfo_events);
1025#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001026 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001027}
1028
Balbir Singhcf475ad2008-04-29 01:00:16 -07001029struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001030{
Balbir Singh31a78f22008-09-28 23:09:31 +01001031 /*
1032 * mm_update_next_owner() may clear mm->owner to NULL
1033 * if it races with swapoff, page migration, etc.
1034 * So this can be called with p == NULL.
1035 */
1036 if (unlikely(!p))
1037 return NULL;
1038
Tejun Heo073219e2014-02-08 10:36:58 -05001039 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001040}
1041
Johannes Weinerdf381972014-04-07 15:37:43 -07001042static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001043{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001044 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001045
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001046 rcu_read_lock();
1047 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -07001048 /*
1049 * Page cache insertions can happen withou an
1050 * actual mm context, e.g. during disk probing
1051 * on boot, loopback IO, acct() writes etc.
1052 */
1053 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -07001054 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -07001055 else {
1056 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1057 if (unlikely(!memcg))
1058 memcg = root_mem_cgroup;
1059 }
Tejun Heoec903c02014-05-13 12:11:01 -04001060 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001061 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001062 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001063}
1064
Michal Hocko16248d82013-04-29 15:07:19 -07001065/*
1066 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1067 * ref. count) or NULL if the whole root's subtree has been visited.
1068 *
1069 * helper function to be used by mem_cgroup_iter
1070 */
1071static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001072 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001073{
Tejun Heo492eb212013-08-08 20:11:25 -04001074 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001075
Tejun Heobd8815a2013-08-08 20:11:27 -04001076 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001077skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001078 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001079
1080 /*
1081 * Even if we found a group we have to make sure it is
1082 * alive. css && !memcg means that the groups should be
1083 * skipped and we should continue the tree walk.
1084 * last_visited css is safe to use because it is
1085 * protected by css_get and the tree walk is rcu safe.
Michal Hocko0eef6152014-01-23 15:53:37 -08001086 *
1087 * We do not take a reference on the root of the tree walk
1088 * because we might race with the root removal when it would
1089 * be the only node in the iterated hierarchy and mem_cgroup_iter
1090 * would end up in an endless loop because it expects that at
1091 * least one valid node will be returned. Root cannot disappear
1092 * because caller of the iterator should hold it already so
1093 * skipping css reference should be safe.
Michal Hocko16248d82013-04-29 15:07:19 -07001094 */
Tejun Heo492eb212013-08-08 20:11:25 -04001095 if (next_css) {
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07001096 struct mem_cgroup *memcg = mem_cgroup_from_css(next_css);
1097
1098 if (next_css == &root->css)
1099 return memcg;
1100
1101 if (css_tryget_online(next_css)) {
1102 /*
1103 * Make sure the memcg is initialized:
1104 * mem_cgroup_css_online() orders the the
1105 * initialization against setting the flag.
1106 */
1107 if (smp_load_acquire(&memcg->initialized))
1108 return memcg;
1109 css_put(next_css);
1110 }
Michal Hocko0eef6152014-01-23 15:53:37 -08001111
1112 prev_css = next_css;
1113 goto skip_node;
Michal Hocko16248d82013-04-29 15:07:19 -07001114 }
1115
1116 return NULL;
1117}
1118
Johannes Weiner519ebea2013-07-03 15:04:51 -07001119static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1120{
1121 /*
1122 * When a group in the hierarchy below root is destroyed, the
1123 * hierarchy iterator can no longer be trusted since it might
1124 * have pointed to the destroyed group. Invalidate it.
1125 */
1126 atomic_inc(&root->dead_count);
1127}
1128
1129static struct mem_cgroup *
1130mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1131 struct mem_cgroup *root,
1132 int *sequence)
1133{
1134 struct mem_cgroup *position = NULL;
1135 /*
1136 * A cgroup destruction happens in two stages: offlining and
1137 * release. They are separated by a RCU grace period.
1138 *
1139 * If the iterator is valid, we may still race with an
1140 * offlining. The RCU lock ensures the object won't be
1141 * released, tryget will fail if we lost the race.
1142 */
1143 *sequence = atomic_read(&root->dead_count);
1144 if (iter->last_dead_count == *sequence) {
1145 smp_rmb();
1146 position = iter->last_visited;
Michal Hockoecc736f2014-01-23 15:53:35 -08001147
1148 /*
1149 * We cannot take a reference to root because we might race
1150 * with root removal and returning NULL would end up in
1151 * an endless loop on the iterator user level when root
1152 * would be returned all the time.
1153 */
1154 if (position && position != root &&
Tejun Heoec903c02014-05-13 12:11:01 -04001155 !css_tryget_online(&position->css))
Johannes Weiner519ebea2013-07-03 15:04:51 -07001156 position = NULL;
1157 }
1158 return position;
1159}
1160
1161static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1162 struct mem_cgroup *last_visited,
1163 struct mem_cgroup *new_position,
Michal Hockoecc736f2014-01-23 15:53:35 -08001164 struct mem_cgroup *root,
Johannes Weiner519ebea2013-07-03 15:04:51 -07001165 int sequence)
1166{
Michal Hockoecc736f2014-01-23 15:53:35 -08001167 /* root reference counting symmetric to mem_cgroup_iter_load */
1168 if (last_visited && last_visited != root)
Johannes Weiner519ebea2013-07-03 15:04:51 -07001169 css_put(&last_visited->css);
1170 /*
1171 * We store the sequence count from the time @last_visited was
1172 * loaded successfully instead of rereading it here so that we
1173 * don't lose destruction events in between. We could have
1174 * raced with the destruction of @new_position after all.
1175 */
1176 iter->last_visited = new_position;
1177 smp_wmb();
1178 iter->last_dead_count = sequence;
1179}
1180
Johannes Weiner56600482012-01-12 17:17:59 -08001181/**
1182 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1183 * @root: hierarchy root
1184 * @prev: previously returned memcg, NULL on first invocation
1185 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1186 *
1187 * Returns references to children of the hierarchy below @root, or
1188 * @root itself, or %NULL after a full round-trip.
1189 *
1190 * Caller must pass the return value in @prev on subsequent
1191 * invocations for reference counting, or use mem_cgroup_iter_break()
1192 * to cancel a hierarchy walk before the round-trip is complete.
1193 *
1194 * Reclaimers can specify a zone and a priority level in @reclaim to
1195 * divide up the memcgs in the hierarchy among all concurrent
1196 * reclaimers operating on the same zone and priority.
1197 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001198struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001199 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001200 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001201{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001202 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001203 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001204
Andrew Morton694fbc02013-09-24 15:27:37 -07001205 if (mem_cgroup_disabled())
1206 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001207
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001208 if (!root)
1209 root = root_mem_cgroup;
1210
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001211 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001212 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001213
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001214 if (!root->use_hierarchy && root != root_mem_cgroup) {
1215 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001216 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001217 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001218 }
1219
Michal Hocko542f85f2013-04-29 15:07:15 -07001220 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001221 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001222 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001223 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001224
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001225 if (reclaim) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001226 struct mem_cgroup_per_zone *mz;
1227
Jianyu Zhane2318752014-06-06 14:38:20 -07001228 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001229 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001230 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001231 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001232 goto out_unlock;
1233 }
Michal Hocko5f578162013-04-29 15:07:17 -07001234
Johannes Weiner519ebea2013-07-03 15:04:51 -07001235 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001236 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001237
Andrew Morton694fbc02013-09-24 15:27:37 -07001238 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001239
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001240 if (reclaim) {
Michal Hockoecc736f2014-01-23 15:53:35 -08001241 mem_cgroup_iter_update(iter, last_visited, memcg, root,
1242 seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001243
Michal Hocko19f39402013-04-29 15:07:18 -07001244 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001245 iter->generation++;
1246 else if (!prev && memcg)
1247 reclaim->generation = iter->generation;
1248 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001249
Andrew Morton694fbc02013-09-24 15:27:37 -07001250 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001251 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001252 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001253out_unlock:
1254 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001255out_css_put:
1256 if (prev && prev != root)
1257 css_put(&prev->css);
1258
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001259 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001260}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001261
Johannes Weiner56600482012-01-12 17:17:59 -08001262/**
1263 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1264 * @root: hierarchy root
1265 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1266 */
1267void mem_cgroup_iter_break(struct mem_cgroup *root,
1268 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001269{
1270 if (!root)
1271 root = root_mem_cgroup;
1272 if (prev && prev != root)
1273 css_put(&prev->css);
1274}
1275
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001276/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001277 * Iteration constructs for visiting all cgroups (under a tree). If
1278 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1279 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001280 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001281#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001282 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001283 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001284 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001285
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001286#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001287 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001288 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001289 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001290
David Rientjes68ae5642012-12-12 13:51:57 -08001291void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001292{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001293 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001294
Ying Han456f9982011-05-26 16:25:38 -07001295 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001296 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1297 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001298 goto out;
1299
1300 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001301 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001302 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1303 break;
1304 case PGMAJFAULT:
1305 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001306 break;
1307 default:
1308 BUG();
1309 }
1310out:
1311 rcu_read_unlock();
1312}
David Rientjes68ae5642012-12-12 13:51:57 -08001313EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001314
Johannes Weiner925b7672012-01-12 17:18:15 -08001315/**
1316 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1317 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001318 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001319 *
1320 * Returns the lru list vector holding pages for the given @zone and
1321 * @mem. This can be the global zone lruvec, if the memory controller
1322 * is disabled.
1323 */
1324struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1325 struct mem_cgroup *memcg)
1326{
1327 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001328 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001329
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001330 if (mem_cgroup_disabled()) {
1331 lruvec = &zone->lruvec;
1332 goto out;
1333 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001334
Jianyu Zhane2318752014-06-06 14:38:20 -07001335 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001336 lruvec = &mz->lruvec;
1337out:
1338 /*
1339 * Since a node can be onlined after the mem_cgroup was created,
1340 * we have to be prepared to initialize lruvec->zone here;
1341 * and if offlined then reonlined, we need to reinitialize it.
1342 */
1343 if (unlikely(lruvec->zone != zone))
1344 lruvec->zone = zone;
1345 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001346}
1347
Johannes Weiner925b7672012-01-12 17:18:15 -08001348/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001349 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001350 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001351 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001352 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001353struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001354{
1355 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001356 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001357 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001358 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001359
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001360 if (mem_cgroup_disabled()) {
1361 lruvec = &zone->lruvec;
1362 goto out;
1363 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001364
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001365 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001366 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001367
1368 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001369 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001370 * an uncharged page off lru does nothing to secure
1371 * its former mem_cgroup from sudden removal.
1372 *
1373 * Our caller holds lru_lock, and PageCgroupUsed is updated
1374 * under page_cgroup lock: between them, they make all uses
1375 * of pc->mem_cgroup safe.
1376 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001377 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001378 pc->mem_cgroup = memcg = root_mem_cgroup;
1379
Jianyu Zhane2318752014-06-06 14:38:20 -07001380 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001381 lruvec = &mz->lruvec;
1382out:
1383 /*
1384 * Since a node can be onlined after the mem_cgroup was created,
1385 * we have to be prepared to initialize lruvec->zone here;
1386 * and if offlined then reonlined, we need to reinitialize it.
1387 */
1388 if (unlikely(lruvec->zone != zone))
1389 lruvec->zone = zone;
1390 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001391}
1392
1393/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001394 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1395 * @lruvec: mem_cgroup per zone lru vector
1396 * @lru: index of lru list the page is sitting on
1397 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001398 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001399 * This function must be called when a page is added to or removed from an
1400 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001401 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001402void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1403 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001404{
1405 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001406 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001407
1408 if (mem_cgroup_disabled())
1409 return;
1410
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001411 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1412 lru_size = mz->lru_size + lru;
1413 *lru_size += nr_pages;
1414 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001415}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001416
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001417/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001418 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001419 * hierarchy subtree
1420 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001421bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1422 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001423{
Johannes Weiner91c637342012-05-29 15:06:24 -07001424 if (root_memcg == memcg)
1425 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001426 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001427 return false;
Li Zefanb47f77b2013-09-23 16:55:43 +08001428 return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001429}
1430
1431static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1432 struct mem_cgroup *memcg)
1433{
1434 bool ret;
1435
Johannes Weiner91c637342012-05-29 15:06:24 -07001436 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001437 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001438 rcu_read_unlock();
1439 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001440}
1441
David Rientjesffbdccf2013-07-03 15:01:23 -07001442bool task_in_mem_cgroup(struct task_struct *task,
1443 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001444{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001445 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001446 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001447 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001448
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001449 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001450 if (p) {
Johannes Weinerdf381972014-04-07 15:37:43 -07001451 curr = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001452 task_unlock(p);
1453 } else {
1454 /*
1455 * All threads may have already detached their mm's, but the oom
1456 * killer still needs to detect if they have already been oom
1457 * killed to prevent needlessly killing additional tasks.
1458 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001459 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001460 curr = mem_cgroup_from_task(task);
1461 if (curr)
1462 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001463 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001464 }
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001465 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001466 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001467 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001468 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1469 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001470 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001471 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001472 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001473 return ret;
1474}
1475
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001476int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001477{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001478 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001479 unsigned long inactive;
1480 unsigned long active;
1481 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001482
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001483 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1484 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001485
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001486 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1487 if (gb)
1488 inactive_ratio = int_sqrt(10 * gb);
1489 else
1490 inactive_ratio = 1;
1491
Johannes Weiner9b272972011-11-02 13:38:23 -07001492 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001493}
1494
Balbir Singh6d61ef42009-01-07 18:08:06 -08001495#define mem_cgroup_from_res_counter(counter, member) \
1496 container_of(counter, struct mem_cgroup, member)
1497
Johannes Weiner19942822011-02-01 15:52:43 -08001498/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001499 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001500 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001501 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001502 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001503 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001504 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001505static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001506{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001507 unsigned long long margin;
1508
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001509 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001510 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001511 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001512 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001513}
1514
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001515int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001516{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001517 /* root ? */
Linus Torvalds14208b02014-06-09 15:03:33 -07001518 if (mem_cgroup_disabled() || !memcg->css.parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001519 return vm_swappiness;
1520
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001521 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001522}
1523
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001524/*
1525 * memcg->moving_account is used for checking possibility that some thread is
1526 * calling move_account(). When a thread on CPU-A starts moving pages under
1527 * a memcg, other threads should check memcg->moving_account under
1528 * rcu_read_lock(), like this:
1529 *
1530 * CPU-A CPU-B
1531 * rcu_read_lock()
1532 * memcg->moving_account+1 if (memcg->mocing_account)
1533 * take heavy locks.
1534 * synchronize_rcu() update something.
1535 * rcu_read_unlock()
1536 * start move here.
1537 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001538
1539/* for quick checking without looking up memcg */
1540atomic_t memcg_moving __read_mostly;
1541
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001542static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001543{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001544 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001545 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001546 synchronize_rcu();
1547}
1548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001549static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001550{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001551 /*
1552 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1553 * We check NULL in callee rather than caller.
1554 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001555 if (memcg) {
1556 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001557 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001558 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001559}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001560
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001561/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001562 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001563 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001564 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1565 * moving cgroups. This is for waiting at high-memory pressure
1566 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001567 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001568static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001569{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001570 struct mem_cgroup *from;
1571 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001572 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001573 /*
1574 * Unlike task_move routines, we access mc.to, mc.from not under
1575 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1576 */
1577 spin_lock(&mc.lock);
1578 from = mc.from;
1579 to = mc.to;
1580 if (!from)
1581 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001582
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583 ret = mem_cgroup_same_or_subtree(memcg, from)
1584 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001585unlock:
1586 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001587 return ret;
1588}
1589
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001590static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001591{
1592 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001593 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001594 DEFINE_WAIT(wait);
1595 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1596 /* moving charge context might have finished. */
1597 if (mc.moving_task)
1598 schedule();
1599 finish_wait(&mc.waitq, &wait);
1600 return true;
1601 }
1602 }
1603 return false;
1604}
1605
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001606/*
1607 * Take this lock when
1608 * - a code tries to modify page's memcg while it's USED.
1609 * - a code tries to modify page state accounting in a memcg.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001610 */
1611static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1612 unsigned long *flags)
1613{
1614 spin_lock_irqsave(&memcg->move_lock, *flags);
1615}
1616
1617static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1618 unsigned long *flags)
1619{
1620 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1621}
1622
Sha Zhengju58cf1882013-02-22 16:32:05 -08001623#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001624/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001625 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001626 * @memcg: The memory cgroup that went over limit
1627 * @p: Task that is going to be killed
1628 *
1629 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1630 * enabled
1631 */
1632void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1633{
Tejun Heoe61734c2014-02-12 09:29:50 -05001634 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001635 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001636 struct mem_cgroup *iter;
1637 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001638
Sha Zhengju58cf1882013-02-22 16:32:05 -08001639 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001640 return;
1641
Michal Hocko08088cb2014-02-25 15:01:44 -08001642 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001643 rcu_read_lock();
1644
Tejun Heoe61734c2014-02-12 09:29:50 -05001645 pr_info("Task in ");
1646 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1647 pr_info(" killed as a result of limit of ");
1648 pr_cont_cgroup_path(memcg->css.cgroup);
1649 pr_info("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001650
Balbir Singhe2224322009-04-02 16:57:39 -07001651 rcu_read_unlock();
1652
Andrew Mortond0451972013-02-22 16:32:06 -08001653 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001654 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1655 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1656 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001657 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001658 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1659 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1660 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001661 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001662 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1663 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1664 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001665
1666 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001667 pr_info("Memory cgroup stats for ");
1668 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001669 pr_cont(":");
1670
1671 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1672 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1673 continue;
1674 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1675 K(mem_cgroup_read_stat(iter, i)));
1676 }
1677
1678 for (i = 0; i < NR_LRU_LISTS; i++)
1679 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1680 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1681
1682 pr_cont("\n");
1683 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001684 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001685}
1686
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001687/*
1688 * This function returns the number of memcg under hierarchy tree. Returns
1689 * 1(self count) if no children.
1690 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001691static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001692{
1693 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001694 struct mem_cgroup *iter;
1695
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001696 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001697 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001698 return num;
1699}
1700
Balbir Singh6d61ef42009-01-07 18:08:06 -08001701/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001702 * Return the memory (and swap, if configured) limit for a memcg.
1703 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001704static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001705{
1706 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001707
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001708 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001709
David Rientjesa63d83f2010-08-09 17:19:46 -07001710 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001711 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001712 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001713 if (mem_cgroup_swappiness(memcg)) {
1714 u64 memsw;
1715
1716 limit += total_swap_pages << PAGE_SHIFT;
1717 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1718
1719 /*
1720 * If memsw is finite and limits the amount of swap space
1721 * available to this memcg, return that limit.
1722 */
1723 limit = min(limit, memsw);
1724 }
1725
1726 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001727}
1728
David Rientjes19965462012-12-11 16:00:26 -08001729static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1730 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001731{
1732 struct mem_cgroup *iter;
1733 unsigned long chosen_points = 0;
1734 unsigned long totalpages;
1735 unsigned int points = 0;
1736 struct task_struct *chosen = NULL;
1737
David Rientjes876aafb2012-07-31 16:43:48 -07001738 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001739 * If current has a pending SIGKILL or is exiting, then automatically
1740 * select it. The goal is to allow it to allocate so that it may
1741 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001742 */
David Rientjes465adcf2013-04-29 15:08:45 -07001743 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001744 set_thread_flag(TIF_MEMDIE);
1745 return;
1746 }
1747
1748 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001749 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1750 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001751 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001752 struct task_struct *task;
1753
Tejun Heo72ec7022013-08-08 20:11:26 -04001754 css_task_iter_start(&iter->css, &it);
1755 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001756 switch (oom_scan_process_thread(task, totalpages, NULL,
1757 false)) {
1758 case OOM_SCAN_SELECT:
1759 if (chosen)
1760 put_task_struct(chosen);
1761 chosen = task;
1762 chosen_points = ULONG_MAX;
1763 get_task_struct(chosen);
1764 /* fall through */
1765 case OOM_SCAN_CONTINUE:
1766 continue;
1767 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001768 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001769 mem_cgroup_iter_break(memcg, iter);
1770 if (chosen)
1771 put_task_struct(chosen);
1772 return;
1773 case OOM_SCAN_OK:
1774 break;
1775 };
1776 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001777 if (!points || points < chosen_points)
1778 continue;
1779 /* Prefer thread group leaders for display purposes */
1780 if (points == chosen_points &&
1781 thread_group_leader(chosen))
1782 continue;
1783
1784 if (chosen)
1785 put_task_struct(chosen);
1786 chosen = task;
1787 chosen_points = points;
1788 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001789 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001790 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001791 }
1792
1793 if (!chosen)
1794 return;
1795 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001796 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1797 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001798}
1799
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001800/**
1801 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001802 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001803 * @nid: the node ID to be checked.
1804 * @noswap : specify true here if the user wants flle only information.
1805 *
1806 * This function returns whether the specified memcg contains any
1807 * reclaimable pages on a node. Returns true if there are any reclaimable
1808 * pages in the node.
1809 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001810static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001811 int nid, bool noswap)
1812{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001813 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001814 return true;
1815 if (noswap || !total_swap_pages)
1816 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001817 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001818 return true;
1819 return false;
1820
1821}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001822#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001823
1824/*
1825 * Always updating the nodemask is not very good - even if we have an empty
1826 * list or the wrong list here, we can start from some node and traverse all
1827 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1828 *
1829 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001830static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001831{
1832 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001833 /*
1834 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1835 * pagein/pageout changes since the last update.
1836 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001837 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001838 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001839 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001840 return;
1841
Ying Han889976d2011-05-26 16:25:33 -07001842 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001843 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001844
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001845 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001846
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001847 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1848 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001849 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001850
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001851 atomic_set(&memcg->numainfo_events, 0);
1852 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001853}
1854
1855/*
1856 * Selecting a node where we start reclaim from. Because what we need is just
1857 * reducing usage counter, start from anywhere is O,K. Considering
1858 * memory reclaim from current node, there are pros. and cons.
1859 *
1860 * Freeing memory from current node means freeing memory from a node which
1861 * we'll use or we've used. So, it may make LRU bad. And if several threads
1862 * hit limits, it will see a contention on a node. But freeing from remote
1863 * node means more costs for memory reclaim because of memory latency.
1864 *
1865 * Now, we use round-robin. Better algorithm is welcomed.
1866 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001867int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001868{
1869 int node;
1870
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001871 mem_cgroup_may_update_nodemask(memcg);
1872 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001873
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001874 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001875 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001876 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001877 /*
1878 * We call this when we hit limit, not when pages are added to LRU.
1879 * No LRU may hold pages because all pages are UNEVICTABLE or
1880 * memcg is too small and all pages are not on LRU. In that case,
1881 * we use curret node.
1882 */
1883 if (unlikely(node == MAX_NUMNODES))
1884 node = numa_node_id();
1885
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001886 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001887 return node;
1888}
1889
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001890/*
1891 * Check all nodes whether it contains reclaimable pages or not.
1892 * For quick scan, we make use of scan_nodes. This will allow us to skip
1893 * unused nodes. But scan_nodes is lazily updated and may not cotain
1894 * enough new information. We need to do double check.
1895 */
1896static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1897{
1898 int nid;
1899
1900 /*
1901 * quick check...making use of scan_node.
1902 * We can skip unused nodes.
1903 */
1904 if (!nodes_empty(memcg->scan_nodes)) {
1905 for (nid = first_node(memcg->scan_nodes);
1906 nid < MAX_NUMNODES;
1907 nid = next_node(nid, memcg->scan_nodes)) {
1908
1909 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1910 return true;
1911 }
1912 }
1913 /*
1914 * Check rest of nodes.
1915 */
1916 for_each_node_state(nid, N_MEMORY) {
1917 if (node_isset(nid, memcg->scan_nodes))
1918 continue;
1919 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1920 return true;
1921 }
1922 return false;
1923}
1924
Ying Han889976d2011-05-26 16:25:33 -07001925#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001927{
1928 return 0;
1929}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001930
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001931static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1932{
1933 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
1934}
Ying Han889976d2011-05-26 16:25:33 -07001935#endif
1936
Andrew Morton0608f432013-09-24 15:27:41 -07001937static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1938 struct zone *zone,
1939 gfp_t gfp_mask,
1940 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001941{
Andrew Morton0608f432013-09-24 15:27:41 -07001942 struct mem_cgroup *victim = NULL;
1943 int total = 0;
1944 int loop = 0;
1945 unsigned long excess;
1946 unsigned long nr_scanned;
1947 struct mem_cgroup_reclaim_cookie reclaim = {
1948 .zone = zone,
1949 .priority = 0,
1950 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001951
Andrew Morton0608f432013-09-24 15:27:41 -07001952 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001953
Andrew Morton0608f432013-09-24 15:27:41 -07001954 while (1) {
1955 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1956 if (!victim) {
1957 loop++;
1958 if (loop >= 2) {
1959 /*
1960 * If we have not been able to reclaim
1961 * anything, it might because there are
1962 * no reclaimable pages under this hierarchy
1963 */
1964 if (!total)
1965 break;
1966 /*
1967 * We want to do more targeted reclaim.
1968 * excess >> 2 is not to excessive so as to
1969 * reclaim too much, nor too less that we keep
1970 * coming back to reclaim from this cgroup
1971 */
1972 if (total >= (excess >> 2) ||
1973 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1974 break;
1975 }
1976 continue;
1977 }
1978 if (!mem_cgroup_reclaimable(victim, false))
1979 continue;
1980 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1981 zone, &nr_scanned);
1982 *total_scanned += nr_scanned;
1983 if (!res_counter_soft_limit_excess(&root_memcg->res))
1984 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001985 }
Andrew Morton0608f432013-09-24 15:27:41 -07001986 mem_cgroup_iter_break(root_memcg, victim);
1987 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001988}
1989
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001990#ifdef CONFIG_LOCKDEP
1991static struct lockdep_map memcg_oom_lock_dep_map = {
1992 .name = "memcg_oom_lock",
1993};
1994#endif
1995
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001996static DEFINE_SPINLOCK(memcg_oom_lock);
1997
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001998/*
1999 * Check OOM-Killer is already running under our hierarchy.
2000 * If someone is running, return false.
2001 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002002static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002003{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002004 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002005
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002006 spin_lock(&memcg_oom_lock);
2007
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002008 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002009 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002010 /*
2011 * this subtree of our hierarchy is already locked
2012 * so we cannot give a lock.
2013 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002014 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002015 mem_cgroup_iter_break(memcg, iter);
2016 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002017 } else
2018 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002019 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002020
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002021 if (failed) {
2022 /*
2023 * OK, we failed to lock the whole subtree so we have
2024 * to clean up what we set up to the failing subtree
2025 */
2026 for_each_mem_cgroup_tree(iter, memcg) {
2027 if (iter == failed) {
2028 mem_cgroup_iter_break(memcg, iter);
2029 break;
2030 }
2031 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002032 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002033 } else
2034 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002035
2036 spin_unlock(&memcg_oom_lock);
2037
2038 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002039}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002040
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002041static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002042{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002043 struct mem_cgroup *iter;
2044
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002045 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002046 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002047 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002048 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002049 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002050}
2051
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002052static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002053{
2054 struct mem_cgroup *iter;
2055
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002056 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002057 atomic_inc(&iter->under_oom);
2058}
2059
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002060static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002061{
2062 struct mem_cgroup *iter;
2063
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002064 /*
2065 * When a new child is created while the hierarchy is under oom,
2066 * mem_cgroup_oom_lock() may not be called. We have to use
2067 * atomic_add_unless() here.
2068 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002069 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002070 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002071}
2072
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002073static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2074
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002075struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002076 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002077 wait_queue_t wait;
2078};
2079
2080static int memcg_oom_wake_function(wait_queue_t *wait,
2081 unsigned mode, int sync, void *arg)
2082{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002083 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2084 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002085 struct oom_wait_info *oom_wait_info;
2086
2087 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002088 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002089
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002090 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002091 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002092 * Then we can use css_is_ancestor without taking care of RCU.
2093 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002094 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2095 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002096 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002097 return autoremove_wake_function(wait, mode, sync, arg);
2098}
2099
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002100static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002101{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002102 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002103 /* for filtering, pass "memcg" as argument. */
2104 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002105}
2106
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002107static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002108{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002109 if (memcg && atomic_read(&memcg->under_oom))
2110 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002111}
2112
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002113static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002114{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002115 if (!current->memcg_oom.may_oom)
2116 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002117 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002118 * We are in the middle of the charge context here, so we
2119 * don't want to block when potentially sitting on a callstack
2120 * that holds all kinds of filesystem and mm locks.
2121 *
2122 * Also, the caller may handle a failed allocation gracefully
2123 * (like optional page cache readahead) and so an OOM killer
2124 * invocation might not even be necessary.
2125 *
2126 * That's why we don't do anything here except remember the
2127 * OOM context and then deal with it at the end of the page
2128 * fault when the stack is unwound, the locks are released,
2129 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002130 */
Johannes Weiner49426422013-10-16 13:46:59 -07002131 css_get(&memcg->css);
2132 current->memcg_oom.memcg = memcg;
2133 current->memcg_oom.gfp_mask = mask;
2134 current->memcg_oom.order = order;
2135}
2136
2137/**
2138 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2139 * @handle: actually kill/wait or just clean up the OOM state
2140 *
2141 * This has to be called at the end of a page fault if the memcg OOM
2142 * handler was enabled.
2143 *
2144 * Memcg supports userspace OOM handling where failed allocations must
2145 * sleep on a waitqueue until the userspace task resolves the
2146 * situation. Sleeping directly in the charge context with all kinds
2147 * of locks held is not a good idea, instead we remember an OOM state
2148 * in the task and mem_cgroup_oom_synchronize() has to be called at
2149 * the end of the page fault to complete the OOM handling.
2150 *
2151 * Returns %true if an ongoing memcg OOM situation was detected and
2152 * completed, %false otherwise.
2153 */
2154bool mem_cgroup_oom_synchronize(bool handle)
2155{
2156 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2157 struct oom_wait_info owait;
2158 bool locked;
2159
2160 /* OOM is global, do not handle */
2161 if (!memcg)
2162 return false;
2163
2164 if (!handle)
2165 goto cleanup;
2166
2167 owait.memcg = memcg;
2168 owait.wait.flags = 0;
2169 owait.wait.func = memcg_oom_wake_function;
2170 owait.wait.private = current;
2171 INIT_LIST_HEAD(&owait.wait.task_list);
2172
2173 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002174 mem_cgroup_mark_under_oom(memcg);
2175
2176 locked = mem_cgroup_oom_trylock(memcg);
2177
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002178 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002179 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002180
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002181 if (locked && !memcg->oom_kill_disable) {
2182 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002183 finish_wait(&memcg_oom_waitq, &owait.wait);
2184 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2185 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002186 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002187 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002188 mem_cgroup_unmark_under_oom(memcg);
2189 finish_wait(&memcg_oom_waitq, &owait.wait);
2190 }
2191
2192 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002193 mem_cgroup_oom_unlock(memcg);
2194 /*
2195 * There is no guarantee that an OOM-lock contender
2196 * sees the wakeups triggered by the OOM kill
2197 * uncharges. Wake any sleepers explicitely.
2198 */
2199 memcg_oom_recover(memcg);
2200 }
Johannes Weiner49426422013-10-16 13:46:59 -07002201cleanup:
2202 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002203 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002204 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002205}
2206
Balbir Singhd69b0422009-06-17 16:26:34 -07002207/*
Qiang Huangb5ffc852014-06-04 16:08:22 -07002208 * Used to update mapped file or writeback or other statistics.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002209 *
2210 * Notes: Race condition
2211 *
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002212 * Charging occurs during page instantiation, while the page is
2213 * unmapped and locked in page migration, or while the page table is
2214 * locked in THP migration. No race is possible.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002215 *
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002216 * Uncharge happens to pages with zero references, no race possible.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002217 *
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002218 * Charge moving between groups is protected by checking mm->moving
2219 * account and taking the move_lock in the slowpath.
Balbir Singhd69b0422009-06-17 16:26:34 -07002220 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002221
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002222void __mem_cgroup_begin_update_page_stat(struct page *page,
2223 bool *locked, unsigned long *flags)
2224{
2225 struct mem_cgroup *memcg;
2226 struct page_cgroup *pc;
2227
2228 pc = lookup_page_cgroup(page);
2229again:
2230 memcg = pc->mem_cgroup;
2231 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2232 return;
2233 /*
2234 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002235 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002236 * rcu_read_lock(), any calls to move_account will be delayed until
Qiang Huangbdcbb652014-06-04 16:08:21 -07002237 * rcu_read_unlock().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002238 */
Qiang Huangbdcbb652014-06-04 16:08:21 -07002239 VM_BUG_ON(!rcu_read_lock_held());
2240 if (atomic_read(&memcg->moving_account) <= 0)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002241 return;
2242
2243 move_lock_mem_cgroup(memcg, flags);
2244 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2245 move_unlock_mem_cgroup(memcg, flags);
2246 goto again;
2247 }
2248 *locked = true;
2249}
2250
2251void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2252{
2253 struct page_cgroup *pc = lookup_page_cgroup(page);
2254
2255 /*
2256 * It's guaranteed that pc->mem_cgroup never changes while
2257 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002258 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002259 */
2260 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2261}
2262
Greg Thelen2a7106f2011-01-13 15:47:37 -08002263void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002264 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002265{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002266 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002267 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002268 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002269
Johannes Weinercfa44942012-01-12 17:18:38 -08002270 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002271 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002272
Sha Zhengju658b72c2013-09-12 15:13:52 -07002273 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002274 memcg = pc->mem_cgroup;
2275 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002276 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002277
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002278 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002279}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002280
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002281/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002282 * size of first charge trial. "32" comes from vmscan.c's magic value.
2283 * TODO: maybe necessary to use big numbers in big irons.
2284 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002285#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002286struct memcg_stock_pcp {
2287 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002288 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002289 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002290 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002291#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002292};
2293static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002294static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002295
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002296/**
2297 * consume_stock: Try to consume stocked charge on this cpu.
2298 * @memcg: memcg to consume from.
2299 * @nr_pages: how many pages to charge.
2300 *
2301 * The charges will only happen if @memcg matches the current cpu's memcg
2302 * stock, and at least @nr_pages are available in that stock. Failure to
2303 * service an allocation will refill the stock.
2304 *
2305 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002306 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002307static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002308{
2309 struct memcg_stock_pcp *stock;
2310 bool ret = true;
2311
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002312 if (nr_pages > CHARGE_BATCH)
2313 return false;
2314
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002315 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002316 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2317 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002318 else /* need to call res_counter_charge */
2319 ret = false;
2320 put_cpu_var(memcg_stock);
2321 return ret;
2322}
2323
2324/*
2325 * Returns stocks cached in percpu to res_counter and reset cached information.
2326 */
2327static void drain_stock(struct memcg_stock_pcp *stock)
2328{
2329 struct mem_cgroup *old = stock->cached;
2330
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002331 if (stock->nr_pages) {
2332 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2333
2334 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002335 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002336 res_counter_uncharge(&old->memsw, bytes);
2337 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002338 }
2339 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002340}
2341
2342/*
2343 * This must be called under preempt disabled or must be called by
2344 * a thread which is pinned to local cpu.
2345 */
2346static void drain_local_stock(struct work_struct *dummy)
2347{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002348 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002349 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002350 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002351}
2352
Michal Hockoe4777492013-02-22 16:35:40 -08002353static void __init memcg_stock_init(void)
2354{
2355 int cpu;
2356
2357 for_each_possible_cpu(cpu) {
2358 struct memcg_stock_pcp *stock =
2359 &per_cpu(memcg_stock, cpu);
2360 INIT_WORK(&stock->work, drain_local_stock);
2361 }
2362}
2363
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002364/*
2365 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002366 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002367 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002368static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369{
2370 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2371
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002372 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002373 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002374 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002375 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002376 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002377 put_cpu_var(memcg_stock);
2378}
2379
2380/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002381 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002382 * of the hierarchy under it. sync flag says whether we should block
2383 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002384 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002385static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002386{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002387 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002388
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002389 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002390 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002391 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002392 for_each_online_cpu(cpu) {
2393 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002394 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002395
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002396 memcg = stock->cached;
2397 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002398 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002399 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002400 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002401 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2402 if (cpu == curcpu)
2403 drain_local_stock(&stock->work);
2404 else
2405 schedule_work_on(cpu, &stock->work);
2406 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002407 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002408 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002409
2410 if (!sync)
2411 goto out;
2412
2413 for_each_online_cpu(cpu) {
2414 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002415 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002416 flush_work(&stock->work);
2417 }
2418out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002419 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002420}
2421
2422/*
2423 * Tries to drain stocked charges in other cpus. This function is asynchronous
2424 * and just put a work per cpu for draining localy on each cpu. Caller can
2425 * expects some charges will be back to res_counter later but cannot wait for
2426 * it.
2427 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002428static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002429{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002430 /*
2431 * If someone calls draining, avoid adding more kworker runs.
2432 */
2433 if (!mutex_trylock(&percpu_charge_mutex))
2434 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002435 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002436 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002437}
2438
2439/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002440static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002441{
2442 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002443 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002444 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002445 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002446}
2447
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002448/*
2449 * This function drains percpu counter value from DEAD cpu and
2450 * move it to local cpu. Note that this function can be preempted.
2451 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002452static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002453{
2454 int i;
2455
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002456 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002457 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002458 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002459
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002460 per_cpu(memcg->stat->count[i], cpu) = 0;
2461 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002462 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002463 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002464 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002465
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002466 per_cpu(memcg->stat->events[i], cpu) = 0;
2467 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002468 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002469 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002470}
2471
Paul Gortmaker0db06282013-06-19 14:53:51 -04002472static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002473 unsigned long action,
2474 void *hcpu)
2475{
2476 int cpu = (unsigned long)hcpu;
2477 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002478 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002479
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002480 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002481 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002482
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002483 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002484 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002485
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002486 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002487 mem_cgroup_drain_pcp_counter(iter, cpu);
2488
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002489 stock = &per_cpu(memcg_stock, cpu);
2490 drain_stock(stock);
2491 return NOTIFY_OK;
2492}
2493
Johannes Weiner00501b52014-08-08 14:19:20 -07002494static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2495 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002496{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002497 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07002498 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002499 struct mem_cgroup *mem_over_limit;
2500 struct res_counter *fail_res;
2501 unsigned long nr_reclaimed;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002502 unsigned long long size;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002503 bool may_swap = true;
2504 bool drained = false;
Johannes Weiner05b84302014-08-06 16:05:59 -07002505 int ret = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002506
Johannes Weinerce00a962014-09-05 08:43:57 -04002507 if (mem_cgroup_is_root(memcg))
2508 goto done;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002509retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002510 if (consume_stock(memcg, nr_pages))
2511 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002512
Johannes Weiner6539cc02014-08-06 16:05:42 -07002513 size = batch * PAGE_SIZE;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002514 if (!do_swap_account ||
2515 !res_counter_charge(&memcg->memsw, size, &fail_res)) {
2516 if (!res_counter_charge(&memcg->res, size, &fail_res))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002517 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002518 if (do_swap_account)
2519 res_counter_uncharge(&memcg->memsw, size);
2520 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2521 } else {
Johannes Weiner6539cc02014-08-06 16:05:42 -07002522 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002523 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002524 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002525
Johannes Weiner6539cc02014-08-06 16:05:42 -07002526 if (batch > nr_pages) {
2527 batch = nr_pages;
2528 goto retry;
2529 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002530
Johannes Weiner06b078f2014-08-06 16:05:44 -07002531 /*
2532 * Unlike in global OOM situations, memcg is not in a physical
2533 * memory shortage. Allow dying and OOM-killed tasks to
2534 * bypass the last charges so that they can exit quickly and
2535 * free their memory.
2536 */
2537 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2538 fatal_signal_pending(current) ||
2539 current->flags & PF_EXITING))
2540 goto bypass;
2541
2542 if (unlikely(task_in_memcg_oom(current)))
2543 goto nomem;
2544
Johannes Weiner6539cc02014-08-06 16:05:42 -07002545 if (!(gfp_mask & __GFP_WAIT))
2546 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002547
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002548 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2549 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002550
Johannes Weiner61e02c72014-08-06 16:08:16 -07002551 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002552 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002553
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002554 if (!drained) {
2555 drain_all_stock_async(mem_over_limit);
2556 drained = true;
2557 goto retry;
2558 }
2559
Johannes Weiner28c34c22014-08-06 16:05:47 -07002560 if (gfp_mask & __GFP_NORETRY)
2561 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002562 /*
2563 * Even though the limit is exceeded at this point, reclaim
2564 * may have been able to free some pages. Retry the charge
2565 * before killing the task.
2566 *
2567 * Only for regular pages, though: huge pages are rather
2568 * unlikely to succeed so close to the limit, and we fall back
2569 * to regular pages anyway in case of failure.
2570 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002571 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002572 goto retry;
2573 /*
2574 * At task move, charge accounts can be doubly counted. So, it's
2575 * better to wait until the end of task_move if something is going on.
2576 */
2577 if (mem_cgroup_wait_acct_move(mem_over_limit))
2578 goto retry;
2579
Johannes Weiner9b130612014-08-06 16:05:51 -07002580 if (nr_retries--)
2581 goto retry;
2582
Johannes Weiner06b078f2014-08-06 16:05:44 -07002583 if (gfp_mask & __GFP_NOFAIL)
2584 goto bypass;
2585
Johannes Weiner6539cc02014-08-06 16:05:42 -07002586 if (fatal_signal_pending(current))
2587 goto bypass;
2588
Johannes Weiner61e02c72014-08-06 16:08:16 -07002589 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002590nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002591 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002592 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002593bypass:
Johannes Weinerce00a962014-09-05 08:43:57 -04002594 return -EINTR;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002595
2596done_restock:
2597 if (batch > nr_pages)
2598 refill_stock(memcg, batch - nr_pages);
2599done:
Johannes Weiner05b84302014-08-06 16:05:59 -07002600 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002601}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002602
Johannes Weiner00501b52014-08-08 14:19:20 -07002603static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002604{
Johannes Weiner05b84302014-08-06 16:05:59 -07002605 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002606
Johannes Weinerce00a962014-09-05 08:43:57 -04002607 if (mem_cgroup_is_root(memcg))
2608 return;
2609
Johannes Weiner05b84302014-08-06 16:05:59 -07002610 res_counter_uncharge(&memcg->res, bytes);
2611 if (do_swap_account)
2612 res_counter_uncharge(&memcg->memsw, bytes);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002613}
2614
2615/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002616 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2617 * This is useful when moving usage to parent cgroup.
2618 */
2619static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2620 unsigned int nr_pages)
2621{
2622 unsigned long bytes = nr_pages * PAGE_SIZE;
2623
Johannes Weinerce00a962014-09-05 08:43:57 -04002624 if (mem_cgroup_is_root(memcg))
2625 return;
2626
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002627 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2628 if (do_swap_account)
2629 res_counter_uncharge_until(&memcg->memsw,
2630 memcg->memsw.parent, bytes);
2631}
2632
2633/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002634 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoec903c02014-05-13 12:11:01 -04002635 * rcu_read_lock(). The caller is responsible for calling
2636 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
2637 * refcnt from swap can be called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002638 */
2639static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2640{
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002641 /* ID 0 is unused ID */
2642 if (!id)
2643 return NULL;
Li Zefan34c00c32013-09-23 16:56:01 +08002644 return mem_cgroup_from_id(id);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002645}
2646
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002647/*
2648 * try_get_mem_cgroup_from_page - look up page's memcg association
2649 * @page: the page
2650 *
2651 * Look up, get a css reference, and return the memcg that owns @page.
2652 *
2653 * The page must be locked to prevent racing with swap-in and page
2654 * cache charges. If coming from an unlocked page table, the caller
2655 * must ensure the page is on the LRU or this can race with charging.
2656 */
Wu Fengguange42d9d52009-12-16 12:19:59 +01002657struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002658{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002659 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002660 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002661 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002662 swp_entry_t ent;
2663
Sasha Levin309381fea2014-01-23 15:52:54 -08002664 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002665
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002666 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002667 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002668 memcg = pc->mem_cgroup;
Tejun Heoec903c02014-05-13 12:11:01 -04002669 if (memcg && !css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002670 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002671 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002672 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002673 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002674 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002675 memcg = mem_cgroup_lookup(id);
Tejun Heoec903c02014-05-13 12:11:01 -04002676 if (memcg && !css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002677 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002678 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002679 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002680 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002681}
2682
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002683static void lock_page_lru(struct page *page, int *isolated)
2684{
2685 struct zone *zone = page_zone(page);
2686
2687 spin_lock_irq(&zone->lru_lock);
2688 if (PageLRU(page)) {
2689 struct lruvec *lruvec;
2690
2691 lruvec = mem_cgroup_page_lruvec(page, zone);
2692 ClearPageLRU(page);
2693 del_page_from_lru_list(page, lruvec, page_lru(page));
2694 *isolated = 1;
2695 } else
2696 *isolated = 0;
2697}
2698
2699static void unlock_page_lru(struct page *page, int isolated)
2700{
2701 struct zone *zone = page_zone(page);
2702
2703 if (isolated) {
2704 struct lruvec *lruvec;
2705
2706 lruvec = mem_cgroup_page_lruvec(page, zone);
2707 VM_BUG_ON_PAGE(PageLRU(page), page);
2708 SetPageLRU(page);
2709 add_page_to_lru_list(page, lruvec, page_lru(page));
2710 }
2711 spin_unlock_irq(&zone->lru_lock);
2712}
2713
Johannes Weiner00501b52014-08-08 14:19:20 -07002714static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002715 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002716{
Johannes Weinerce587e62012-04-24 20:22:33 +02002717 struct page_cgroup *pc = lookup_page_cgroup(page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002718 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002719
Sasha Levin309381fea2014-01-23 15:52:54 -08002720 VM_BUG_ON_PAGE(PageCgroupUsed(pc), page);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002721 /*
2722 * we don't need page_cgroup_lock about tail pages, becase they are not
2723 * accessed by any other context at this point.
2724 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002725
2726 /*
2727 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2728 * may already be on some other mem_cgroup's LRU. Take care of it.
2729 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002730 if (lrucare)
2731 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002732
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002733 /*
2734 * Nobody should be changing or seriously looking at
2735 * pc->mem_cgroup and pc->flags at this point:
2736 *
2737 * - the page is uncharged
2738 *
2739 * - the page is off-LRU
2740 *
2741 * - an anonymous fault has exclusive page access, except for
2742 * a locked page table
2743 *
2744 * - a page cache insertion, a swapin fault, or a migration
2745 * have the page locked
2746 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002747 pc->mem_cgroup = memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002748 pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0);
Hugh Dickins3be91272008-02-07 00:14:19 -08002749
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002750 if (lrucare)
2751 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002752}
2753
Glauber Costa7cf27982012-12-18 14:22:55 -08002754static DEFINE_MUTEX(set_limit_mutex);
2755
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002756#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovbd673142014-06-04 16:07:40 -07002757/*
2758 * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or
2759 * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists.
2760 */
2761static DEFINE_MUTEX(memcg_slab_mutex);
2762
Vladimir Davydovd6441632014-01-23 15:53:09 -08002763static DEFINE_MUTEX(activate_kmem_mutex);
2764
Glauber Costa1f458cb2012-12-18 14:22:50 -08002765/*
2766 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2767 * in the memcg_cache_params struct.
2768 */
2769static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2770{
2771 struct kmem_cache *cachep;
2772
2773 VM_BUG_ON(p->is_root_cache);
2774 cachep = p->root_cache;
Qiang Huang7a67d7a2013-11-12 15:08:24 -08002775 return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg));
Glauber Costa1f458cb2012-12-18 14:22:50 -08002776}
2777
Glauber Costa749c5412012-12-18 14:23:01 -08002778#ifdef CONFIG_SLABINFO
Tejun Heo2da8ca82013-12-05 12:28:04 -05002779static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v)
Glauber Costa749c5412012-12-18 14:23:01 -08002780{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002781 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Glauber Costa749c5412012-12-18 14:23:01 -08002782 struct memcg_cache_params *params;
2783
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07002784 if (!memcg_kmem_is_active(memcg))
Glauber Costa749c5412012-12-18 14:23:01 -08002785 return -EIO;
2786
2787 print_slabinfo_header(m);
2788
Vladimir Davydovbd673142014-06-04 16:07:40 -07002789 mutex_lock(&memcg_slab_mutex);
Glauber Costa749c5412012-12-18 14:23:01 -08002790 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2791 cache_show(memcg_params_to_cache(params), m);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002792 mutex_unlock(&memcg_slab_mutex);
Glauber Costa749c5412012-12-18 14:23:01 -08002793
2794 return 0;
2795}
2796#endif
2797
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002798static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002799{
2800 struct res_counter *fail_res;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002801 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002802
2803 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2804 if (ret)
2805 return ret;
2806
Johannes Weiner00501b52014-08-08 14:19:20 -07002807 ret = try_charge(memcg, gfp, size >> PAGE_SHIFT);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002808 if (ret == -EINTR) {
2809 /*
Johannes Weiner00501b52014-08-08 14:19:20 -07002810 * try_charge() chose to bypass to root due to OOM kill or
2811 * fatal signal. Since our only options are to either fail
2812 * the allocation or charge it to this cgroup, do it as a
2813 * temporary condition. But we can't fail. From a kmem/slab
2814 * perspective, the cache has already been selected, by
2815 * mem_cgroup_kmem_get_cache(), so it is too late to change
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002816 * our minds.
2817 *
2818 * This condition will only trigger if the task entered
Johannes Weiner00501b52014-08-08 14:19:20 -07002819 * memcg_charge_kmem in a sane state, but was OOM-killed
2820 * during try_charge() above. Tasks that were already dying
2821 * when the allocation triggers should have been already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002822 * directed to the root cgroup in memcontrol.h
2823 */
2824 res_counter_charge_nofail(&memcg->res, size, &fail_res);
2825 if (do_swap_account)
2826 res_counter_charge_nofail(&memcg->memsw, size,
2827 &fail_res);
2828 ret = 0;
2829 } else if (ret)
2830 res_counter_uncharge(&memcg->kmem, size);
2831
2832 return ret;
2833}
2834
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002835static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002836{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002837 res_counter_uncharge(&memcg->res, size);
2838 if (do_swap_account)
2839 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08002840
2841 /* Not down to 0 */
2842 if (res_counter_uncharge(&memcg->kmem, size))
2843 return;
2844
Li Zefan10d5ebf2013-07-08 16:00:33 -07002845 /*
2846 * Releases a reference taken in kmem_cgroup_css_offline in case
2847 * this last uncharge is racing with the offlining code or it is
2848 * outliving the memcg existence.
2849 *
2850 * The memory barrier imposed by test&clear is paired with the
2851 * explicit one in memcg_kmem_mark_dead().
2852 */
Glauber Costa7de37682012-12-18 14:22:07 -08002853 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07002854 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002855}
2856
Glauber Costa2633d7a2012-12-18 14:22:34 -08002857/*
2858 * helper for acessing a memcg's index. It will be used as an index in the
2859 * child cache array in kmem_cache, and also to derive its name. This function
2860 * will return -1 when this is not a kmem-limited memcg.
2861 */
2862int memcg_cache_id(struct mem_cgroup *memcg)
2863{
2864 return memcg ? memcg->kmemcg_id : -1;
2865}
2866
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002867static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002868{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002869 int id, size;
2870 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002871
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002872 id = ida_simple_get(&kmem_limited_groups,
2873 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2874 if (id < 0)
2875 return id;
2876
2877 if (id < memcg_limited_groups_array_size)
2878 return id;
2879
2880 /*
2881 * There's no space for the new id in memcg_caches arrays,
2882 * so we have to grow them.
2883 */
2884
2885 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002886 if (size < MEMCG_CACHES_MIN_SIZE)
2887 size = MEMCG_CACHES_MIN_SIZE;
2888 else if (size > MEMCG_CACHES_MAX_SIZE)
2889 size = MEMCG_CACHES_MAX_SIZE;
2890
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002891 mutex_lock(&memcg_slab_mutex);
2892 err = memcg_update_all_caches(size);
2893 mutex_unlock(&memcg_slab_mutex);
2894
2895 if (err) {
2896 ida_simple_remove(&kmem_limited_groups, id);
2897 return err;
2898 }
2899 return id;
2900}
2901
2902static void memcg_free_cache_id(int id)
2903{
2904 ida_simple_remove(&kmem_limited_groups, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002905}
2906
2907/*
2908 * We should update the current array size iff all caches updates succeed. This
2909 * can only be done from the slab side. The slab mutex needs to be held when
2910 * calling this.
2911 */
2912void memcg_update_array_size(int num)
2913{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002914 memcg_limited_groups_array_size = num;
Glauber Costa55007d82012-12-18 14:22:38 -08002915}
2916
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002917static void memcg_register_cache(struct mem_cgroup *memcg,
2918 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08002919{
Vladimir Davydov93f39ee2014-06-04 16:08:24 -07002920 static char memcg_name_buf[NAME_MAX + 1]; /* protected by
2921 memcg_slab_mutex */
Vladimir Davydovbd673142014-06-04 16:07:40 -07002922 struct kmem_cache *cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002923 int id;
2924
Vladimir Davydovbd673142014-06-04 16:07:40 -07002925 lockdep_assert_held(&memcg_slab_mutex);
2926
2927 id = memcg_cache_id(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08002928
Vladimir Davydov2edefe12014-01-23 15:53:02 -08002929 /*
Vladimir Davydovbd673142014-06-04 16:07:40 -07002930 * Since per-memcg caches are created asynchronously on first
2931 * allocation (see memcg_kmem_get_cache()), several threads can try to
2932 * create the same cache, but only one of them may succeed.
Vladimir Davydov2edefe12014-01-23 15:53:02 -08002933 */
Vladimir Davydovbd673142014-06-04 16:07:40 -07002934 if (cache_from_memcg_idx(root_cache, id))
2935 return;
Vladimir Davydov2edefe12014-01-23 15:53:02 -08002936
Vladimir Davydov073ee1c2014-06-04 16:08:23 -07002937 cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1);
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002938 cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002939 /*
2940 * If we could not create a memcg cache, do not complain, because
2941 * that's not critical at all as we can always proceed with the root
2942 * cache.
2943 */
2944 if (!cachep)
2945 return;
2946
Vladimir Davydov33a690c2014-10-09 15:28:43 -07002947 css_get(&memcg->css);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002948 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002949
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002950 /*
Vladimir Davydov959c8962014-01-23 15:52:59 -08002951 * Since readers won't lock (see cache_from_memcg_idx()), we need a
2952 * barrier here to ensure nobody will see the kmem_cache partially
2953 * initialized.
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002954 */
Vladimir Davydov959c8962014-01-23 15:52:59 -08002955 smp_wmb();
2956
Vladimir Davydovbd673142014-06-04 16:07:40 -07002957 BUG_ON(root_cache->memcg_params->memcg_caches[id]);
2958 root_cache->memcg_params->memcg_caches[id] = cachep;
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002959}
2960
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002961static void memcg_unregister_cache(struct kmem_cache *cachep)
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002962{
Vladimir Davydovbd673142014-06-04 16:07:40 -07002963 struct kmem_cache *root_cache;
Vladimir Davydov1aa13252014-01-23 15:52:58 -08002964 struct mem_cgroup *memcg;
2965 int id;
2966
Vladimir Davydovbd673142014-06-04 16:07:40 -07002967 lockdep_assert_held(&memcg_slab_mutex);
Glauber Costad7f25f82012-12-18 14:22:40 -08002968
Vladimir Davydovbd673142014-06-04 16:07:40 -07002969 BUG_ON(is_root_cache(cachep));
Vladimir Davydov2edefe12014-01-23 15:53:02 -08002970
Vladimir Davydovbd673142014-06-04 16:07:40 -07002971 root_cache = cachep->memcg_params->root_cache;
2972 memcg = cachep->memcg_params->memcg;
Vladimir Davydov96403da2014-01-23 15:53:01 -08002973 id = memcg_cache_id(memcg);
Glauber Costad7f25f82012-12-18 14:22:40 -08002974
Vladimir Davydovbd673142014-06-04 16:07:40 -07002975 BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep);
2976 root_cache->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08002977
Vladimir Davydovbd673142014-06-04 16:07:40 -07002978 list_del(&cachep->memcg_params->list);
2979
2980 kmem_cache_destroy(cachep);
Vladimir Davydov33a690c2014-10-09 15:28:43 -07002981
2982 /* drop the reference taken in memcg_register_cache */
2983 css_put(&memcg->css);
Glauber Costa2633d7a2012-12-18 14:22:34 -08002984}
2985
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002986/*
2987 * During the creation a new cache, we need to disable our accounting mechanism
2988 * altogether. This is true even if we are not creating, but rather just
2989 * enqueing new caches to be created.
2990 *
2991 * This is because that process will trigger allocations; some visible, like
2992 * explicit kmallocs to auxiliary data structures, name strings and internal
2993 * cache structures; some well concealed, like INIT_WORK() that can allocate
2994 * objects during debug.
2995 *
2996 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
2997 * to it. This may not be a bounded recursion: since the first cache creation
2998 * failed to complete (waiting on the allocation), we'll just try to create the
2999 * cache again, failing at the same point.
3000 *
3001 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3002 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3003 * inside the following two functions.
3004 */
3005static inline void memcg_stop_kmem_account(void)
3006{
3007 VM_BUG_ON(!current->mm);
3008 current->memcg_kmem_skip_account++;
3009}
3010
3011static inline void memcg_resume_kmem_account(void)
3012{
3013 VM_BUG_ON(!current->mm);
3014 current->memcg_kmem_skip_account--;
3015}
3016
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003017int __memcg_cleanup_cache_params(struct kmem_cache *s)
Glauber Costa7cf27982012-12-18 14:22:55 -08003018{
3019 struct kmem_cache *c;
Vladimir Davydovb8529902014-04-07 15:39:28 -07003020 int i, failed = 0;
Glauber Costa7cf27982012-12-18 14:22:55 -08003021
Vladimir Davydovbd673142014-06-04 16:07:40 -07003022 mutex_lock(&memcg_slab_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003023 for_each_memcg_cache_index(i) {
3024 c = cache_from_memcg_idx(s, i);
Glauber Costa7cf27982012-12-18 14:22:55 -08003025 if (!c)
3026 continue;
3027
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003028 memcg_unregister_cache(c);
Vladimir Davydovb8529902014-04-07 15:39:28 -07003029
3030 if (cache_from_memcg_idx(s, i))
3031 failed++;
Glauber Costa7cf27982012-12-18 14:22:55 -08003032 }
Vladimir Davydovbd673142014-06-04 16:07:40 -07003033 mutex_unlock(&memcg_slab_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -07003034 return failed;
Glauber Costa7cf27982012-12-18 14:22:55 -08003035}
3036
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003037static void memcg_unregister_all_caches(struct mem_cgroup *memcg)
Glauber Costa1f458cb2012-12-18 14:22:50 -08003038{
3039 struct kmem_cache *cachep;
Vladimir Davydovbd673142014-06-04 16:07:40 -07003040 struct memcg_cache_params *params, *tmp;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003041
3042 if (!memcg_kmem_is_active(memcg))
3043 return;
3044
Vladimir Davydovbd673142014-06-04 16:07:40 -07003045 mutex_lock(&memcg_slab_mutex);
3046 list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) {
Glauber Costa1f458cb2012-12-18 14:22:50 -08003047 cachep = memcg_params_to_cache(params);
Vladimir Davydovbd673142014-06-04 16:07:40 -07003048 kmem_cache_shrink(cachep);
3049 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003050 memcg_unregister_cache(cachep);
Glauber Costa1f458cb2012-12-18 14:22:50 -08003051 }
Vladimir Davydovbd673142014-06-04 16:07:40 -07003052 mutex_unlock(&memcg_slab_mutex);
Glauber Costa1f458cb2012-12-18 14:22:50 -08003053}
3054
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003055struct memcg_register_cache_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07003056 struct mem_cgroup *memcg;
3057 struct kmem_cache *cachep;
3058 struct work_struct work;
3059};
3060
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003061static void memcg_register_cache_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08003062{
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003063 struct memcg_register_cache_work *cw =
3064 container_of(w, struct memcg_register_cache_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07003065 struct mem_cgroup *memcg = cw->memcg;
3066 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003067
Vladimir Davydovbd673142014-06-04 16:07:40 -07003068 mutex_lock(&memcg_slab_mutex);
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003069 memcg_register_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07003070 mutex_unlock(&memcg_slab_mutex);
3071
Vladimir Davydov5722d092014-04-07 15:39:24 -07003072 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003073 kfree(cw);
3074}
3075
3076/*
3077 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003078 */
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003079static void __memcg_schedule_register_cache(struct mem_cgroup *memcg,
3080 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003081{
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003082 struct memcg_register_cache_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08003083
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003084 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003085 if (cw == NULL) {
3086 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003087 return;
3088 }
3089
3090 cw->memcg = memcg;
3091 cw->cachep = cachep;
3092
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003093 INIT_WORK(&cw->work, memcg_register_cache_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08003094 schedule_work(&cw->work);
3095}
3096
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003097static void memcg_schedule_register_cache(struct mem_cgroup *memcg,
3098 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003099{
3100 /*
3101 * We need to stop accounting when we kmalloc, because if the
3102 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003103 * in __memcg_schedule_register_cache will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003104 *
3105 * However, it is better to enclose the whole function. Depending on
3106 * the debugging options enabled, INIT_WORK(), for instance, can
3107 * trigger an allocation. This too, will make us recurse. Because at
3108 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3109 * the safest choice is to do it like this, wrapping the whole function.
3110 */
3111 memcg_stop_kmem_account();
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003112 __memcg_schedule_register_cache(memcg, cachep);
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003113 memcg_resume_kmem_account();
3114}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07003115
3116int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order)
3117{
3118 int res;
3119
3120 res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp,
3121 PAGE_SIZE << order);
3122 if (!res)
3123 atomic_add(1 << order, &cachep->memcg_params->nr_pages);
3124 return res;
3125}
3126
3127void __memcg_uncharge_slab(struct kmem_cache *cachep, int order)
3128{
3129 memcg_uncharge_kmem(cachep->memcg_params->memcg, PAGE_SIZE << order);
3130 atomic_sub(1 << order, &cachep->memcg_params->nr_pages);
3131}
3132
Glauber Costad7f25f82012-12-18 14:22:40 -08003133/*
3134 * Return the kmem_cache we're supposed to use for a slab allocation.
3135 * We try to use the current memcg's version of the cache.
3136 *
3137 * If the cache does not exist yet, if we are the first user of it,
3138 * we either create it immediately, if possible, or create it asynchronously
3139 * in a workqueue.
3140 * In the latter case, we will let the current allocation go through with
3141 * the original cache.
3142 *
3143 * Can't be called in interrupt context or from kernel threads.
3144 * This function needs to be called with rcu_read_lock() held.
3145 */
3146struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3147 gfp_t gfp)
3148{
3149 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08003150 struct kmem_cache *memcg_cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003151
3152 VM_BUG_ON(!cachep->memcg_params);
3153 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3154
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003155 if (!current->mm || current->memcg_kmem_skip_account)
3156 return cachep;
3157
Glauber Costad7f25f82012-12-18 14:22:40 -08003158 rcu_read_lock();
3159 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003160
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07003161 if (!memcg_kmem_is_active(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003162 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003163
Vladimir Davydov959c8962014-01-23 15:52:59 -08003164 memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg));
3165 if (likely(memcg_cachep)) {
3166 cachep = memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07003167 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003168 }
3169
Li Zefanca0dde92013-04-29 15:08:57 -07003170 /* The corresponding put will be done in the workqueue. */
Tejun Heoec903c02014-05-13 12:11:01 -04003171 if (!css_tryget_online(&memcg->css))
Li Zefanca0dde92013-04-29 15:08:57 -07003172 goto out;
3173 rcu_read_unlock();
3174
3175 /*
3176 * If we are in a safe context (can wait, and not in interrupt
3177 * context), we could be be predictable and return right away.
3178 * This would guarantee that the allocation being performed
3179 * already belongs in the new cache.
3180 *
3181 * However, there are some clashes that can arrive from locking.
3182 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003183 * memcg_create_kmem_cache, this means no further allocation
3184 * could happen with the slab_mutex held. So it's better to
3185 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07003186 */
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003187 memcg_schedule_register_cache(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07003188 return cachep;
3189out:
3190 rcu_read_unlock();
3191 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003192}
Glauber Costad7f25f82012-12-18 14:22:40 -08003193
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003194/*
3195 * We need to verify if the allocation against current->mm->owner's memcg is
3196 * possible for the given order. But the page is not allocated yet, so we'll
3197 * need a further commit step to do the final arrangements.
3198 *
3199 * It is possible for the task to switch cgroups in this mean time, so at
3200 * commit time, we can't rely on task conversion any longer. We'll then use
3201 * the handle argument to return to the caller which cgroup we should commit
3202 * against. We could also return the memcg directly and avoid the pointer
3203 * passing, but a boolean return value gives better semantics considering
3204 * the compiled-out case as well.
3205 *
3206 * Returning true means the allocation is possible.
3207 */
3208bool
3209__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3210{
3211 struct mem_cgroup *memcg;
3212 int ret;
3213
3214 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003215
3216 /*
3217 * Disabling accounting is only relevant for some specific memcg
3218 * internal allocations. Therefore we would initially not have such
Vladimir Davydov52383432014-06-04 16:06:39 -07003219 * check here, since direct calls to the page allocator that are
3220 * accounted to kmemcg (alloc_kmem_pages and friends) only happen
3221 * outside memcg core. We are mostly concerned with cache allocations,
3222 * and by having this test at memcg_kmem_get_cache, we are already able
3223 * to relay the allocation to the root cache and bypass the memcg cache
3224 * altogether.
Glauber Costa6d42c232013-07-08 16:00:00 -07003225 *
3226 * There is one exception, though: the SLUB allocator does not create
3227 * large order caches, but rather service large kmallocs directly from
3228 * the page allocator. Therefore, the following sequence when backed by
3229 * the SLUB allocator:
3230 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003231 * memcg_stop_kmem_account();
3232 * kmalloc(<large_number>)
3233 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003234 *
3235 * would effectively ignore the fact that we should skip accounting,
3236 * since it will drive us directly to this function without passing
3237 * through the cache selector memcg_kmem_get_cache. Such large
3238 * allocations are extremely rare but can happen, for instance, for the
3239 * cache arrays. We bring this test here.
3240 */
3241 if (!current->mm || current->memcg_kmem_skip_account)
3242 return true;
3243
Johannes Weinerdf381972014-04-07 15:37:43 -07003244 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003245
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07003246 if (!memcg_kmem_is_active(memcg)) {
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003247 css_put(&memcg->css);
3248 return true;
3249 }
3250
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003251 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3252 if (!ret)
3253 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003254
3255 css_put(&memcg->css);
3256 return (ret == 0);
3257}
3258
3259void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3260 int order)
3261{
3262 struct page_cgroup *pc;
3263
3264 VM_BUG_ON(mem_cgroup_is_root(memcg));
3265
3266 /* The page allocation failed. Revert */
3267 if (!page) {
3268 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003269 return;
3270 }
Johannes Weinera840cda2014-08-06 16:06:04 -07003271 /*
3272 * The page is freshly allocated and not visible to any
3273 * outside callers yet. Set up pc non-atomically.
3274 */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003275 pc = lookup_page_cgroup(page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003276 pc->mem_cgroup = memcg;
Johannes Weinera840cda2014-08-06 16:06:04 -07003277 pc->flags = PCG_USED;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003278}
3279
3280void __memcg_kmem_uncharge_pages(struct page *page, int order)
3281{
3282 struct mem_cgroup *memcg = NULL;
3283 struct page_cgroup *pc;
3284
3285
3286 pc = lookup_page_cgroup(page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003287 if (!PageCgroupUsed(pc))
3288 return;
3289
Johannes Weinera840cda2014-08-06 16:06:04 -07003290 memcg = pc->mem_cgroup;
3291 pc->flags = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003292
3293 /*
3294 * We trust that only if there is a memcg associated with the page, it
3295 * is a valid allocation
3296 */
3297 if (!memcg)
3298 return;
3299
Sasha Levin309381fea2014-01-23 15:52:54 -08003300 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003301 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003302}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003303#else
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07003304static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg)
Glauber Costa1f458cb2012-12-18 14:22:50 -08003305{
3306}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003307#endif /* CONFIG_MEMCG_KMEM */
3308
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003309#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3310
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003311/*
3312 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003313 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3314 * charge/uncharge will be never happen and move_account() is done under
3315 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003316 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003317void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003318{
3319 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003320 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003321 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003322 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003323
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003324 if (mem_cgroup_disabled())
3325 return;
David Rientjesb070e652013-05-07 16:18:09 -07003326
3327 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003328 for (i = 1; i < HPAGE_PMD_NR; i++) {
3329 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003330 pc->mem_cgroup = memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003331 pc->flags = head_pc->flags;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003332 }
David Rientjesb070e652013-05-07 16:18:09 -07003333 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3334 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003335}
Hugh Dickins12d27102012-01-12 17:19:52 -08003336#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003337
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003338/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003339 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003340 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003341 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003342 * @pc: page_cgroup of the page.
3343 * @from: mem_cgroup which the page is moved from.
3344 * @to: mem_cgroup which the page is moved to. @from != @to.
3345 *
3346 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003347 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003348 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003349 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003350 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3351 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003352 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003353static int mem_cgroup_move_account(struct page *page,
3354 unsigned int nr_pages,
3355 struct page_cgroup *pc,
3356 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003357 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003358{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003359 unsigned long flags;
3360 int ret;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003361
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003362 VM_BUG_ON(from == to);
Sasha Levin309381fea2014-01-23 15:52:54 -08003363 VM_BUG_ON_PAGE(PageLRU(page), page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003364 /*
3365 * The page is isolated from LRU. So, collapse function
3366 * will not handle this page. But page splitting can happen.
3367 * Do this check under compound_page_lock(). The caller should
3368 * hold it.
3369 */
3370 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003371 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003372 goto out;
3373
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003374 /*
3375 * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup
3376 * of its source page while we change it: page migration takes
3377 * both pages off the LRU, but page cache replacement doesn't.
3378 */
3379 if (!trylock_page(page))
3380 goto out;
Johannes Weinerde3638d2011-03-23 16:42:28 -07003381
3382 ret = -EINVAL;
3383 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003384 goto out_unlock;
Johannes Weinerde3638d2011-03-23 16:42:28 -07003385
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003386 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003387
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003388 if (!PageAnon(page) && page_mapped(page)) {
Johannes Weiner59d1d252014-04-07 15:37:40 -07003389 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3390 nr_pages);
3391 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
3392 nr_pages);
3393 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003394
Johannes Weiner59d1d252014-04-07 15:37:40 -07003395 if (PageWriteback(page)) {
3396 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3397 nr_pages);
3398 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
3399 nr_pages);
3400 }
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003401
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003402 /*
3403 * It is safe to change pc->mem_cgroup here because the page
3404 * is referenced, charged, and isolated - we can't race with
3405 * uncharging, charging, migration, or LRU putback.
3406 */
Balbir Singhd69b0422009-06-17 16:26:34 -07003407
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003408 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003409 pc->mem_cgroup = to;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003410 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003411 ret = 0;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003412
3413 local_irq_disable();
3414 mem_cgroup_charge_statistics(to, page, nr_pages);
Johannes Weiner5564e882011-03-23 16:42:29 -07003415 memcg_check_events(to, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003416 mem_cgroup_charge_statistics(from, page, -nr_pages);
Johannes Weiner5564e882011-03-23 16:42:29 -07003417 memcg_check_events(from, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003418 local_irq_enable();
3419out_unlock:
3420 unlock_page(page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003421out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003422 return ret;
3423}
3424
Michal Hocko2ef37d32012-10-26 13:37:30 +02003425/**
3426 * mem_cgroup_move_parent - moves page to the parent group
3427 * @page: the page to move
3428 * @pc: page_cgroup of the page
3429 * @child: page's cgroup
3430 *
3431 * move charges to its parent or the root cgroup if the group has no
3432 * parent (aka use_hierarchy==0).
3433 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3434 * mem_cgroup_move_account fails) the failure is always temporary and
3435 * it signals a race with a page removal/uncharge or migration. In the
3436 * first case the page is on the way out and it will vanish from the LRU
3437 * on the next attempt and the call should be retried later.
3438 * Isolation from the LRU fails only if page has been isolated from
3439 * the LRU since we looked at it and that usually means either global
3440 * reclaim or migration going on. The page will either get back to the
3441 * LRU or vanish.
3442 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3443 * (!PageCgroupUsed) or moved to a different group. The page will
3444 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003445 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003446static int mem_cgroup_move_parent(struct page *page,
3447 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003448 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003449{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003450 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003451 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003452 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003453 int ret;
3454
Michal Hockod8423012012-10-26 13:37:29 +02003455 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003456
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003457 ret = -EBUSY;
3458 if (!get_page_unless_zero(page))
3459 goto out;
3460 if (isolate_lru_page(page))
3461 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003462
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003463 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003464
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003465 parent = parent_mem_cgroup(child);
3466 /*
3467 * If no parent, move charges to root cgroup.
3468 */
3469 if (!parent)
3470 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003471
Michal Hocko2ef37d32012-10-26 13:37:30 +02003472 if (nr_pages > 1) {
Sasha Levin309381fea2014-01-23 15:52:54 -08003473 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003474 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003475 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003476
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003477 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003478 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003479 if (!ret)
3480 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003481
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003482 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003483 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003484 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003485put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003486 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003487out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003488 return ret;
3489}
3490
Andrew Mortonc255a452012-07-31 16:43:02 -07003491#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003492static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
3493 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003494{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003495 int val = (charge) ? 1 : -1;
3496 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003497}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003498
3499/**
3500 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3501 * @entry: swap entry to be moved
3502 * @from: mem_cgroup which the entry is moved from
3503 * @to: mem_cgroup which the entry is moved to
3504 *
3505 * It succeeds only when the swap_cgroup's record for this entry is the same
3506 * as the mem_cgroup's id of @from.
3507 *
3508 * Returns 0 on success, -EINVAL on failure.
3509 *
3510 * The caller must have charged to @to, IOW, called res_counter_charge() about
3511 * both res and memsw, and called css_get().
3512 */
3513static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003514 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003515{
3516 unsigned short old_id, new_id;
3517
Li Zefan34c00c32013-09-23 16:56:01 +08003518 old_id = mem_cgroup_id(from);
3519 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003520
3521 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003522 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003523 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003524 /*
3525 * This function is only called from task migration context now.
3526 * It postpones res_counter and refcount handling till the end
3527 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07003528 * improvement. But we cannot postpone css_get(to) because if
3529 * the process that has been moved to @to does swap-in, the
3530 * refcount of @to might be decreased to 0.
3531 *
3532 * We are in attach() phase, so the cgroup is guaranteed to be
3533 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003534 */
Li Zefan40503772013-07-08 16:00:34 -07003535 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003536 return 0;
3537 }
3538 return -EINVAL;
3539}
3540#else
3541static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003542 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003543{
3544 return -EINVAL;
3545}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003546#endif
3547
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003548#ifdef CONFIG_DEBUG_VM
3549static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3550{
3551 struct page_cgroup *pc;
3552
3553 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003554 /*
3555 * Can be NULL while feeding pages into the page allocator for
3556 * the first time, i.e. during boot or memory hotplug;
3557 * or when mem_cgroup_disabled().
3558 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003559 if (likely(pc) && PageCgroupUsed(pc))
3560 return pc;
3561 return NULL;
3562}
3563
3564bool mem_cgroup_bad_page_check(struct page *page)
3565{
3566 if (mem_cgroup_disabled())
3567 return false;
3568
3569 return lookup_page_cgroup_used(page) != NULL;
3570}
3571
3572void mem_cgroup_print_bad_page(struct page *page)
3573{
3574 struct page_cgroup *pc;
3575
3576 pc = lookup_page_cgroup_used(page);
3577 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08003578 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3579 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003580 }
3581}
3582#endif
3583
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003584static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003585 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003586{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003587 int retry_count;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003588 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003589 int children = mem_cgroup_count_children(memcg);
3590 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003591 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003592
3593 /*
3594 * For keeping hierarchical_reclaim simple, how long we should retry
3595 * is depends on callers. We set our retry-count to be function
3596 * of # of children which we should visit in this loop.
3597 */
3598 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3599
3600 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003601
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003602 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003603 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003604 if (signal_pending(current)) {
3605 ret = -EINTR;
3606 break;
3607 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003608 /*
3609 * Rather than hide all in some function, I do this in
3610 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07003611 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003612 */
3613 mutex_lock(&set_limit_mutex);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07003614 if (res_counter_read_u64(&memcg->memsw, RES_LIMIT) < val) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003615 ret = -EINVAL;
3616 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003617 break;
3618 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003619
Johannes Weiner3fbe7242014-10-09 15:28:54 -07003620 if (res_counter_read_u64(&memcg->res, RES_LIMIT) < val)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003621 enlarge = 1;
3622
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003623 ret = res_counter_set_limit(&memcg->res, val);
3624 mutex_unlock(&set_limit_mutex);
3625
3626 if (!ret)
3627 break;
3628
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003629 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
3630
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003631 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3632 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003633 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003634 retry_count--;
3635 else
3636 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003637 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003638 if (!ret && enlarge)
3639 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003640
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003641 return ret;
3642}
3643
Li Zefan338c8432009-06-17 16:27:15 -07003644static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3645 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003646{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003647 int retry_count;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07003648 u64 oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003649 int children = mem_cgroup_count_children(memcg);
3650 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003651 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003652
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003653 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003654 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003655 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003656 while (retry_count) {
3657 if (signal_pending(current)) {
3658 ret = -EINTR;
3659 break;
3660 }
3661 /*
3662 * Rather than hide all in some function, I do this in
3663 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07003664 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003665 */
3666 mutex_lock(&set_limit_mutex);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07003667 if (res_counter_read_u64(&memcg->res, RES_LIMIT) > val) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003668 ret = -EINVAL;
3669 mutex_unlock(&set_limit_mutex);
3670 break;
3671 }
Johannes Weiner3fbe7242014-10-09 15:28:54 -07003672 if (res_counter_read_u64(&memcg->memsw, RES_LIMIT) < val)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003673 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003674 ret = res_counter_set_limit(&memcg->memsw, val);
3675 mutex_unlock(&set_limit_mutex);
3676
3677 if (!ret)
3678 break;
3679
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003680 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
3681
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003682 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003683 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003684 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003685 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003686 else
3687 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003688 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003689 if (!ret && enlarge)
3690 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003691 return ret;
3692}
3693
Andrew Morton0608f432013-09-24 15:27:41 -07003694unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3695 gfp_t gfp_mask,
3696 unsigned long *total_scanned)
3697{
3698 unsigned long nr_reclaimed = 0;
3699 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3700 unsigned long reclaimed;
3701 int loop = 0;
3702 struct mem_cgroup_tree_per_zone *mctz;
3703 unsigned long long excess;
3704 unsigned long nr_scanned;
3705
3706 if (order > 0)
3707 return 0;
3708
3709 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3710 /*
3711 * This loop can run a while, specially if mem_cgroup's continuously
3712 * keep exceeding their soft limit and putting the system under
3713 * pressure
3714 */
3715 do {
3716 if (next_mz)
3717 mz = next_mz;
3718 else
3719 mz = mem_cgroup_largest_soft_limit_node(mctz);
3720 if (!mz)
3721 break;
3722
3723 nr_scanned = 0;
3724 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
3725 gfp_mask, &nr_scanned);
3726 nr_reclaimed += reclaimed;
3727 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003728 spin_lock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003729
3730 /*
3731 * If we failed to reclaim anything from this memory cgroup
3732 * it is time to move on to the next cgroup
3733 */
3734 next_mz = NULL;
3735 if (!reclaimed) {
3736 do {
3737 /*
3738 * Loop until we find yet another one.
3739 *
3740 * By the time we get the soft_limit lock
3741 * again, someone might have aded the
3742 * group back on the RB tree. Iterate to
3743 * make sure we get a different mem.
3744 * mem_cgroup_largest_soft_limit_node returns
3745 * NULL if no other cgroup is present on
3746 * the tree
3747 */
3748 next_mz =
3749 __mem_cgroup_largest_soft_limit_node(mctz);
3750 if (next_mz == mz)
3751 css_put(&next_mz->memcg->css);
3752 else /* next_mz == NULL or other memcg */
3753 break;
3754 } while (1);
3755 }
Johannes Weinercf2c8122014-06-06 14:38:21 -07003756 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07003757 excess = res_counter_soft_limit_excess(&mz->memcg->res);
3758 /*
3759 * One school of thought says that we should not add
3760 * back the node to the tree if reclaim returns 0.
3761 * But our reclaim could return 0, simply because due
3762 * to priority we are exposing a smaller subset of
3763 * memory to reclaim from. Consider this as a longer
3764 * term TODO.
3765 */
3766 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07003767 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003768 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003769 css_put(&mz->memcg->css);
3770 loop++;
3771 /*
3772 * Could not reclaim anything and there are no more
3773 * mem cgroups to try or we seem to be looping without
3774 * reclaiming anything.
3775 */
3776 if (!nr_reclaimed &&
3777 (next_mz == NULL ||
3778 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3779 break;
3780 } while (!nr_reclaimed);
3781 if (next_mz)
3782 css_put(&next_mz->memcg->css);
3783 return nr_reclaimed;
3784}
3785
Michal Hocko2ef37d32012-10-26 13:37:30 +02003786/**
3787 * mem_cgroup_force_empty_list - clears LRU of a group
3788 * @memcg: group to clear
3789 * @node: NUMA node
3790 * @zid: zone id
3791 * @lru: lru to to clear
3792 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003793 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02003794 * reclaim the pages page themselves - pages are moved to the parent (or root)
3795 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003796 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02003797static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003798 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003799{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08003800 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02003801 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003802 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003803 struct page *busy;
3804 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003805
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003806 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08003807 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
3808 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003809
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003810 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02003811 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08003812 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003813 struct page *page;
3814
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003815 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003816 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003817 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003818 break;
3819 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003820 page = list_entry(list->prev, struct page, lru);
3821 if (busy == page) {
3822 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003823 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003824 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003825 continue;
3826 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003827 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003828
Johannes Weiner925b7672012-01-12 17:18:15 -08003829 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003830
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07003831 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003832 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003833 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003834 } else
3835 busy = NULL;
Hugh Dickins2a7a0e02014-06-04 16:11:04 -07003836 cond_resched();
Michal Hocko2ef37d32012-10-26 13:37:30 +02003837 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003838}
3839
3840/*
Michal Hockoc26251f2012-10-26 13:37:28 +02003841 * make mem_cgroup's charge to be 0 if there is no task by moving
3842 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003843 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02003844 *
3845 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003846 */
Michal Hockoab5196c2012-10-26 13:37:32 +02003847static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003848{
Michal Hockoc26251f2012-10-26 13:37:28 +02003849 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08003850 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003851
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003852 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003853 /* This is for making all *used* pages to be on LRU. */
3854 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003855 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003856 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08003857 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02003858 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003859 enum lru_list lru;
3860 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02003861 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07003862 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003863 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003864 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003865 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003866 mem_cgroup_end_move(memcg);
3867 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003868 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003869
Michal Hocko2ef37d32012-10-26 13:37:30 +02003870 /*
Glauber Costabea207c2012-12-18 14:22:11 -08003871 * Kernel memory may not necessarily be trackable to a specific
3872 * process. So they are not migrated, and therefore we can't
3873 * expect their value to drop to 0 here.
3874 * Having res filled up with kmem only is enough.
3875 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02003876 * This is a safety check because mem_cgroup_force_empty_list
3877 * could have raced with mem_cgroup_replace_page_cache callers
3878 * so the lru seemed empty but the page could have been added
3879 * right after the check. RES_USAGE should be safe as we always
3880 * charge before adding to the LRU.
3881 */
Glauber Costabea207c2012-12-18 14:22:11 -08003882 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
3883 res_counter_read_u64(&memcg->kmem, RES_USAGE);
3884 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02003885}
3886
Tejun Heoea280e72014-05-16 13:22:48 -04003887/*
3888 * Test whether @memcg has children, dead or alive. Note that this
3889 * function doesn't care whether @memcg has use_hierarchy enabled and
3890 * returns %true if there are child csses according to the cgroup
3891 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
3892 */
Glauber Costab5f99b52013-02-22 16:34:53 -08003893static inline bool memcg_has_children(struct mem_cgroup *memcg)
3894{
Tejun Heoea280e72014-05-16 13:22:48 -04003895 bool ret;
3896
Johannes Weiner696ac172013-10-31 16:34:15 -07003897 /*
Tejun Heoea280e72014-05-16 13:22:48 -04003898 * The lock does not prevent addition or deletion of children, but
3899 * it prevents a new child from being initialized based on this
3900 * parent in css_online(), so it's enough to decide whether
3901 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07003902 */
Tejun Heoea280e72014-05-16 13:22:48 -04003903 lockdep_assert_held(&memcg_create_mutex);
3904
3905 rcu_read_lock();
3906 ret = css_next_child(NULL, &memcg->css);
3907 rcu_read_unlock();
3908 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08003909}
3910
3911/*
Michal Hockoc26251f2012-10-26 13:37:28 +02003912 * Reclaims as many pages from the given memcg as possible and moves
3913 * the rest to the parent.
3914 *
3915 * Caller is responsible for holding css reference for memcg.
3916 */
3917static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3918{
3919 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003920
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003921 /* we call try-to-free pages for make this cgroup empty */
3922 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003923 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07003924 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003925 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003926
Michal Hockoc26251f2012-10-26 13:37:28 +02003927 if (signal_pending(current))
3928 return -EINTR;
3929
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003930 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3931 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003932 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003933 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003934 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003935 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003936 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003937
3938 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003939
3940 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003941}
3942
Tejun Heo6770c642014-05-13 12:16:21 -04003943static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3944 char *buf, size_t nbytes,
3945 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003946{
Tejun Heo6770c642014-05-13 12:16:21 -04003947 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003948
Michal Hockod8423012012-10-26 13:37:29 +02003949 if (mem_cgroup_is_root(memcg))
3950 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003951 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003952}
3953
Tejun Heo182446d2013-08-08 20:11:24 -04003954static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3955 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003956{
Tejun Heo182446d2013-08-08 20:11:24 -04003957 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003958}
3959
Tejun Heo182446d2013-08-08 20:11:24 -04003960static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3961 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003962{
3963 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04003964 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04003965 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003966
Glauber Costa09998212013-02-22 16:34:55 -08003967 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07003968
3969 if (memcg->use_hierarchy == val)
3970 goto out;
3971
Balbir Singh18f59ea2009-01-07 18:08:07 -08003972 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003973 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003974 * in the child subtrees. If it is unset, then the change can
3975 * occur, provided the current cgroup has no children.
3976 *
3977 * For the root cgroup, parent_mem is NULL, we allow value to be
3978 * set if there are no children.
3979 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003980 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003981 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04003982 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003983 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003984 else
3985 retval = -EBUSY;
3986 } else
3987 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07003988
3989out:
Glauber Costa09998212013-02-22 16:34:55 -08003990 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003991
3992 return retval;
3993}
3994
Johannes Weinerce00a962014-09-05 08:43:57 -04003995static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
3996 enum mem_cgroup_stat_index idx)
3997{
3998 struct mem_cgroup *iter;
3999 long val = 0;
4000
4001 /* Per-cpu values can be negative, use a signed accumulator */
4002 for_each_mem_cgroup_tree(iter, memcg)
4003 val += mem_cgroup_read_stat(iter, idx);
4004
4005 if (val < 0) /* race ? */
4006 val = 0;
4007 return val;
4008}
4009
4010static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
4011{
4012 u64 val;
4013
4014 if (!mem_cgroup_is_root(memcg)) {
4015 if (!swap)
4016 return res_counter_read_u64(&memcg->res, RES_USAGE);
4017 else
4018 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
4019 }
4020
4021 /*
4022 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
4023 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
4024 */
4025 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
4026 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
4027
4028 if (swap)
4029 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
4030
4031 return val << PAGE_SHIFT;
4032}
4033
4034
Tejun Heo791badb2013-12-05 12:28:02 -05004035static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07004036 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004037{
Tejun Heo182446d2013-08-08 20:11:24 -04004038 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner05b84302014-08-06 16:05:59 -07004039 enum res_type type = MEMFILE_TYPE(cft->private);
4040 int name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07004041
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004042 switch (type) {
4043 case _MEM:
Johannes Weinerce00a962014-09-05 08:43:57 -04004044 if (name == RES_USAGE)
4045 return mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004046 return res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004047 case _MEMSWAP:
Johannes Weinerce00a962014-09-05 08:43:57 -04004048 if (name == RES_USAGE)
4049 return mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004050 return res_counter_read_u64(&memcg->memsw, name);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004051 case _KMEM:
Johannes Weiner05b84302014-08-06 16:05:59 -07004052 return res_counter_read_u64(&memcg->kmem, name);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004053 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004054 default:
4055 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004056 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004057}
Glauber Costa510fc4e2012-12-18 14:21:47 -08004058
Glauber Costa510fc4e2012-12-18 14:21:47 -08004059#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovd6441632014-01-23 15:53:09 -08004060/* should be called with activate_kmem_mutex held */
4061static int __memcg_activate_kmem(struct mem_cgroup *memcg,
4062 unsigned long long limit)
4063{
4064 int err = 0;
4065 int memcg_id;
4066
4067 if (memcg_kmem_is_active(memcg))
4068 return 0;
4069
4070 /*
4071 * We are going to allocate memory for data shared by all memory
4072 * cgroups so let's stop accounting here.
4073 */
4074 memcg_stop_kmem_account();
4075
Glauber Costa510fc4e2012-12-18 14:21:47 -08004076 /*
4077 * For simplicity, we won't allow this to be disabled. It also can't
4078 * be changed if the cgroup has children already, or if tasks had
4079 * already joined.
4080 *
4081 * If tasks join before we set the limit, a person looking at
4082 * kmem.usage_in_bytes will have no way to determine when it took
4083 * place, which makes the value quite meaningless.
4084 *
4085 * After it first became limited, changes in the value of the limit are
4086 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08004087 */
Glauber Costa09998212013-02-22 16:34:55 -08004088 mutex_lock(&memcg_create_mutex);
Tejun Heoea280e72014-05-16 13:22:48 -04004089 if (cgroup_has_tasks(memcg->css.cgroup) ||
4090 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08004091 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08004092 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004093 if (err)
4094 goto out;
4095
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07004096 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08004097 if (memcg_id < 0) {
4098 err = memcg_id;
4099 goto out;
4100 }
4101
Vladimir Davydovd6441632014-01-23 15:53:09 -08004102 memcg->kmemcg_id = memcg_id;
4103 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004104
4105 /*
4106 * We couldn't have accounted to this cgroup, because it hasn't got the
4107 * active bit set yet, so this should succeed.
4108 */
4109 err = res_counter_set_limit(&memcg->kmem, limit);
4110 VM_BUG_ON(err);
4111
4112 static_key_slow_inc(&memcg_kmem_enabled_key);
4113 /*
4114 * Setting the active bit after enabling static branching will
4115 * guarantee no one starts accounting before all call sites are
4116 * patched.
4117 */
4118 memcg_kmem_set_active(memcg);
4119out:
4120 memcg_resume_kmem_account();
4121 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08004122}
4123
4124static int memcg_activate_kmem(struct mem_cgroup *memcg,
4125 unsigned long long limit)
4126{
4127 int ret;
4128
4129 mutex_lock(&activate_kmem_mutex);
4130 ret = __memcg_activate_kmem(memcg, limit);
4131 mutex_unlock(&activate_kmem_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004132 return ret;
4133}
4134
Vladimir Davydovd6441632014-01-23 15:53:09 -08004135static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
4136 unsigned long long val)
4137{
4138 int ret;
4139
4140 if (!memcg_kmem_is_active(memcg))
4141 ret = memcg_activate_kmem(memcg, val);
4142 else
4143 ret = res_counter_set_limit(&memcg->kmem, val);
4144 return ret;
4145}
4146
Glauber Costa55007d82012-12-18 14:22:38 -08004147static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08004148{
Glauber Costa55007d82012-12-18 14:22:38 -08004149 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08004150 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004151
Glauber Costa510fc4e2012-12-18 14:21:47 -08004152 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08004153 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08004154
Vladimir Davydovd6441632014-01-23 15:53:09 -08004155 mutex_lock(&activate_kmem_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08004156 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08004157 * If the parent cgroup is not kmem-active now, it cannot be activated
4158 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08004159 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08004160 if (memcg_kmem_is_active(parent))
4161 ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX);
4162 mutex_unlock(&activate_kmem_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08004163 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08004164}
Vladimir Davydovd6441632014-01-23 15:53:09 -08004165#else
4166static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
4167 unsigned long long val)
4168{
4169 return -EINVAL;
4170}
Hugh Dickins6d0439902013-02-22 16:35:50 -08004171#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08004172
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004173/*
4174 * The user of this function is...
4175 * RES_LIMIT.
4176 */
Tejun Heo451af502014-05-13 12:16:21 -04004177static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
4178 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004179{
Tejun Heo451af502014-05-13 12:16:21 -04004180 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Glauber Costa86ae53e2012-12-18 14:21:45 -08004181 enum res_type type;
4182 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004183 unsigned long long val;
4184 int ret;
4185
Tejun Heo451af502014-05-13 12:16:21 -04004186 buf = strstrip(buf);
4187 type = MEMFILE_TYPE(of_cft(of)->private);
4188 name = MEMFILE_ATTR(of_cft(of)->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07004189
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004190 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004191 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07004192 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
4193 ret = -EINVAL;
4194 break;
4195 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004196 /* This function does all necessary parse...reuse it */
Tejun Heo451af502014-05-13 12:16:21 -04004197 ret = res_counter_memparse_write_strategy(buf, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004198 if (ret)
4199 break;
4200 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004201 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004202 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004203 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004204 else if (type == _KMEM)
Vladimir Davydovd6441632014-01-23 15:53:09 -08004205 ret = memcg_update_kmem_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004206 else
4207 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004208 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07004209 case RES_SOFT_LIMIT:
Tejun Heo451af502014-05-13 12:16:21 -04004210 ret = res_counter_memparse_write_strategy(buf, &val);
Balbir Singh296c81d2009-09-23 15:56:36 -07004211 if (ret)
4212 break;
4213 /*
4214 * For memsw, soft limits are hard to implement in terms
4215 * of semantics, for now, we support soft limits for
4216 * control without swap
4217 */
4218 if (type == _MEM)
4219 ret = res_counter_set_soft_limit(&memcg->res, val);
4220 else
4221 ret = -EINVAL;
4222 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004223 default:
4224 ret = -EINVAL; /* should be BUG() ? */
4225 break;
4226 }
Tejun Heo451af502014-05-13 12:16:21 -04004227 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004228}
4229
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004230static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
4231 unsigned long long *mem_limit, unsigned long long *memsw_limit)
4232{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004233 unsigned long long min_limit, min_memsw_limit, tmp;
4234
4235 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4236 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004237 if (!memcg->use_hierarchy)
4238 goto out;
4239
Tejun Heo5c9d5352014-05-16 13:22:48 -04004240 while (memcg->css.parent) {
4241 memcg = mem_cgroup_from_css(memcg->css.parent);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004242 if (!memcg->use_hierarchy)
4243 break;
4244 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4245 min_limit = min(min_limit, tmp);
4246 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4247 min_memsw_limit = min(min_memsw_limit, tmp);
4248 }
4249out:
4250 *mem_limit = min_limit;
4251 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004252}
4253
Tejun Heo6770c642014-05-13 12:16:21 -04004254static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
4255 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004256{
Tejun Heo6770c642014-05-13 12:16:21 -04004257 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Glauber Costa86ae53e2012-12-18 14:21:45 -08004258 int name;
4259 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004260
Tejun Heo6770c642014-05-13 12:16:21 -04004261 type = MEMFILE_TYPE(of_cft(of)->private);
4262 name = MEMFILE_ATTR(of_cft(of)->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07004263
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004264 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004265 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004266 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004267 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004268 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004269 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004270 else if (type == _KMEM)
4271 res_counter_reset_max(&memcg->kmem);
4272 else
4273 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004274 break;
4275 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004276 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004277 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004278 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004279 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08004280 else if (type == _KMEM)
4281 res_counter_reset_failcnt(&memcg->kmem);
4282 else
4283 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004284 break;
4285 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004286
Tejun Heo6770c642014-05-13 12:16:21 -04004287 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004288}
4289
Tejun Heo182446d2013-08-08 20:11:24 -04004290static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004291 struct cftype *cft)
4292{
Tejun Heo182446d2013-08-08 20:11:24 -04004293 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004294}
4295
Daisuke Nishimura02491442010-03-10 15:22:17 -08004296#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04004297static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004298 struct cftype *cft, u64 val)
4299{
Tejun Heo182446d2013-08-08 20:11:24 -04004300 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004301
4302 if (val >= (1 << NR_MOVE_TYPE))
4303 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004304
Glauber Costaee5e8472013-02-22 16:34:50 -08004305 /*
4306 * No kind of locking is needed in here, because ->can_attach() will
4307 * check this value once in the beginning of the process, and then carry
4308 * on with stale data. This means that changes to this value will only
4309 * affect task migrations starting after the change.
4310 */
4311 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004312 return 0;
4313}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004314#else
Tejun Heo182446d2013-08-08 20:11:24 -04004315static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004316 struct cftype *cft, u64 val)
4317{
4318 return -ENOSYS;
4319}
4320#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004321
Ying Han406eb0c2011-05-26 16:25:37 -07004322#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05004323static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07004324{
Greg Thelen25485de2013-11-12 15:07:40 -08004325 struct numa_stat {
4326 const char *name;
4327 unsigned int lru_mask;
4328 };
4329
4330 static const struct numa_stat stats[] = {
4331 { "total", LRU_ALL },
4332 { "file", LRU_ALL_FILE },
4333 { "anon", LRU_ALL_ANON },
4334 { "unevictable", BIT(LRU_UNEVICTABLE) },
4335 };
4336 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07004337 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08004338 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05004339 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07004340
Greg Thelen25485de2013-11-12 15:07:40 -08004341 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
4342 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
4343 seq_printf(m, "%s=%lu", stat->name, nr);
4344 for_each_node_state(nid, N_MEMORY) {
4345 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4346 stat->lru_mask);
4347 seq_printf(m, " N%d=%lu", nid, nr);
4348 }
4349 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004350 }
Ying Han406eb0c2011-05-26 16:25:37 -07004351
Ying Han071aee12013-11-12 15:07:41 -08004352 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
4353 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07004354
Ying Han071aee12013-11-12 15:07:41 -08004355 nr = 0;
4356 for_each_mem_cgroup_tree(iter, memcg)
4357 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
4358 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
4359 for_each_node_state(nid, N_MEMORY) {
4360 nr = 0;
4361 for_each_mem_cgroup_tree(iter, memcg)
4362 nr += mem_cgroup_node_nr_lru_pages(
4363 iter, nid, stat->lru_mask);
4364 seq_printf(m, " N%d=%lu", nid, nr);
4365 }
4366 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004367 }
Ying Han406eb0c2011-05-26 16:25:37 -07004368
Ying Han406eb0c2011-05-26 16:25:37 -07004369 return 0;
4370}
4371#endif /* CONFIG_NUMA */
4372
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004373static inline void mem_cgroup_lru_names_not_uptodate(void)
4374{
4375 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4376}
4377
Tejun Heo2da8ca82013-12-05 12:28:04 -05004378static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004379{
Tejun Heo2da8ca82013-12-05 12:28:04 -05004380 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004381 struct mem_cgroup *mi;
4382 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004383
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004384 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004385 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004386 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004387 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4388 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004389 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004390
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004391 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4392 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4393 mem_cgroup_read_events(memcg, i));
4394
4395 for (i = 0; i < NR_LRU_LISTS; i++)
4396 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4397 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4398
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004399 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004400 {
4401 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004402 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004403 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004404 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004405 seq_printf(m, "hierarchical_memsw_limit %llu\n",
4406 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004407 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004408
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004409 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4410 long long val = 0;
4411
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004412 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004413 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004414 for_each_mem_cgroup_tree(mi, memcg)
4415 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4416 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4417 }
4418
4419 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4420 unsigned long long val = 0;
4421
4422 for_each_mem_cgroup_tree(mi, memcg)
4423 val += mem_cgroup_read_events(mi, i);
4424 seq_printf(m, "total_%s %llu\n",
4425 mem_cgroup_events_names[i], val);
4426 }
4427
4428 for (i = 0; i < NR_LRU_LISTS; i++) {
4429 unsigned long long val = 0;
4430
4431 for_each_mem_cgroup_tree(mi, memcg)
4432 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4433 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004434 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004435
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004436#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004437 {
4438 int nid, zid;
4439 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07004440 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004441 unsigned long recent_rotated[2] = {0, 0};
4442 unsigned long recent_scanned[2] = {0, 0};
4443
4444 for_each_online_node(nid)
4445 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07004446 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07004447 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004448
Hugh Dickins89abfab2012-05-29 15:06:53 -07004449 recent_rotated[0] += rstat->recent_rotated[0];
4450 recent_rotated[1] += rstat->recent_rotated[1];
4451 recent_scanned[0] += rstat->recent_scanned[0];
4452 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004453 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07004454 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4455 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4456 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4457 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004458 }
4459#endif
4460
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004461 return 0;
4462}
4463
Tejun Heo182446d2013-08-08 20:11:24 -04004464static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
4465 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004466{
Tejun Heo182446d2013-08-08 20:11:24 -04004467 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004468
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004469 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004470}
4471
Tejun Heo182446d2013-08-08 20:11:24 -04004472static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
4473 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004474{
Tejun Heo182446d2013-08-08 20:11:24 -04004475 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08004476
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004477 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004478 return -EINVAL;
4479
Linus Torvalds14208b02014-06-09 15:03:33 -07004480 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004481 memcg->swappiness = val;
4482 else
4483 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08004484
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004485 return 0;
4486}
4487
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004488static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4489{
4490 struct mem_cgroup_threshold_ary *t;
4491 u64 usage;
4492 int i;
4493
4494 rcu_read_lock();
4495 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004496 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004497 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004498 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004499
4500 if (!t)
4501 goto unlock;
4502
Johannes Weinerce00a962014-09-05 08:43:57 -04004503 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004504
4505 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004506 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004507 * If it's not true, a threshold was crossed after last
4508 * call of __mem_cgroup_threshold().
4509 */
Phil Carmody5407a562010-05-26 14:42:42 -07004510 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511
4512 /*
4513 * Iterate backward over array of thresholds starting from
4514 * current_threshold and check if a threshold is crossed.
4515 * If none of thresholds below usage is crossed, we read
4516 * only one element of the array here.
4517 */
4518 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4519 eventfd_signal(t->entries[i].eventfd, 1);
4520
4521 /* i = current_threshold + 1 */
4522 i++;
4523
4524 /*
4525 * Iterate forward over array of thresholds starting from
4526 * current_threshold+1 and check if a threshold is crossed.
4527 * If none of thresholds above usage is crossed, we read
4528 * only one element of the array here.
4529 */
4530 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4531 eventfd_signal(t->entries[i].eventfd, 1);
4532
4533 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004534 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004535unlock:
4536 rcu_read_unlock();
4537}
4538
4539static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4540{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004541 while (memcg) {
4542 __mem_cgroup_threshold(memcg, false);
4543 if (do_swap_account)
4544 __mem_cgroup_threshold(memcg, true);
4545
4546 memcg = parent_mem_cgroup(memcg);
4547 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004548}
4549
4550static int compare_thresholds(const void *a, const void *b)
4551{
4552 const struct mem_cgroup_threshold *_a = a;
4553 const struct mem_cgroup_threshold *_b = b;
4554
Greg Thelen2bff24a2013-09-11 14:23:08 -07004555 if (_a->threshold > _b->threshold)
4556 return 1;
4557
4558 if (_a->threshold < _b->threshold)
4559 return -1;
4560
4561 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004562}
4563
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004564static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004565{
4566 struct mem_cgroup_eventfd_list *ev;
4567
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004568 spin_lock(&memcg_oom_lock);
4569
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004570 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004571 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004572
4573 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004574 return 0;
4575}
4576
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004577static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004578{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004579 struct mem_cgroup *iter;
4580
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004581 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004582 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004583}
4584
Tejun Heo59b6f872013-11-22 18:20:43 -05004585static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004586 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004587{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004588 struct mem_cgroup_thresholds *thresholds;
4589 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004590 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004591 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004592
4593 ret = res_counter_memparse_write_strategy(args, &threshold);
4594 if (ret)
4595 return ret;
4596
4597 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004598
Johannes Weiner05b84302014-08-06 16:05:59 -07004599 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004600 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004601 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004602 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004603 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004604 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004605 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004606 BUG();
4607
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004608 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004609 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004610 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4611
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004612 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004613
4614 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004615 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004616 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004617 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004618 ret = -ENOMEM;
4619 goto unlock;
4620 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004621 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004622
4623 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004624 if (thresholds->primary) {
4625 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004626 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004627 }
4628
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004629 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004630 new->entries[size - 1].eventfd = eventfd;
4631 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004632
4633 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004634 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004635 compare_thresholds, NULL);
4636
4637 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004638 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004639 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004640 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004641 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004642 * new->current_threshold will not be used until
4643 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004644 * it here.
4645 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004646 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004647 } else
4648 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004649 }
4650
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004651 /* Free old spare buffer and save old primary buffer as spare */
4652 kfree(thresholds->spare);
4653 thresholds->spare = thresholds->primary;
4654
4655 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004656
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004657 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004658 synchronize_rcu();
4659
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004660unlock:
4661 mutex_unlock(&memcg->thresholds_lock);
4662
4663 return ret;
4664}
4665
Tejun Heo59b6f872013-11-22 18:20:43 -05004666static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004667 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004668{
Tejun Heo59b6f872013-11-22 18:20:43 -05004669 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004670}
4671
Tejun Heo59b6f872013-11-22 18:20:43 -05004672static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004673 struct eventfd_ctx *eventfd, const char *args)
4674{
Tejun Heo59b6f872013-11-22 18:20:43 -05004675 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004676}
4677
Tejun Heo59b6f872013-11-22 18:20:43 -05004678static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004679 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004680{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004681 struct mem_cgroup_thresholds *thresholds;
4682 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004683 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004684 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004685
4686 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07004687
4688 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004689 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004690 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004691 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004692 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004693 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004694 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004695 BUG();
4696
Anton Vorontsov371528c2012-02-24 05:14:46 +04004697 if (!thresholds->primary)
4698 goto unlock;
4699
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004700 /* Check if a threshold crossed before removing */
4701 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4702
4703 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004704 size = 0;
4705 for (i = 0; i < thresholds->primary->size; i++) {
4706 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004707 size++;
4708 }
4709
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004710 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004711
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004712 /* Set thresholds array to NULL if we don't have thresholds */
4713 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004714 kfree(new);
4715 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004716 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004717 }
4718
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004719 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004720
4721 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004722 new->current_threshold = -1;
4723 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4724 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004725 continue;
4726
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004727 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004728 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004729 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004730 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004731 * until rcu_assign_pointer(), so it's safe to increment
4732 * it here.
4733 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004734 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004735 }
4736 j++;
4737 }
4738
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004739swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004740 /* Swap primary and spare array */
4741 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004742 /* If all events are unregistered, free the spare array */
4743 if (!new) {
4744 kfree(thresholds->spare);
4745 thresholds->spare = NULL;
4746 }
4747
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004748 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004749
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004750 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004751 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004752unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004753 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004754}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004755
Tejun Heo59b6f872013-11-22 18:20:43 -05004756static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004757 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004758{
Tejun Heo59b6f872013-11-22 18:20:43 -05004759 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004760}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004761
Tejun Heo59b6f872013-11-22 18:20:43 -05004762static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004763 struct eventfd_ctx *eventfd)
4764{
Tejun Heo59b6f872013-11-22 18:20:43 -05004765 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004766}
4767
Tejun Heo59b6f872013-11-22 18:20:43 -05004768static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004769 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004770{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004771 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004772
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004773 event = kmalloc(sizeof(*event), GFP_KERNEL);
4774 if (!event)
4775 return -ENOMEM;
4776
Michal Hocko1af8efe2011-07-26 16:08:24 -07004777 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004778
4779 event->eventfd = eventfd;
4780 list_add(&event->list, &memcg->oom_notify);
4781
4782 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004783 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004784 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004785 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004786
4787 return 0;
4788}
4789
Tejun Heo59b6f872013-11-22 18:20:43 -05004790static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004791 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004792{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004793 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004794
Michal Hocko1af8efe2011-07-26 16:08:24 -07004795 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004796
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004797 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004798 if (ev->eventfd == eventfd) {
4799 list_del(&ev->list);
4800 kfree(ev);
4801 }
4802 }
4803
Michal Hocko1af8efe2011-07-26 16:08:24 -07004804 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004805}
4806
Tejun Heo2da8ca82013-12-05 12:28:04 -05004807static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004808{
Tejun Heo2da8ca82013-12-05 12:28:04 -05004809 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004810
Tejun Heo791badb2013-12-05 12:28:02 -05004811 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
4812 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004813 return 0;
4814}
4815
Tejun Heo182446d2013-08-08 20:11:24 -04004816static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004817 struct cftype *cft, u64 val)
4818{
Tejun Heo182446d2013-08-08 20:11:24 -04004819 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004820
4821 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07004822 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004823 return -EINVAL;
4824
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004825 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004826 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004827 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004828
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004829 return 0;
4830}
4831
Andrew Mortonc255a452012-07-31 16:43:02 -07004832#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03004833static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004834{
Glauber Costa55007d82012-12-18 14:22:38 -08004835 int ret;
4836
Glauber Costa2633d7a2012-12-18 14:22:34 -08004837 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08004838 ret = memcg_propagate_kmem(memcg);
4839 if (ret)
4840 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08004841
Glauber Costa1d62e432012-04-09 19:36:33 -03004842 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07004843}
Glauber Costae5671df2011-12-11 21:47:01 +00004844
Li Zefan10d5ebf2013-07-08 16:00:33 -07004845static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004846{
Glauber Costa1d62e432012-04-09 19:36:33 -03004847 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07004848}
4849
4850static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
4851{
4852 if (!memcg_kmem_is_active(memcg))
4853 return;
4854
4855 /*
4856 * kmem charges can outlive the cgroup. In the case of slab
4857 * pages, for instance, a page contain objects from various
4858 * processes. As we prevent from taking a reference for every
4859 * such allocation we have to be careful when doing uncharge
4860 * (see memcg_uncharge_kmem) and here during offlining.
4861 *
4862 * The idea is that that only the _last_ uncharge which sees
4863 * the dead memcg will drop the last reference. An additional
4864 * reference is taken here before the group is marked dead
4865 * which is then paired with css_put during uncharge resp. here.
4866 *
4867 * Although this might sound strange as this path is called from
Tejun Heoec903c02014-05-13 12:11:01 -04004868 * css_offline() when the referencemight have dropped down to 0 and
4869 * shouldn't be incremented anymore (css_tryget_online() would
4870 * fail) we do not have other options because of the kmem
4871 * allocations lifetime.
Li Zefan10d5ebf2013-07-08 16:00:33 -07004872 */
4873 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08004874
4875 memcg_kmem_mark_dead(memcg);
4876
4877 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
4878 return;
4879
Glauber Costa7de37682012-12-18 14:22:07 -08004880 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07004881 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004882}
Glauber Costae5671df2011-12-11 21:47:01 +00004883#else
Glauber Costacbe128e32012-04-09 19:36:34 -03004884static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00004885{
4886 return 0;
4887}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004888
Li Zefan10d5ebf2013-07-08 16:00:33 -07004889static void memcg_destroy_kmem(struct mem_cgroup *memcg)
4890{
4891}
4892
4893static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004894{
4895}
Glauber Costae5671df2011-12-11 21:47:01 +00004896#endif
4897
Tejun Heo79bd9812013-11-22 18:20:42 -05004898/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004899 * DO NOT USE IN NEW FILES.
4900 *
4901 * "cgroup.event_control" implementation.
4902 *
4903 * This is way over-engineered. It tries to support fully configurable
4904 * events for each user. Such level of flexibility is completely
4905 * unnecessary especially in the light of the planned unified hierarchy.
4906 *
4907 * Please deprecate this and replace with something simpler if at all
4908 * possible.
4909 */
4910
4911/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004912 * Unregister event and free resources.
4913 *
4914 * Gets called from workqueue.
4915 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004916static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004917{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004918 struct mem_cgroup_event *event =
4919 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004920 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004921
4922 remove_wait_queue(event->wqh, &event->wait);
4923
Tejun Heo59b6f872013-11-22 18:20:43 -05004924 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004925
4926 /* Notify userspace the event is going away. */
4927 eventfd_signal(event->eventfd, 1);
4928
4929 eventfd_ctx_put(event->eventfd);
4930 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004931 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004932}
4933
4934/*
4935 * Gets called on POLLHUP on eventfd when user closes it.
4936 *
4937 * Called with wqh->lock held and interrupts disabled.
4938 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004939static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
4940 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004941{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004942 struct mem_cgroup_event *event =
4943 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004944 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004945 unsigned long flags = (unsigned long)key;
4946
4947 if (flags & POLLHUP) {
4948 /*
4949 * If the event has been detached at cgroup removal, we
4950 * can simply return knowing the other side will cleanup
4951 * for us.
4952 *
4953 * We can't race against event freeing since the other
4954 * side will require wqh->lock via remove_wait_queue(),
4955 * which we hold.
4956 */
Tejun Heofba94802013-11-22 18:20:43 -05004957 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004958 if (!list_empty(&event->list)) {
4959 list_del_init(&event->list);
4960 /*
4961 * We are in atomic context, but cgroup_event_remove()
4962 * may sleep, so we have to call it in workqueue.
4963 */
4964 schedule_work(&event->remove);
4965 }
Tejun Heofba94802013-11-22 18:20:43 -05004966 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004967 }
4968
4969 return 0;
4970}
4971
Tejun Heo3bc942f2013-11-22 18:20:44 -05004972static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004973 wait_queue_head_t *wqh, poll_table *pt)
4974{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004975 struct mem_cgroup_event *event =
4976 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004977
4978 event->wqh = wqh;
4979 add_wait_queue(wqh, &event->wait);
4980}
4981
4982/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004983 * DO NOT USE IN NEW FILES.
4984 *
Tejun Heo79bd9812013-11-22 18:20:42 -05004985 * Parse input and register new cgroup event handler.
4986 *
4987 * Input must be in format '<event_fd> <control_fd> <args>'.
4988 * Interpretation of args is defined by control file implementation.
4989 */
Tejun Heo451af502014-05-13 12:16:21 -04004990static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4991 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05004992{
Tejun Heo451af502014-05-13 12:16:21 -04004993 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05004994 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004995 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05004996 struct cgroup_subsys_state *cfile_css;
4997 unsigned int efd, cfd;
4998 struct fd efile;
4999 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05005000 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05005001 char *endp;
5002 int ret;
5003
Tejun Heo451af502014-05-13 12:16:21 -04005004 buf = strstrip(buf);
5005
5006 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05005007 if (*endp != ' ')
5008 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04005009 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05005010
Tejun Heo451af502014-05-13 12:16:21 -04005011 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05005012 if ((*endp != ' ') && (*endp != '\0'))
5013 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04005014 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05005015
5016 event = kzalloc(sizeof(*event), GFP_KERNEL);
5017 if (!event)
5018 return -ENOMEM;
5019
Tejun Heo59b6f872013-11-22 18:20:43 -05005020 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005021 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005022 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
5023 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
5024 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05005025
5026 efile = fdget(efd);
5027 if (!efile.file) {
5028 ret = -EBADF;
5029 goto out_kfree;
5030 }
5031
5032 event->eventfd = eventfd_ctx_fileget(efile.file);
5033 if (IS_ERR(event->eventfd)) {
5034 ret = PTR_ERR(event->eventfd);
5035 goto out_put_efile;
5036 }
5037
5038 cfile = fdget(cfd);
5039 if (!cfile.file) {
5040 ret = -EBADF;
5041 goto out_put_eventfd;
5042 }
5043
5044 /* the process need read permission on control file */
5045 /* AV: shouldn't we check that it's been opened for read instead? */
5046 ret = inode_permission(file_inode(cfile.file), MAY_READ);
5047 if (ret < 0)
5048 goto out_put_cfile;
5049
Tejun Heo79bd9812013-11-22 18:20:42 -05005050 /*
Tejun Heofba94802013-11-22 18:20:43 -05005051 * Determine the event callbacks and set them in @event. This used
5052 * to be done via struct cftype but cgroup core no longer knows
5053 * about these events. The following is crude but the whole thing
5054 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05005055 *
5056 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05005057 */
5058 name = cfile.file->f_dentry->d_name.name;
5059
5060 if (!strcmp(name, "memory.usage_in_bytes")) {
5061 event->register_event = mem_cgroup_usage_register_event;
5062 event->unregister_event = mem_cgroup_usage_unregister_event;
5063 } else if (!strcmp(name, "memory.oom_control")) {
5064 event->register_event = mem_cgroup_oom_register_event;
5065 event->unregister_event = mem_cgroup_oom_unregister_event;
5066 } else if (!strcmp(name, "memory.pressure_level")) {
5067 event->register_event = vmpressure_register_event;
5068 event->unregister_event = vmpressure_unregister_event;
5069 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05005070 event->register_event = memsw_cgroup_usage_register_event;
5071 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05005072 } else {
5073 ret = -EINVAL;
5074 goto out_put_cfile;
5075 }
5076
5077 /*
Tejun Heob5557c42013-11-22 18:20:42 -05005078 * Verify @cfile should belong to @css. Also, remaining events are
5079 * automatically removed on cgroup destruction but the removal is
5080 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05005081 */
Tejun Heoec903c02014-05-13 12:11:01 -04005082 cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent,
5083 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05005084 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05005085 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05005086 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05005087 if (cfile_css != css) {
5088 css_put(cfile_css);
5089 goto out_put_cfile;
5090 }
Tejun Heo79bd9812013-11-22 18:20:42 -05005091
Tejun Heo451af502014-05-13 12:16:21 -04005092 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05005093 if (ret)
5094 goto out_put_css;
5095
5096 efile.file->f_op->poll(efile.file, &event->pt);
5097
Tejun Heofba94802013-11-22 18:20:43 -05005098 spin_lock(&memcg->event_list_lock);
5099 list_add(&event->list, &memcg->event_list);
5100 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05005101
5102 fdput(cfile);
5103 fdput(efile);
5104
Tejun Heo451af502014-05-13 12:16:21 -04005105 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05005106
5107out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05005108 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05005109out_put_cfile:
5110 fdput(cfile);
5111out_put_eventfd:
5112 eventfd_ctx_put(event->eventfd);
5113out_put_efile:
5114 fdput(efile);
5115out_kfree:
5116 kfree(event);
5117
5118 return ret;
5119}
5120
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005121static struct cftype mem_cgroup_files[] = {
5122 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005123 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005124 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005125 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005126 },
5127 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005128 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005129 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005130 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005131 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005132 },
5133 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005134 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005135 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04005136 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005137 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005138 },
5139 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005140 .name = "soft_limit_in_bytes",
5141 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04005142 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005143 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07005144 },
5145 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005146 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005147 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005148 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005149 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005150 },
Balbir Singh8697d332008-02-07 00:13:59 -08005151 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005152 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005153 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005154 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005155 {
5156 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04005157 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005158 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005159 {
5160 .name = "use_hierarchy",
5161 .write_u64 = mem_cgroup_hierarchy_write,
5162 .read_u64 = mem_cgroup_hierarchy_read,
5163 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005164 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05005165 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04005166 .write = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05005167 .flags = CFTYPE_NO_PREFIX,
5168 .mode = S_IWUGO,
5169 },
5170 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005171 .name = "swappiness",
5172 .read_u64 = mem_cgroup_swappiness_read,
5173 .write_u64 = mem_cgroup_swappiness_write,
5174 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005175 {
5176 .name = "move_charge_at_immigrate",
5177 .read_u64 = mem_cgroup_move_charge_read,
5178 .write_u64 = mem_cgroup_move_charge_write,
5179 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005180 {
5181 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005182 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005183 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005184 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
5185 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005186 {
5187 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005188 },
Ying Han406eb0c2011-05-26 16:25:37 -07005189#ifdef CONFIG_NUMA
5190 {
5191 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005192 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07005193 },
5194#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005195#ifdef CONFIG_MEMCG_KMEM
5196 {
5197 .name = "kmem.limit_in_bytes",
5198 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04005199 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005200 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005201 },
5202 {
5203 .name = "kmem.usage_in_bytes",
5204 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005205 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005206 },
5207 {
5208 .name = "kmem.failcnt",
5209 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005210 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005211 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005212 },
5213 {
5214 .name = "kmem.max_usage_in_bytes",
5215 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005216 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005217 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005218 },
Glauber Costa749c5412012-12-18 14:23:01 -08005219#ifdef CONFIG_SLABINFO
5220 {
5221 .name = "kmem.slabinfo",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005222 .seq_show = mem_cgroup_slabinfo_read,
Glauber Costa749c5412012-12-18 14:23:01 -08005223 },
5224#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005225#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07005226 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07005227};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005228
Michal Hocko2d110852013-02-22 16:34:43 -08005229#ifdef CONFIG_MEMCG_SWAP
5230static struct cftype memsw_cgroup_files[] = {
5231 {
5232 .name = "memsw.usage_in_bytes",
5233 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005234 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08005235 },
5236 {
5237 .name = "memsw.max_usage_in_bytes",
5238 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005239 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005240 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08005241 },
5242 {
5243 .name = "memsw.limit_in_bytes",
5244 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04005245 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005246 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08005247 },
5248 {
5249 .name = "memsw.failcnt",
5250 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005251 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005252 .read_u64 = mem_cgroup_read_u64,
Michal Hocko2d110852013-02-22 16:34:43 -08005253 },
5254 { }, /* terminate */
5255};
5256#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005257static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005258{
5259 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005260 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005261 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005262 /*
5263 * This routine is called against possible nodes.
5264 * But it's BUG to call kmalloc() against offline node.
5265 *
5266 * TODO: this routine can waste much memory for nodes which will
5267 * never be onlined. It's better to use memory hotplug callback
5268 * function.
5269 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005270 if (!node_state(node, N_NORMAL_MEMORY))
5271 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08005272 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005273 if (!pn)
5274 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005275
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005276 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5277 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005278 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07005279 mz->usage_in_excess = 0;
5280 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005281 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005282 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005283 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005284 return 0;
5285}
5286
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005287static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005288{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005289 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005290}
5291
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005292static struct mem_cgroup *mem_cgroup_alloc(void)
5293{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005294 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005295 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005296
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005297 size = sizeof(struct mem_cgroup);
5298 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005299
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005300 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005301 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005302 return NULL;
5303
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005304 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
5305 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08005306 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005307 spin_lock_init(&memcg->pcp_counter_lock);
5308 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08005309
5310out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005311 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08005312 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005313}
5314
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005315/*
Glauber Costac8b2a362012-12-18 14:22:13 -08005316 * At destroying mem_cgroup, references from swap_cgroup can remain.
5317 * (scanning all at force_empty is too costly...)
5318 *
5319 * Instead of clearing all references at force_empty, we remember
5320 * the number of reference from swap_cgroup and free mem_cgroup when
5321 * it goes down to 0.
5322 *
5323 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07005324 */
Glauber Costac8b2a362012-12-18 14:22:13 -08005325
5326static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07005327{
Glauber Costac8b2a362012-12-18 14:22:13 -08005328 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07005329
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07005330 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08005331
5332 for_each_node(node)
5333 free_mem_cgroup_per_zone_info(memcg, node);
5334
5335 free_percpu(memcg->stat);
5336
Glauber Costa3f134612012-05-29 15:07:11 -07005337 /*
5338 * We need to make sure that (at least for now), the jump label
5339 * destruction code runs outside of the cgroup lock. This is because
5340 * get_online_cpus(), which is called from the static_branch update,
5341 * can't be called inside the cgroup_lock. cpusets are the ones
5342 * enforcing this dependency, so if they ever change, we might as well.
5343 *
5344 * schedule_work() will guarantee this happens. Be careful if you need
5345 * to move this code around, and make sure it is outside
5346 * the cgroup_lock.
5347 */
Glauber Costaa8964b92012-12-18 14:22:09 -08005348 disarm_static_keys(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005349 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07005350}
Glauber Costa3afe36b2012-05-29 15:07:10 -07005351
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005352/*
5353 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
5354 */
Glauber Costae1aab162011-12-11 21:47:03 +00005355struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005356{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005357 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005358 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005359 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005360}
Glauber Costae1aab162011-12-11 21:47:03 +00005361EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005362
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07005363static void __init mem_cgroup_soft_limit_tree_init(void)
5364{
5365 struct mem_cgroup_tree_per_node *rtpn;
5366 struct mem_cgroup_tree_per_zone *rtpz;
5367 int tmp, node, zone;
5368
5369 for_each_node(node) {
5370 tmp = node;
5371 if (!node_state(node, N_NORMAL_MEMORY))
5372 tmp = -1;
5373 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
5374 BUG_ON(!rtpn);
5375
5376 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5377
5378 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5379 rtpz = &rtpn->rb_tree_per_zone[zone];
5380 rtpz->rb_root = RB_ROOT;
5381 spin_lock_init(&rtpz->lock);
5382 }
5383 }
5384}
5385
Li Zefan0eb253e2009-01-15 13:51:25 -08005386static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04005387mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005388{
Glauber Costad142e3e2013-02-22 16:34:52 -08005389 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005390 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005391 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005392
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005393 memcg = mem_cgroup_alloc();
5394 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005395 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005396
Bob Liu3ed28fa2012-01-12 17:19:04 -08005397 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005398 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005399 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005400
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005401 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04005402 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08005403 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08005404 res_counter_init(&memcg->res, NULL);
5405 res_counter_init(&memcg->memsw, NULL);
5406 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005407 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005408
Glauber Costad142e3e2013-02-22 16:34:52 -08005409 memcg->last_scanned_node = MAX_NUMNODES;
5410 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08005411 memcg->move_charge_at_immigrate = 0;
5412 mutex_init(&memcg->thresholds_lock);
5413 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005414 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05005415 INIT_LIST_HEAD(&memcg->event_list);
5416 spin_lock_init(&memcg->event_list_lock);
Glauber Costad142e3e2013-02-22 16:34:52 -08005417
5418 return &memcg->css;
5419
5420free_out:
5421 __mem_cgroup_free(memcg);
5422 return ERR_PTR(error);
5423}
5424
5425static int
Tejun Heoeb954192013-08-08 20:11:23 -04005426mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08005427{
Tejun Heoeb954192013-08-08 20:11:23 -04005428 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04005429 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005430 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08005431
Tejun Heo15a4c832014-05-04 15:09:14 -04005432 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08005433 return -ENOSPC;
5434
Tejun Heo63876982013-08-08 20:11:23 -04005435 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08005436 return 0;
5437
Glauber Costa09998212013-02-22 16:34:55 -08005438 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08005439
5440 memcg->use_hierarchy = parent->use_hierarchy;
5441 memcg->oom_kill_disable = parent->oom_kill_disable;
5442 memcg->swappiness = mem_cgroup_swappiness(parent);
5443
5444 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005445 res_counter_init(&memcg->res, &parent->res);
5446 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005447 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08005448
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005449 /*
Li Zefan8d76a972013-07-08 16:00:36 -07005450 * No need to take a reference to the parent because cgroup
5451 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005452 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08005453 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04005454 res_counter_init(&memcg->res, NULL);
5455 res_counter_init(&memcg->memsw, NULL);
5456 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07005457 /*
5458 * Deeper hierachy with use_hierarchy == false doesn't make
5459 * much sense so let cgroup subsystem know about this
5460 * unfortunate state in our controller.
5461 */
Glauber Costad142e3e2013-02-22 16:34:52 -08005462 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05005463 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005464 }
Glauber Costa09998212013-02-22 16:34:55 -08005465 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005466
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005467 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
5468 if (ret)
5469 return ret;
5470
5471 /*
5472 * Make sure the memcg is initialized: mem_cgroup_iter()
5473 * orders reading memcg->initialized against its callers
5474 * reading the memcg members.
5475 */
5476 smp_store_release(&memcg->initialized, 1);
5477
5478 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005479}
5480
Michal Hocko5f578162013-04-29 15:07:17 -07005481/*
5482 * Announce all parents that a group from their hierarchy is gone.
5483 */
5484static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
5485{
5486 struct mem_cgroup *parent = memcg;
5487
5488 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07005489 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07005490
5491 /*
5492 * if the root memcg is not hierarchical we have to check it
5493 * explicitely.
5494 */
5495 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07005496 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07005497}
5498
Tejun Heoeb954192013-08-08 20:11:23 -04005499static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005500{
Tejun Heoeb954192013-08-08 20:11:23 -04005501 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005502 struct mem_cgroup_event *event, *tmp;
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08005503 struct cgroup_subsys_state *iter;
Tejun Heo79bd9812013-11-22 18:20:42 -05005504
5505 /*
5506 * Unregister events and notify userspace.
5507 * Notify userspace about cgroup removing only after rmdir of cgroup
5508 * directory to avoid race between userspace and kernelspace.
5509 */
Tejun Heofba94802013-11-22 18:20:43 -05005510 spin_lock(&memcg->event_list_lock);
5511 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05005512 list_del_init(&event->list);
5513 schedule_work(&event->remove);
5514 }
Tejun Heofba94802013-11-22 18:20:43 -05005515 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005516
Li Zefan10d5ebf2013-07-08 16:00:33 -07005517 kmem_cgroup_css_offline(memcg);
5518
Michal Hocko5f578162013-04-29 15:07:17 -07005519 mem_cgroup_invalidate_reclaim_iterators(memcg);
Filipe Brandenburger4fb1a862014-03-03 15:38:25 -08005520
5521 /*
5522 * This requires that offlining is serialized. Right now that is
5523 * guaranteed because css_killed_work_fn() holds the cgroup_mutex.
5524 */
5525 css_for_each_descendant_post(iter, css)
5526 mem_cgroup_reparent_charges(mem_cgroup_from_css(iter));
5527
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07005528 memcg_unregister_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07005529 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005530}
5531
Tejun Heoeb954192013-08-08 20:11:23 -04005532static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005533{
Tejun Heoeb954192013-08-08 20:11:23 -04005534 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner96f1c582013-12-12 17:12:34 -08005535 /*
5536 * XXX: css_offline() would be where we should reparent all
5537 * memory to prepare the cgroup for destruction. However,
Tejun Heoec903c02014-05-13 12:11:01 -04005538 * memcg does not do css_tryget_online() and res_counter charging
Johannes Weiner96f1c582013-12-12 17:12:34 -08005539 * under the same RCU lock region, which means that charging
5540 * could race with offlining. Offlining only happens to
5541 * cgroups with no tasks in them but charges can show up
5542 * without any tasks from the swapin path when the target
5543 * memcg is looked up from the swapout record and not from the
5544 * current task as it usually is. A race like this can leak
5545 * charges and put pages with stale cgroup pointers into
5546 * circulation:
5547 *
5548 * #0 #1
5549 * lookup_swap_cgroup_id()
5550 * rcu_read_lock()
5551 * mem_cgroup_lookup()
Tejun Heoec903c02014-05-13 12:11:01 -04005552 * css_tryget_online()
Johannes Weiner96f1c582013-12-12 17:12:34 -08005553 * rcu_read_unlock()
Tejun Heoec903c02014-05-13 12:11:01 -04005554 * disable css_tryget_online()
Johannes Weiner96f1c582013-12-12 17:12:34 -08005555 * call_rcu()
5556 * offline_css()
5557 * reparent_charges()
5558 * res_counter_charge()
5559 * css_put()
5560 * css_free()
5561 * pc->mem_cgroup = dead memcg
5562 * add page to lru
5563 *
5564 * The bulk of the charges are still moved in offline_css() to
5565 * avoid pinning a lot of pages in case a long-term reference
5566 * like a swapout record is deferring the css_free() to long
5567 * after offlining. But this makes sure we catch any charges
5568 * made after offlining:
5569 */
5570 mem_cgroup_reparent_charges(memcg);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005571
Li Zefan10d5ebf2013-07-08 16:00:33 -07005572 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07005573 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005574}
5575
Tejun Heo1ced9532014-07-08 18:02:57 -04005576/**
5577 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5578 * @css: the target css
5579 *
5580 * Reset the states of the mem_cgroup associated with @css. This is
5581 * invoked when the userland requests disabling on the default hierarchy
5582 * but the memcg is pinned through dependency. The memcg should stop
5583 * applying policies and should revert to the vanilla state as it may be
5584 * made visible again.
5585 *
5586 * The current implementation only resets the essential configurations.
5587 * This needs to be expanded to cover all the visible parts.
5588 */
5589static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
5590{
5591 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5592
5593 mem_cgroup_resize_limit(memcg, ULLONG_MAX);
5594 mem_cgroup_resize_memsw_limit(memcg, ULLONG_MAX);
5595 memcg_update_kmem_limit(memcg, ULLONG_MAX);
5596 res_counter_set_soft_limit(&memcg->res, ULLONG_MAX);
5597}
5598
Daisuke Nishimura02491442010-03-10 15:22:17 -08005599#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005600/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005601static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005602{
Johannes Weiner05b84302014-08-06 16:05:59 -07005603 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07005604
5605 /* Try a single bulk charge without reclaim first */
Johannes Weiner00501b52014-08-08 14:19:20 -07005606 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07005607 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005608 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005609 return ret;
5610 }
Johannes Weiner692e7c42014-08-06 16:05:57 -07005611 if (ret == -EINTR) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005612 cancel_charge(root_mem_cgroup, count);
Johannes Weiner692e7c42014-08-06 16:05:57 -07005613 return ret;
5614 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005615
5616 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005617 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005618 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
Johannes Weiner9476db92014-08-06 16:05:55 -07005619 /*
5620 * In case of failure, any residual charges against
5621 * mc.to will be dropped by mem_cgroup_clear_mc()
Johannes Weiner692e7c42014-08-06 16:05:57 -07005622 * later on. However, cancel any charges that are
5623 * bypassed to root right away or they'll be lost.
Johannes Weiner9476db92014-08-06 16:05:55 -07005624 */
Johannes Weiner692e7c42014-08-06 16:05:57 -07005625 if (ret == -EINTR)
Johannes Weiner00501b52014-08-08 14:19:20 -07005626 cancel_charge(root_mem_cgroup, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005627 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005628 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005629 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07005630 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005631 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005632 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005633}
5634
5635/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005636 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005637 * @vma: the vma the pte to be checked belongs
5638 * @addr: the address corresponding to the pte to be checked
5639 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005640 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005641 *
5642 * Returns
5643 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5644 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5645 * move charge. if @target is not NULL, the page is stored in target->page
5646 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005647 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5648 * target for charge migration. if @target is not NULL, the entry is stored
5649 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005650 *
5651 * Called with pte lock held.
5652 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005653union mc_target {
5654 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005655 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005656};
5657
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005658enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005659 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005660 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005661 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005662};
5663
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005664static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5665 unsigned long addr, pte_t ptent)
5666{
5667 struct page *page = vm_normal_page(vma, addr, ptent);
5668
5669 if (!page || !page_mapped(page))
5670 return NULL;
5671 if (PageAnon(page)) {
5672 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005673 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005674 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005675 } else if (!move_file())
5676 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005677 return NULL;
5678 if (!get_page_unless_zero(page))
5679 return NULL;
5680
5681 return page;
5682}
5683
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005684#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005685static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5686 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5687{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005688 struct page *page = NULL;
5689 swp_entry_t ent = pte_to_swp_entry(ptent);
5690
5691 if (!move_anon() || non_swap_entry(ent))
5692 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005693 /*
5694 * Because lookup_swap_cache() updates some statistics counter,
5695 * we call find_get_page() with swapper_space directly.
5696 */
Shaohua Li33806f02013-02-22 16:34:37 -08005697 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005698 if (do_swap_account)
5699 entry->val = ent.val;
5700
5701 return page;
5702}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005703#else
5704static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5705 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5706{
5707 return NULL;
5708}
5709#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005710
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005711static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5712 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5713{
5714 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005715 struct address_space *mapping;
5716 pgoff_t pgoff;
5717
5718 if (!vma->vm_file) /* anonymous vma */
5719 return NULL;
5720 if (!move_file())
5721 return NULL;
5722
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005723 mapping = vma->vm_file->f_mapping;
5724 if (pte_none(ptent))
5725 pgoff = linear_page_index(vma, addr);
5726 else /* pte_file(ptent) is true */
5727 pgoff = pte_to_pgoff(ptent);
5728
5729 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005730#ifdef CONFIG_SWAP
5731 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07005732 if (shmem_mapping(mapping)) {
5733 page = find_get_entry(mapping, pgoff);
5734 if (radix_tree_exceptional_entry(page)) {
5735 swp_entry_t swp = radix_to_swp_entry(page);
5736 if (do_swap_account)
5737 *entry = swp;
5738 page = find_get_page(swap_address_space(swp), swp.val);
5739 }
5740 } else
5741 page = find_get_page(mapping, pgoff);
5742#else
5743 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005744#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005745 return page;
5746}
5747
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005748static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005749 unsigned long addr, pte_t ptent, union mc_target *target)
5750{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005751 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005752 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005753 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005754 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005755
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005756 if (pte_present(ptent))
5757 page = mc_handle_present_pte(vma, addr, ptent);
5758 else if (is_swap_pte(ptent))
5759 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005760 else if (pte_none(ptent) || pte_file(ptent))
5761 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005762
5763 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005764 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005765 if (page) {
5766 pc = lookup_page_cgroup(page);
5767 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005768 * Do only loose check w/o serialization.
5769 * mem_cgroup_move_account() checks the pc is valid or
5770 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08005771 */
5772 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5773 ret = MC_TARGET_PAGE;
5774 if (target)
5775 target->page = page;
5776 }
5777 if (!ret || !target)
5778 put_page(page);
5779 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005780 /* There is a swap entry and a page doesn't exist or isn't charged */
5781 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08005782 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005783 ret = MC_TARGET_SWAP;
5784 if (target)
5785 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005786 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005787 return ret;
5788}
5789
Naoya Horiguchi12724852012-03-21 16:34:28 -07005790#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5791/*
5792 * We don't consider swapping or file mapped pages because THP does not
5793 * support them for now.
5794 * Caller should make sure that pmd_trans_huge(pmd) is true.
5795 */
5796static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5797 unsigned long addr, pmd_t pmd, union mc_target *target)
5798{
5799 struct page *page = NULL;
5800 struct page_cgroup *pc;
5801 enum mc_target_type ret = MC_TARGET_NONE;
5802
5803 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08005804 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07005805 if (!move_anon())
5806 return ret;
5807 pc = lookup_page_cgroup(page);
5808 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5809 ret = MC_TARGET_PAGE;
5810 if (target) {
5811 get_page(page);
5812 target->page = page;
5813 }
5814 }
5815 return ret;
5816}
5817#else
5818static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5819 unsigned long addr, pmd_t pmd, union mc_target *target)
5820{
5821 return MC_TARGET_NONE;
5822}
5823#endif
5824
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005825static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5826 unsigned long addr, unsigned long end,
5827 struct mm_walk *walk)
5828{
5829 struct vm_area_struct *vma = walk->private;
5830 pte_t *pte;
5831 spinlock_t *ptl;
5832
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005833 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005834 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5835 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005836 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005837 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005838 }
Dave Hansen03319322011-03-22 16:32:56 -07005839
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005840 if (pmd_trans_unstable(pmd))
5841 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005842 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5843 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005844 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005845 mc.precharge++; /* increment precharge temporarily */
5846 pte_unmap_unlock(pte - 1, ptl);
5847 cond_resched();
5848
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005849 return 0;
5850}
5851
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005852static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5853{
5854 unsigned long precharge;
5855 struct vm_area_struct *vma;
5856
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005857 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005858 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5859 struct mm_walk mem_cgroup_count_precharge_walk = {
5860 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5861 .mm = mm,
5862 .private = vma,
5863 };
5864 if (is_vm_hugetlb_page(vma))
5865 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005866 walk_page_range(vma->vm_start, vma->vm_end,
5867 &mem_cgroup_count_precharge_walk);
5868 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005869 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005870
5871 precharge = mc.precharge;
5872 mc.precharge = 0;
5873
5874 return precharge;
5875}
5876
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005877static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5878{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005879 unsigned long precharge = mem_cgroup_count_precharge(mm);
5880
5881 VM_BUG_ON(mc.moving_task);
5882 mc.moving_task = current;
5883 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005884}
5885
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005886/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5887static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005888{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005889 struct mem_cgroup *from = mc.from;
5890 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07005891 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005892
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005893 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005894 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005895 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005896 mc.precharge = 0;
5897 }
5898 /*
5899 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5900 * we must uncharge here.
5901 */
5902 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005903 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005904 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005905 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005906 /* we must fixup refcnts and charges */
5907 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005908 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04005909 if (!mem_cgroup_is_root(mc.from))
5910 res_counter_uncharge(&mc.from->memsw,
5911 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07005912
5913 for (i = 0; i < mc.moved_swap; i++)
5914 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005915
Johannes Weiner05b84302014-08-06 16:05:59 -07005916 /*
5917 * we charged both to->res and to->memsw, so we should
5918 * uncharge to->res.
5919 */
Johannes Weinerce00a962014-09-05 08:43:57 -04005920 if (!mem_cgroup_is_root(mc.to))
5921 res_counter_uncharge(&mc.to->res,
5922 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07005923 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005924 mc.moved_swap = 0;
5925 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005926 memcg_oom_recover(from);
5927 memcg_oom_recover(to);
5928 wake_up_all(&mc.waitq);
5929}
5930
5931static void mem_cgroup_clear_mc(void)
5932{
5933 struct mem_cgroup *from = mc.from;
5934
5935 /*
5936 * we must clear moving_task before waking up waiters at the end of
5937 * task migration.
5938 */
5939 mc.moving_task = NULL;
5940 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005941 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005942 mc.from = NULL;
5943 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005944 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005945 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005946}
5947
Tejun Heoeb954192013-08-08 20:11:23 -04005948static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005949 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005950{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005951 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005952 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04005953 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08005954 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005955
Glauber Costaee5e8472013-02-22 16:34:50 -08005956 /*
5957 * We are now commited to this value whatever it is. Changes in this
5958 * tunable will only affect upcoming migrations, not the current one.
5959 * So we need to save it, and keep it going.
5960 */
5961 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
5962 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005963 struct mm_struct *mm;
5964 struct mem_cgroup *from = mem_cgroup_from_task(p);
5965
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005966 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005967
5968 mm = get_task_mm(p);
5969 if (!mm)
5970 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005971 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005972 if (mm->owner == p) {
5973 VM_BUG_ON(mc.from);
5974 VM_BUG_ON(mc.to);
5975 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005976 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005977 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005978 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005979 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005980 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005981 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08005982 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005983 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005984 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005985
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005986 ret = mem_cgroup_precharge_mc(mm);
5987 if (ret)
5988 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005989 }
5990 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005991 }
5992 return ret;
5993}
5994
Tejun Heoeb954192013-08-08 20:11:23 -04005995static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005996 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005997{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005998 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005999}
6000
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006001static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6002 unsigned long addr, unsigned long end,
6003 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006004{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006005 int ret = 0;
6006 struct vm_area_struct *vma = walk->private;
6007 pte_t *pte;
6008 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006009 enum mc_target_type target_type;
6010 union mc_target target;
6011 struct page *page;
6012 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006013
Naoya Horiguchi12724852012-03-21 16:34:28 -07006014 /*
6015 * We don't take compound_lock() here but no race with splitting thp
6016 * happens because:
6017 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6018 * under splitting, which means there's no concurrent thp split,
6019 * - if another thread runs into split_huge_page() just after we
6020 * entered this if-block, the thread must wait for page table lock
6021 * to be unlocked in __split_huge_page_splitting(), where the main
6022 * part of thp split is not executed yet.
6023 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006024 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006025 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006026 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006027 return 0;
6028 }
6029 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6030 if (target_type == MC_TARGET_PAGE) {
6031 page = target.page;
6032 if (!isolate_lru_page(page)) {
6033 pc = lookup_page_cgroup(page);
6034 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006035 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006036 mc.precharge -= HPAGE_PMD_NR;
6037 mc.moved_charge += HPAGE_PMD_NR;
6038 }
6039 putback_lru_page(page);
6040 }
6041 put_page(page);
6042 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006043 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006044 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006045 }
6046
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006047 if (pmd_trans_unstable(pmd))
6048 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006049retry:
6050 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6051 for (; addr != end; addr += PAGE_SIZE) {
6052 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006053 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006054
6055 if (!mc.precharge)
6056 break;
6057
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006058 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006059 case MC_TARGET_PAGE:
6060 page = target.page;
6061 if (isolate_lru_page(page))
6062 goto put;
6063 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006064 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006065 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006066 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006067 /* we uncharge from mc.from later. */
6068 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006069 }
6070 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006071put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006072 put_page(page);
6073 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006074 case MC_TARGET_SWAP:
6075 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006076 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006077 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006078 /* we fixup refcnts and charges later. */
6079 mc.moved_swap++;
6080 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006081 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006082 default:
6083 break;
6084 }
6085 }
6086 pte_unmap_unlock(pte - 1, ptl);
6087 cond_resched();
6088
6089 if (addr != end) {
6090 /*
6091 * We have consumed all precharges we got in can_attach().
6092 * We try charge one by one, but don't do any additional
6093 * charges to mc.to if we have failed in charge once in attach()
6094 * phase.
6095 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006096 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006097 if (!ret)
6098 goto retry;
6099 }
6100
6101 return ret;
6102}
6103
6104static void mem_cgroup_move_charge(struct mm_struct *mm)
6105{
6106 struct vm_area_struct *vma;
6107
6108 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006109retry:
6110 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6111 /*
6112 * Someone who are holding the mmap_sem might be waiting in
6113 * waitq. So we cancel all extra charges, wake up all waiters,
6114 * and retry. Because we cancel precharges, we might not be able
6115 * to move enough charges, but moving charge is a best-effort
6116 * feature anyway, so it wouldn't be a big problem.
6117 */
6118 __mem_cgroup_clear_mc();
6119 cond_resched();
6120 goto retry;
6121 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006122 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6123 int ret;
6124 struct mm_walk mem_cgroup_move_charge_walk = {
6125 .pmd_entry = mem_cgroup_move_charge_pte_range,
6126 .mm = mm,
6127 .private = vma,
6128 };
6129 if (is_vm_hugetlb_page(vma))
6130 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006131 ret = walk_page_range(vma->vm_start, vma->vm_end,
6132 &mem_cgroup_move_charge_walk);
6133 if (ret)
6134 /*
6135 * means we have consumed all precharges and failed in
6136 * doing additional charge. Just abandon here.
6137 */
6138 break;
6139 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006140 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006141}
6142
Tejun Heoeb954192013-08-08 20:11:23 -04006143static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006144 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006145{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006146 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006147 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006148
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006149 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006150 if (mc.to)
6151 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006152 mmput(mm);
6153 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006154 if (mc.to)
6155 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006156}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006157#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04006158static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006159 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006160{
6161 return 0;
6162}
Tejun Heoeb954192013-08-08 20:11:23 -04006163static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006164 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006165{
6166}
Tejun Heoeb954192013-08-08 20:11:23 -04006167static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006168 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006169{
6170}
6171#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006172
Tejun Heof00baae2013-04-15 13:41:15 -07006173/*
6174 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04006175 * to verify whether we're attached to the default hierarchy on each mount
6176 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07006177 */
Tejun Heoeb954192013-08-08 20:11:23 -04006178static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006179{
6180 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04006181 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07006182 * guarantees that @root doesn't have any children, so turning it
6183 * on for the root memcg is enough.
6184 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04006185 if (cgroup_on_dfl(root_css->cgroup))
Tejun Heoeb954192013-08-08 20:11:23 -04006186 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07006187}
6188
Tejun Heo073219e2014-02-08 10:36:58 -05006189struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006190 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006191 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006192 .css_offline = mem_cgroup_css_offline,
6193 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04006194 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006195 .can_attach = mem_cgroup_can_attach,
6196 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08006197 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006198 .bind = mem_cgroup_bind,
Tejun Heo55779642014-07-15 11:05:09 -04006199 .legacy_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006200 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006201};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006202
Andrew Mortonc255a452012-07-31 16:43:02 -07006203#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08006204static int __init enable_swap_account(char *s)
6205{
Michal Hockoa2c89902011-05-24 17:12:50 -07006206 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006207 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07006208 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006209 really_do_swap_account = 0;
6210 return 1;
6211}
Michal Hockoa2c89902011-05-24 17:12:50 -07006212__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006213
Michal Hocko2d110852013-02-22 16:34:43 -08006214static void __init memsw_file_init(void)
6215{
Tejun Heo2cf669a2014-07-15 11:05:09 -04006216 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6217 memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08006218}
Michal Hocko6acc8b02013-02-22 16:34:45 -08006219
6220static void __init enable_swap_cgroup(void)
6221{
6222 if (!mem_cgroup_disabled() && really_do_swap_account) {
6223 do_swap_account = 1;
6224 memsw_file_init();
6225 }
6226}
6227
Michal Hocko2d110852013-02-22 16:34:43 -08006228#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08006229static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08006230{
6231}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006232#endif
Michal Hocko2d110852013-02-22 16:34:43 -08006233
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006234#ifdef CONFIG_MEMCG_SWAP
6235/**
6236 * mem_cgroup_swapout - transfer a memsw charge to swap
6237 * @page: page whose memsw charge to transfer
6238 * @entry: swap entry to move the charge to
6239 *
6240 * Transfer the memsw charge of @page to @entry.
6241 */
6242void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
6243{
6244 struct page_cgroup *pc;
6245 unsigned short oldid;
6246
6247 VM_BUG_ON_PAGE(PageLRU(page), page);
6248 VM_BUG_ON_PAGE(page_count(page), page);
6249
6250 if (!do_swap_account)
6251 return;
6252
6253 pc = lookup_page_cgroup(page);
6254
6255 /* Readahead page, never charged */
6256 if (!PageCgroupUsed(pc))
6257 return;
6258
6259 VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page);
6260
6261 oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup));
6262 VM_BUG_ON_PAGE(oldid, page);
6263
6264 pc->flags &= ~PCG_MEMSW;
6265 css_get(&pc->mem_cgroup->css);
6266 mem_cgroup_swap_statistics(pc->mem_cgroup, true);
6267}
6268
6269/**
6270 * mem_cgroup_uncharge_swap - uncharge a swap entry
6271 * @entry: swap entry to uncharge
6272 *
6273 * Drop the memsw charge associated with @entry.
6274 */
6275void mem_cgroup_uncharge_swap(swp_entry_t entry)
6276{
6277 struct mem_cgroup *memcg;
6278 unsigned short id;
6279
6280 if (!do_swap_account)
6281 return;
6282
6283 id = swap_cgroup_record(entry, 0);
6284 rcu_read_lock();
6285 memcg = mem_cgroup_lookup(id);
6286 if (memcg) {
Johannes Weinerce00a962014-09-05 08:43:57 -04006287 if (!mem_cgroup_is_root(memcg))
6288 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006289 mem_cgroup_swap_statistics(memcg, false);
6290 css_put(&memcg->css);
6291 }
6292 rcu_read_unlock();
6293}
6294#endif
6295
Johannes Weiner00501b52014-08-08 14:19:20 -07006296/**
6297 * mem_cgroup_try_charge - try charging a page
6298 * @page: page to charge
6299 * @mm: mm context of the victim
6300 * @gfp_mask: reclaim mode
6301 * @memcgp: charged memcg return
6302 *
6303 * Try to charge @page to the memcg that @mm belongs to, reclaiming
6304 * pages according to @gfp_mask if necessary.
6305 *
6306 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
6307 * Otherwise, an error code is returned.
6308 *
6309 * After page->mapping has been set up, the caller must finalize the
6310 * charge with mem_cgroup_commit_charge(). Or abort the transaction
6311 * with mem_cgroup_cancel_charge() in case page instantiation fails.
6312 */
6313int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
6314 gfp_t gfp_mask, struct mem_cgroup **memcgp)
6315{
6316 struct mem_cgroup *memcg = NULL;
6317 unsigned int nr_pages = 1;
6318 int ret = 0;
6319
6320 if (mem_cgroup_disabled())
6321 goto out;
6322
6323 if (PageSwapCache(page)) {
6324 struct page_cgroup *pc = lookup_page_cgroup(page);
6325 /*
6326 * Every swap fault against a single page tries to charge the
6327 * page, bail as early as possible. shmem_unuse() encounters
6328 * already charged pages, too. The USED bit is protected by
6329 * the page lock, which serializes swap cache removal, which
6330 * in turn serializes uncharging.
6331 */
6332 if (PageCgroupUsed(pc))
6333 goto out;
6334 }
6335
6336 if (PageTransHuge(page)) {
6337 nr_pages <<= compound_order(page);
6338 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
6339 }
6340
6341 if (do_swap_account && PageSwapCache(page))
6342 memcg = try_get_mem_cgroup_from_page(page);
6343 if (!memcg)
6344 memcg = get_mem_cgroup_from_mm(mm);
6345
6346 ret = try_charge(memcg, gfp_mask, nr_pages);
6347
6348 css_put(&memcg->css);
6349
6350 if (ret == -EINTR) {
6351 memcg = root_mem_cgroup;
6352 ret = 0;
6353 }
6354out:
6355 *memcgp = memcg;
6356 return ret;
6357}
6358
6359/**
6360 * mem_cgroup_commit_charge - commit a page charge
6361 * @page: page to charge
6362 * @memcg: memcg to charge the page to
6363 * @lrucare: page might be on LRU already
6364 *
6365 * Finalize a charge transaction started by mem_cgroup_try_charge(),
6366 * after page->mapping has been set up. This must happen atomically
6367 * as part of the page instantiation, i.e. under the page table lock
6368 * for anonymous pages, under the page lock for page and swap cache.
6369 *
6370 * In addition, the page must not be on the LRU during the commit, to
6371 * prevent racing with task migration. If it might be, use @lrucare.
6372 *
6373 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
6374 */
6375void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
6376 bool lrucare)
6377{
6378 unsigned int nr_pages = 1;
6379
6380 VM_BUG_ON_PAGE(!page->mapping, page);
6381 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
6382
6383 if (mem_cgroup_disabled())
6384 return;
6385 /*
6386 * Swap faults will attempt to charge the same page multiple
6387 * times. But reuse_swap_page() might have removed the page
6388 * from swapcache already, so we can't check PageSwapCache().
6389 */
6390 if (!memcg)
6391 return;
6392
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006393 commit_charge(page, memcg, lrucare);
6394
Johannes Weiner00501b52014-08-08 14:19:20 -07006395 if (PageTransHuge(page)) {
6396 nr_pages <<= compound_order(page);
6397 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
6398 }
6399
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006400 local_irq_disable();
6401 mem_cgroup_charge_statistics(memcg, page, nr_pages);
6402 memcg_check_events(memcg, page);
6403 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07006404
6405 if (do_swap_account && PageSwapCache(page)) {
6406 swp_entry_t entry = { .val = page_private(page) };
6407 /*
6408 * The swap entry might not get freed for a long time,
6409 * let's not wait for it. The page already received a
6410 * memory+swap charge, drop the swap entry duplicate.
6411 */
6412 mem_cgroup_uncharge_swap(entry);
6413 }
6414}
6415
6416/**
6417 * mem_cgroup_cancel_charge - cancel a page charge
6418 * @page: page to charge
6419 * @memcg: memcg to charge the page to
6420 *
6421 * Cancel a charge transaction started by mem_cgroup_try_charge().
6422 */
6423void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
6424{
6425 unsigned int nr_pages = 1;
6426
6427 if (mem_cgroup_disabled())
6428 return;
6429 /*
6430 * Swap faults will attempt to charge the same page multiple
6431 * times. But reuse_swap_page() might have removed the page
6432 * from swapcache already, so we can't check PageSwapCache().
6433 */
6434 if (!memcg)
6435 return;
6436
6437 if (PageTransHuge(page)) {
6438 nr_pages <<= compound_order(page);
6439 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
6440 }
6441
6442 cancel_charge(memcg, nr_pages);
6443}
6444
Johannes Weiner747db952014-08-08 14:19:24 -07006445static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
6446 unsigned long nr_mem, unsigned long nr_memsw,
6447 unsigned long nr_anon, unsigned long nr_file,
6448 unsigned long nr_huge, struct page *dummy_page)
6449{
6450 unsigned long flags;
6451
Johannes Weinerce00a962014-09-05 08:43:57 -04006452 if (!mem_cgroup_is_root(memcg)) {
6453 if (nr_mem)
6454 res_counter_uncharge(&memcg->res,
6455 nr_mem * PAGE_SIZE);
6456 if (nr_memsw)
6457 res_counter_uncharge(&memcg->memsw,
6458 nr_memsw * PAGE_SIZE);
6459 memcg_oom_recover(memcg);
6460 }
Johannes Weiner747db952014-08-08 14:19:24 -07006461
6462 local_irq_save(flags);
6463 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
6464 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
6465 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
6466 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
6467 __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file);
6468 memcg_check_events(memcg, dummy_page);
6469 local_irq_restore(flags);
6470}
6471
6472static void uncharge_list(struct list_head *page_list)
6473{
6474 struct mem_cgroup *memcg = NULL;
6475 unsigned long nr_memsw = 0;
6476 unsigned long nr_anon = 0;
6477 unsigned long nr_file = 0;
6478 unsigned long nr_huge = 0;
6479 unsigned long pgpgout = 0;
6480 unsigned long nr_mem = 0;
6481 struct list_head *next;
6482 struct page *page;
6483
6484 next = page_list->next;
6485 do {
6486 unsigned int nr_pages = 1;
6487 struct page_cgroup *pc;
6488
6489 page = list_entry(next, struct page, lru);
6490 next = page->lru.next;
6491
6492 VM_BUG_ON_PAGE(PageLRU(page), page);
6493 VM_BUG_ON_PAGE(page_count(page), page);
6494
6495 pc = lookup_page_cgroup(page);
6496 if (!PageCgroupUsed(pc))
6497 continue;
6498
6499 /*
6500 * Nobody should be changing or seriously looking at
6501 * pc->mem_cgroup and pc->flags at this point, we have
6502 * fully exclusive access to the page.
6503 */
6504
6505 if (memcg != pc->mem_cgroup) {
6506 if (memcg) {
6507 uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw,
6508 nr_anon, nr_file, nr_huge, page);
6509 pgpgout = nr_mem = nr_memsw = 0;
6510 nr_anon = nr_file = nr_huge = 0;
6511 }
6512 memcg = pc->mem_cgroup;
6513 }
6514
6515 if (PageTransHuge(page)) {
6516 nr_pages <<= compound_order(page);
6517 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
6518 nr_huge += nr_pages;
6519 }
6520
6521 if (PageAnon(page))
6522 nr_anon += nr_pages;
6523 else
6524 nr_file += nr_pages;
6525
6526 if (pc->flags & PCG_MEM)
6527 nr_mem += nr_pages;
6528 if (pc->flags & PCG_MEMSW)
6529 nr_memsw += nr_pages;
6530 pc->flags = 0;
6531
6532 pgpgout++;
6533 } while (next != page_list);
6534
6535 if (memcg)
6536 uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw,
6537 nr_anon, nr_file, nr_huge, page);
6538}
6539
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006540/**
6541 * mem_cgroup_uncharge - uncharge a page
6542 * @page: page to uncharge
6543 *
6544 * Uncharge a page previously charged with mem_cgroup_try_charge() and
6545 * mem_cgroup_commit_charge().
6546 */
6547void mem_cgroup_uncharge(struct page *page)
6548{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006549 struct page_cgroup *pc;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006550
6551 if (mem_cgroup_disabled())
6552 return;
6553
Johannes Weiner747db952014-08-08 14:19:24 -07006554 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006555 pc = lookup_page_cgroup(page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006556 if (!PageCgroupUsed(pc))
6557 return;
6558
Johannes Weiner747db952014-08-08 14:19:24 -07006559 INIT_LIST_HEAD(&page->lru);
6560 uncharge_list(&page->lru);
6561}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006562
Johannes Weiner747db952014-08-08 14:19:24 -07006563/**
6564 * mem_cgroup_uncharge_list - uncharge a list of page
6565 * @page_list: list of pages to uncharge
6566 *
6567 * Uncharge a list of pages previously charged with
6568 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
6569 */
6570void mem_cgroup_uncharge_list(struct list_head *page_list)
6571{
6572 if (mem_cgroup_disabled())
6573 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006574
Johannes Weiner747db952014-08-08 14:19:24 -07006575 if (!list_empty(page_list))
6576 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006577}
6578
6579/**
6580 * mem_cgroup_migrate - migrate a charge to another page
6581 * @oldpage: currently charged page
6582 * @newpage: page to transfer the charge to
6583 * @lrucare: both pages might be on the LRU already
6584 *
6585 * Migrate the charge from @oldpage to @newpage.
6586 *
6587 * Both pages must be locked, @newpage->mapping must be set up.
6588 */
6589void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
6590 bool lrucare)
6591{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006592 struct page_cgroup *pc;
6593 int isolated;
6594
6595 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
6596 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
6597 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
6598 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
6599 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006600 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
6601 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006602
6603 if (mem_cgroup_disabled())
6604 return;
6605
6606 /* Page cache replacement: new page already charged? */
6607 pc = lookup_page_cgroup(newpage);
6608 if (PageCgroupUsed(pc))
6609 return;
6610
6611 /* Re-entrant migration: old page already uncharged? */
6612 pc = lookup_page_cgroup(oldpage);
6613 if (!PageCgroupUsed(pc))
6614 return;
6615
6616 VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage);
6617 VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage);
6618
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006619 if (lrucare)
6620 lock_page_lru(oldpage, &isolated);
6621
6622 pc->flags = 0;
6623
6624 if (lrucare)
6625 unlock_page_lru(oldpage, isolated);
6626
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006627 commit_charge(newpage, pc->mem_cgroup, lrucare);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006628}
6629
Michal Hocko2d110852013-02-22 16:34:43 -08006630/*
Michal Hocko10813122013-02-22 16:35:41 -08006631 * subsys_initcall() for memory controller.
6632 *
6633 * Some parts like hotcpu_notifier() have to be initialized from this context
6634 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
6635 * everything that doesn't depend on a specific mem_cgroup structure should
6636 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08006637 */
6638static int __init mem_cgroup_init(void)
6639{
6640 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08006641 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006642 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08006643 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08006644 return 0;
6645}
6646subsys_initcall(mem_cgroup_init);