blob: 6c32271a31c5ca314337e9f592710017a1415048 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080017 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
21 *
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
26 */
27
28#include <linux/res_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080031#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080032#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080033#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
48#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080049#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080050#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070051#include <linux/vmalloc.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070054#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080057#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000058#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070059#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000060#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080061
Balbir Singh8697d332008-02-07 00:13:59 -080062#include <asm/uaccess.h>
63
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070064#include <trace/events/vmscan.h>
65
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070066struct cgroup_subsys mem_cgroup_subsys __read_mostly;
David Rientjes68ae5642012-12-12 13:51:57 -080067EXPORT_SYMBOL(mem_cgroup_subsys);
68
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070069#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070070static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080071
Andrew Mortonc255a452012-07-31 16:43:02 -070072#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070073/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080074int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080075
76/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070077#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080078static int really_do_swap_account __initdata = 1;
79#else
80static int really_do_swap_account __initdata = 0;
81#endif
82
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080083#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070084#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#endif
86
87
Balbir Singh8cdea7c2008-02-07 00:13:50 -080088/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080089 * Statistics for memory cgroup.
90 */
91enum mem_cgroup_stat_index {
92 /*
93 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
94 */
David Rientjesb070e652013-05-07 16:18:09 -070095 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
96 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
97 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
98 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
99 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800100 MEM_CGROUP_STAT_NSTATS,
101};
102
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700103static const char * const mem_cgroup_stat_names[] = {
104 "cache",
105 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700106 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107 "mapped_file",
108 "swap",
109};
110
Johannes Weinere9f89742011-03-23 16:42:37 -0700111enum mem_cgroup_events_index {
112 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
113 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700114 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
115 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700116 MEM_CGROUP_EVENTS_NSTATS,
117};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700118
119static const char * const mem_cgroup_events_names[] = {
120 "pgpgin",
121 "pgpgout",
122 "pgfault",
123 "pgmajfault",
124};
125
Sha Zhengju58cf1882013-02-22 16:32:05 -0800126static const char * const mem_cgroup_lru_names[] = {
127 "inactive_anon",
128 "active_anon",
129 "inactive_file",
130 "active_file",
131 "unevictable",
132};
133
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700134/*
135 * Per memcg event counter is incremented at every pagein/pageout. With THP,
136 * it will be incremated by the number of pages. This counter is used for
137 * for trigger some periodic events. This is straightforward and better
138 * than using jiffies etc. to handle periodic memcg event.
139 */
140enum mem_cgroup_events_target {
141 MEM_CGROUP_TARGET_THRESH,
142 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700143 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700144 MEM_CGROUP_NTARGETS,
145};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700146#define THRESHOLDS_EVENTS_TARGET 128
147#define SOFTLIMIT_EVENTS_TARGET 1024
148#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700149
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800150struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700151 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700152 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700153 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700154 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800155};
156
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800157struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700158 /*
159 * last scanned hierarchy member. Valid only if last_dead_count
160 * matches memcg->dead_count of the hierarchy root group.
161 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700162 struct mem_cgroup *last_visited;
Michal Hocko5f578162013-04-29 15:07:17 -0700163 unsigned long last_dead_count;
164
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800165 /* scan generation, increased every round-trip */
166 unsigned int generation;
167};
168
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800169/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800170 * per-zone information in memory controller.
171 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800172struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800173 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700174 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800175
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800176 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
177
Balbir Singhf64c3f52009-09-23 15:56:37 -0700178 struct rb_node tree_node; /* RB tree node */
179 unsigned long long usage_in_excess;/* Set to the value by which */
180 /* the soft limit is exceeded*/
181 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700182 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700183 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800184};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800185
186struct mem_cgroup_per_node {
187 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
188};
189
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800190/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700191 * Cgroups above their limits are maintained in a RB-Tree, independent of
192 * their hierarchy representation
193 */
194
195struct mem_cgroup_tree_per_zone {
196 struct rb_root rb_root;
197 spinlock_t lock;
198};
199
200struct mem_cgroup_tree_per_node {
201 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
202};
203
204struct mem_cgroup_tree {
205 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
206};
207
208static struct mem_cgroup_tree soft_limit_tree __read_mostly;
209
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800210struct mem_cgroup_threshold {
211 struct eventfd_ctx *eventfd;
212 u64 threshold;
213};
214
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700215/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800216struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700217 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700218 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800219 /* Size of entries[] */
220 unsigned int size;
221 /* Array of thresholds */
222 struct mem_cgroup_threshold entries[0];
223};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700224
225struct mem_cgroup_thresholds {
226 /* Primary thresholds array */
227 struct mem_cgroup_threshold_ary *primary;
228 /*
229 * Spare threshold array.
230 * This is needed to make mem_cgroup_unregister_event() "never fail".
231 * It must be able to store at least primary->size - 1 entries.
232 */
233 struct mem_cgroup_threshold_ary *spare;
234};
235
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700236/* for OOM */
237struct mem_cgroup_eventfd_list {
238 struct list_head list;
239 struct eventfd_ctx *eventfd;
240};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800241
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700242static void mem_cgroup_threshold(struct mem_cgroup *memcg);
243static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800244
Balbir Singhf64c3f52009-09-23 15:56:37 -0700245/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800246 * The memory controller data structure. The memory controller controls both
247 * page cache and RSS per cgroup. We would eventually like to provide
248 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
249 * to help the administrator determine what knobs to tune.
250 *
251 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800252 * we hit the water mark. May be even add a low water mark, such that
253 * no reclaim occurs from a cgroup at it's low water mark, this is
254 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800255 */
256struct mem_cgroup {
257 struct cgroup_subsys_state css;
258 /*
259 * the counter to account for memory usage
260 */
261 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700262
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700263 /* vmpressure notifications */
264 struct vmpressure vmpressure;
265
Li Zefan465939a2013-07-08 16:00:38 -0700266 /*
267 * the counter to account for mem+swap usage.
268 */
269 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700270
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800271 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800272 * the counter to account for kernel memory usage.
273 */
274 struct res_counter kmem;
275 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800276 * Should the accounting and control be hierarchical, per subtree?
277 */
278 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800279 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700280
281 bool oom_lock;
282 atomic_t under_oom;
283
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700284 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700285 /* OOM-Killer disable */
286 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800287
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700288 /* set when res.limit == memsw.limit */
289 bool memsw_is_minimum;
290
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800291 /* protect arrays of thresholds */
292 struct mutex thresholds_lock;
293
294 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700295 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700296
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800297 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700298 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700299
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700300 /* For oom notifier event fd */
301 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700302
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800303 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800304 * Should we move charges of a task when a task is moved into this
305 * mem_cgroup ? And what type of charges should we move ?
306 */
307 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800308 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700309 * set > 0 if pages under this cgroup are moving to other cgroup.
310 */
311 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700312 /* taken only while moving_account > 0 */
313 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700314 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800315 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800316 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700317 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700318 /*
319 * used when a cpu is offlined or other synchronizations
320 * See mem_cgroup_read_stat().
321 */
322 struct mem_cgroup_stat_cpu nocpu_base;
323 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000324
Michal Hocko5f578162013-04-29 15:07:17 -0700325 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700326#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000327 struct tcp_memcontrol tcp_mem;
328#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800329#if defined(CONFIG_MEMCG_KMEM)
330 /* analogous to slab_common's slab_caches list. per-memcg */
331 struct list_head memcg_slab_caches;
332 /* Not a spinlock, we can take a lot of time walking the list */
333 struct mutex slab_caches_mutex;
334 /* Index in the kmem_cache->memcg_params->memcg_caches array */
335 int kmemcg_id;
336#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800337
338 int last_scanned_node;
339#if MAX_NUMNODES > 1
340 nodemask_t scan_nodes;
341 atomic_t numainfo_events;
342 atomic_t numainfo_updating;
343#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700344
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700345 struct mem_cgroup_per_node *nodeinfo[0];
346 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800347};
348
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800349static size_t memcg_size(void)
350{
351 return sizeof(struct mem_cgroup) +
352 nr_node_ids * sizeof(struct mem_cgroup_per_node);
353}
354
Glauber Costa510fc4e2012-12-18 14:21:47 -0800355/* internal only representation about the status of kmem accounting. */
356enum {
357 KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */
Glauber Costaa8964b92012-12-18 14:22:09 -0800358 KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */
Glauber Costa7de37682012-12-18 14:22:07 -0800359 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800360};
361
Glauber Costaa8964b92012-12-18 14:22:09 -0800362/* We account when limit is on, but only after call sites are patched */
363#define KMEM_ACCOUNTED_MASK \
364 ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED))
Glauber Costa510fc4e2012-12-18 14:21:47 -0800365
366#ifdef CONFIG_MEMCG_KMEM
367static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
368{
369 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
370}
Glauber Costa7de37682012-12-18 14:22:07 -0800371
372static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
373{
374 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
375}
376
Glauber Costaa8964b92012-12-18 14:22:09 -0800377static void memcg_kmem_set_activated(struct mem_cgroup *memcg)
378{
379 set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
380}
381
Glauber Costa55007d82012-12-18 14:22:38 -0800382static void memcg_kmem_clear_activated(struct mem_cgroup *memcg)
383{
384 clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
385}
386
Glauber Costa7de37682012-12-18 14:22:07 -0800387static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
388{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700389 /*
390 * Our caller must use css_get() first, because memcg_uncharge_kmem()
391 * will call css_put() if it sees the memcg is dead.
392 */
393 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800394 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
395 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
396}
397
398static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
399{
400 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
401 &memcg->kmem_account_flags);
402}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800403#endif
404
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800405/* Stuffs for move charges at task migration. */
406/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800407 * Types of charges to be moved. "move_charge_at_immitgrate" and
408 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800409 */
410enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800411 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700412 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800413 NR_MOVE_TYPE,
414};
415
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800416/* "mc" and its members are protected by cgroup_mutex */
417static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800418 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800419 struct mem_cgroup *from;
420 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800421 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800422 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800423 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800424 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800425 struct task_struct *moving_task; /* a task moving charges */
426 wait_queue_head_t waitq; /* a waitq for other context */
427} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700428 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800429 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
430};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800431
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700432static bool move_anon(void)
433{
Glauber Costaee5e8472013-02-22 16:34:50 -0800434 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700435}
436
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700437static bool move_file(void)
438{
Glauber Costaee5e8472013-02-22 16:34:50 -0800439 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700440}
441
Balbir Singh4e416952009-09-23 15:56:39 -0700442/*
443 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
444 * limit reclaim to prevent infinite loops, if they ever occur.
445 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700446#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
447#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700448
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800449enum charge_type {
450 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700451 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800452 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700453 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700454 NR_CHARGE_TYPE,
455};
456
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800457/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800458enum res_type {
459 _MEM,
460 _MEMSWAP,
461 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800462 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800463};
464
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700465#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
466#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800467#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700468/* Used for OOM nofiier */
469#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800470
Balbir Singh75822b42009-09-23 15:56:38 -0700471/*
472 * Reclaim flags for mem_cgroup_hierarchical_reclaim
473 */
474#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
475#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
476#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
477#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
478
Glauber Costa09998212013-02-22 16:34:55 -0800479/*
480 * The memcg_create_mutex will be held whenever a new cgroup is created.
481 * As a consequence, any change that needs to protect against new child cgroups
482 * appearing has to hold it as well.
483 */
484static DEFINE_MUTEX(memcg_create_mutex);
485
Wanpeng Lib2145142012-07-31 16:46:01 -0700486struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
487{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400488 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700489}
490
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700491/* Some nice accessors for the vmpressure. */
492struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
493{
494 if (!memcg)
495 memcg = root_mem_cgroup;
496 return &memcg->vmpressure;
497}
498
499struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
500{
501 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
502}
503
504struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
505{
506 return &mem_cgroup_from_css(css)->vmpressure;
507}
508
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700509static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
510{
511 return (memcg == root_mem_cgroup);
512}
513
Glauber Costae1aab162011-12-11 21:47:03 +0000514/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700515#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000516
Glauber Costae1aab162011-12-11 21:47:03 +0000517void sock_update_memcg(struct sock *sk)
518{
Glauber Costa376be5f2012-01-20 04:57:14 +0000519 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000520 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700521 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000522
523 BUG_ON(!sk->sk_prot->proto_cgroup);
524
Glauber Costaf3f511e2012-01-05 20:16:39 +0000525 /* Socket cloning can throw us here with sk_cgrp already
526 * filled. It won't however, necessarily happen from
527 * process context. So the test for root memcg given
528 * the current task's memcg won't help us in this case.
529 *
530 * Respecting the original socket's memcg is a better
531 * decision in this case.
532 */
533 if (sk->sk_cgrp) {
534 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700535 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000536 return;
537 }
538
Glauber Costae1aab162011-12-11 21:47:03 +0000539 rcu_read_lock();
540 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700541 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700542 if (!mem_cgroup_is_root(memcg) &&
543 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700544 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000545 }
546 rcu_read_unlock();
547 }
548}
549EXPORT_SYMBOL(sock_update_memcg);
550
551void sock_release_memcg(struct sock *sk)
552{
Glauber Costa376be5f2012-01-20 04:57:14 +0000553 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000554 struct mem_cgroup *memcg;
555 WARN_ON(!sk->sk_cgrp->memcg);
556 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700557 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000558 }
559}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000560
561struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
562{
563 if (!memcg || mem_cgroup_is_root(memcg))
564 return NULL;
565
566 return &memcg->tcp_mem.cg_proto;
567}
568EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000569
Glauber Costa3f134612012-05-29 15:07:11 -0700570static void disarm_sock_keys(struct mem_cgroup *memcg)
571{
572 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
573 return;
574 static_key_slow_dec(&memcg_socket_limit_enabled);
575}
576#else
577static void disarm_sock_keys(struct mem_cgroup *memcg)
578{
579}
580#endif
581
Glauber Costaa8964b92012-12-18 14:22:09 -0800582#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800583/*
584 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
585 * There are two main reasons for not using the css_id for this:
586 * 1) this works better in sparse environments, where we have a lot of memcgs,
587 * but only a few kmem-limited. Or also, if we have, for instance, 200
588 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
589 * 200 entry array for that.
590 *
591 * 2) In order not to violate the cgroup API, we would like to do all memory
592 * allocation in ->create(). At that point, we haven't yet allocated the
593 * css_id. Having a separate index prevents us from messing with the cgroup
594 * core for this
595 *
596 * The current size of the caches array is stored in
597 * memcg_limited_groups_array_size. It will double each time we have to
598 * increase it.
599 */
600static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800601int memcg_limited_groups_array_size;
602
Glauber Costa55007d82012-12-18 14:22:38 -0800603/*
604 * MIN_SIZE is different than 1, because we would like to avoid going through
605 * the alloc/free process all the time. In a small machine, 4 kmem-limited
606 * cgroups is a reasonable guess. In the future, it could be a parameter or
607 * tunable, but that is strictly not necessary.
608 *
609 * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get
610 * this constant directly from cgroup, but it is understandable that this is
611 * better kept as an internal representation in cgroup.c. In any case, the
612 * css_id space is not getting any smaller, and we don't have to necessarily
613 * increase ours as well if it increases.
614 */
615#define MEMCG_CACHES_MIN_SIZE 4
616#define MEMCG_CACHES_MAX_SIZE 65535
617
Glauber Costad7f25f82012-12-18 14:22:40 -0800618/*
619 * A lot of the calls to the cache allocation functions are expected to be
620 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
621 * conditional to this static branch, we'll have to allow modules that does
622 * kmem_cache_alloc and the such to see this symbol as well
623 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800624struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800625EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800626
627static void disarm_kmem_keys(struct mem_cgroup *memcg)
628{
Glauber Costa55007d82012-12-18 14:22:38 -0800629 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800630 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800631 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
632 }
Glauber Costabea207c2012-12-18 14:22:11 -0800633 /*
634 * This check can't live in kmem destruction function,
635 * since the charges will outlive the cgroup
636 */
637 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800638}
639#else
640static void disarm_kmem_keys(struct mem_cgroup *memcg)
641{
642}
643#endif /* CONFIG_MEMCG_KMEM */
644
645static void disarm_static_keys(struct mem_cgroup *memcg)
646{
647 disarm_sock_keys(memcg);
648 disarm_kmem_keys(memcg);
649}
650
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700651static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800652
Balbir Singhf64c3f52009-09-23 15:56:37 -0700653static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700654mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700655{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800656 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700657 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700658}
659
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700660struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100661{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700662 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100663}
664
Balbir Singhf64c3f52009-09-23 15:56:37 -0700665static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700666page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700667{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700668 int nid = page_to_nid(page);
669 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700670
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700671 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700672}
673
674static struct mem_cgroup_tree_per_zone *
675soft_limit_tree_node_zone(int nid, int zid)
676{
677 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
678}
679
680static struct mem_cgroup_tree_per_zone *
681soft_limit_tree_from_page(struct page *page)
682{
683 int nid = page_to_nid(page);
684 int zid = page_zonenum(page);
685
686 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
687}
688
689static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700690__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700691 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700692 struct mem_cgroup_tree_per_zone *mctz,
693 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700694{
695 struct rb_node **p = &mctz->rb_root.rb_node;
696 struct rb_node *parent = NULL;
697 struct mem_cgroup_per_zone *mz_node;
698
699 if (mz->on_tree)
700 return;
701
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700702 mz->usage_in_excess = new_usage_in_excess;
703 if (!mz->usage_in_excess)
704 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700705 while (*p) {
706 parent = *p;
707 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
708 tree_node);
709 if (mz->usage_in_excess < mz_node->usage_in_excess)
710 p = &(*p)->rb_left;
711 /*
712 * We can't avoid mem cgroups that are over their soft
713 * limit by the same amount
714 */
715 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
716 p = &(*p)->rb_right;
717 }
718 rb_link_node(&mz->tree_node, parent, p);
719 rb_insert_color(&mz->tree_node, &mctz->rb_root);
720 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700721}
722
723static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700724__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700725 struct mem_cgroup_per_zone *mz,
726 struct mem_cgroup_tree_per_zone *mctz)
727{
728 if (!mz->on_tree)
729 return;
730 rb_erase(&mz->tree_node, &mctz->rb_root);
731 mz->on_tree = false;
732}
733
734static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700735mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700736 struct mem_cgroup_per_zone *mz,
737 struct mem_cgroup_tree_per_zone *mctz)
738{
739 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700740 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700741 spin_unlock(&mctz->lock);
742}
743
Balbir Singhf64c3f52009-09-23 15:56:37 -0700744
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700745static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700746{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700747 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700748 struct mem_cgroup_per_zone *mz;
749 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700750 int nid = page_to_nid(page);
751 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700752 mctz = soft_limit_tree_from_page(page);
753
754 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700755 * Necessary to update all ancestors when hierarchy is used.
756 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700757 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700758 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
759 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
760 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700761 /*
762 * We have to update the tree if mz is on RB-tree or
763 * mem is over its softlimit.
764 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700765 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700766 spin_lock(&mctz->lock);
767 /* if on-tree, remove it */
768 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700769 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700770 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700771 * Insert again. mz->usage_in_excess will be updated.
772 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700773 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700774 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700775 spin_unlock(&mctz->lock);
776 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700777 }
778}
779
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700780static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700781{
782 int node, zone;
783 struct mem_cgroup_per_zone *mz;
784 struct mem_cgroup_tree_per_zone *mctz;
785
Bob Liu3ed28fa2012-01-12 17:19:04 -0800786 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700787 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700788 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700789 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700790 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700791 }
792 }
793}
794
Balbir Singh4e416952009-09-23 15:56:39 -0700795static struct mem_cgroup_per_zone *
796__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
797{
798 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700799 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700800
801retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700802 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700803 rightmost = rb_last(&mctz->rb_root);
804 if (!rightmost)
805 goto done; /* Nothing to reclaim from */
806
807 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
808 /*
809 * Remove the node now but someone else can add it back,
810 * we will to add it back at the end of reclaim to its correct
811 * position in the tree.
812 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700813 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
814 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
815 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700816 goto retry;
817done:
818 return mz;
819}
820
821static struct mem_cgroup_per_zone *
822mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
823{
824 struct mem_cgroup_per_zone *mz;
825
826 spin_lock(&mctz->lock);
827 mz = __mem_cgroup_largest_soft_limit_node(mctz);
828 spin_unlock(&mctz->lock);
829 return mz;
830}
831
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700832/*
833 * Implementation Note: reading percpu statistics for memcg.
834 *
835 * Both of vmstat[] and percpu_counter has threshold and do periodic
836 * synchronization to implement "quick" read. There are trade-off between
837 * reading cost and precision of value. Then, we may have a chance to implement
838 * a periodic synchronizion of counter in memcg's counter.
839 *
840 * But this _read() function is used for user interface now. The user accounts
841 * memory usage by memory cgroup and he _always_ requires exact value because
842 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
843 * have to visit all online cpus and make sum. So, for now, unnecessary
844 * synchronization is not implemented. (just implemented for cpu hotplug)
845 *
846 * If there are kernel internal actions which can make use of some not-exact
847 * value, and reading all cpu value can be performance bottleneck in some
848 * common workload, threashold and synchonization as vmstat[] should be
849 * implemented.
850 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700851static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700852 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800853{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700854 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800855 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800856
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700857 get_online_cpus();
858 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700859 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700860#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700861 spin_lock(&memcg->pcp_counter_lock);
862 val += memcg->nocpu_base.count[idx];
863 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700864#endif
865 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800866 return val;
867}
868
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700869static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700870 bool charge)
871{
872 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700873 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700874}
875
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700876static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700877 enum mem_cgroup_events_index idx)
878{
879 unsigned long val = 0;
880 int cpu;
881
882 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700883 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700884#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700885 spin_lock(&memcg->pcp_counter_lock);
886 val += memcg->nocpu_base.events[idx];
887 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700888#endif
889 return val;
890}
891
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700892static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700893 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700894 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800895{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800896 preempt_disable();
897
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700898 /*
899 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
900 * counted as CACHE even if it's on ANON LRU.
901 */
902 if (anon)
903 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700904 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800905 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700906 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700907 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700908
David Rientjesb070e652013-05-07 16:18:09 -0700909 if (PageTransHuge(page))
910 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
911 nr_pages);
912
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800913 /* pagein of a big page is an event. So, ignore page size */
914 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700915 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800916 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700917 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800918 nr_pages = -nr_pages; /* for event */
919 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800920
Johannes Weiner13114712012-05-29 15:07:07 -0700921 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800922
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800923 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800924}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800925
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700926unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700927mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700928{
929 struct mem_cgroup_per_zone *mz;
930
931 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
932 return mz->lru_size[lru];
933}
934
935static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700936mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700937 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700938{
939 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab932012-03-21 16:34:19 -0700940 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700941 unsigned long ret = 0;
942
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700943 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700944
Hugh Dickinsf156ab932012-03-21 16:34:19 -0700945 for_each_lru(lru) {
946 if (BIT(lru) & lru_mask)
947 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700948 }
949 return ret;
950}
951
952static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700953mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700954 int nid, unsigned int lru_mask)
955{
Ying Han889976d2011-05-26 16:25:33 -0700956 u64 total = 0;
957 int zid;
958
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700959 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700960 total += mem_cgroup_zone_nr_lru_pages(memcg,
961 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700962
Ying Han889976d2011-05-26 16:25:33 -0700963 return total;
964}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700965
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700966static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700967 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800968{
Ying Han889976d2011-05-26 16:25:33 -0700969 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800970 u64 total = 0;
971
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800972 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700973 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800974 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800975}
976
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800977static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
978 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800979{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700980 unsigned long val, next;
981
Johannes Weiner13114712012-05-29 15:07:07 -0700982 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700983 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700984 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800985 if ((long)next - (long)val < 0) {
986 switch (target) {
987 case MEM_CGROUP_TARGET_THRESH:
988 next = val + THRESHOLDS_EVENTS_TARGET;
989 break;
990 case MEM_CGROUP_TARGET_SOFTLIMIT:
991 next = val + SOFTLIMIT_EVENTS_TARGET;
992 break;
993 case MEM_CGROUP_TARGET_NUMAINFO:
994 next = val + NUMAINFO_EVENTS_TARGET;
995 break;
996 default:
997 break;
998 }
999 __this_cpu_write(memcg->stat->targets[target], next);
1000 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001001 }
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001002 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001003}
1004
1005/*
1006 * Check events in order.
1007 *
1008 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001009static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001010{
Steven Rostedt47994012011-11-02 13:38:33 -07001011 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001012 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001013 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1014 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001015 bool do_softlimit;
1016 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001017
1018 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1019 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001020#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001021 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1022 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001023#endif
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001024 preempt_enable();
1025
1026 mem_cgroup_threshold(memcg);
1027 if (unlikely(do_softlimit))
1028 mem_cgroup_update_tree(memcg, page);
1029#if MAX_NUMNODES > 1
1030 if (unlikely(do_numainfo))
1031 atomic_inc(&memcg->numainfo_events);
1032#endif
1033 } else
1034 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001035}
1036
Balbir Singhcf475ad2008-04-29 01:00:16 -07001037struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001038{
Balbir Singh31a78f22008-09-28 23:09:31 +01001039 /*
1040 * mm_update_next_owner() may clear mm->owner to NULL
1041 * if it races with swapoff, page migration, etc.
1042 * So this can be called with p == NULL.
1043 */
1044 if (unlikely(!p))
1045 return NULL;
1046
Tejun Heo8af01f52013-08-08 20:11:22 -04001047 return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001048}
1049
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001050struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001051{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001052 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001053
1054 if (!mm)
1055 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001056 /*
1057 * Because we have no locks, mm->owner's may be being moved to other
1058 * cgroup. We use css_tryget() here even if this looks
1059 * pessimistic (rather than adding locks here).
1060 */
1061 rcu_read_lock();
1062 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001063 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1064 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001065 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001066 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001067 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001068 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001069}
1070
Michal Hocko16248d82013-04-29 15:07:19 -07001071/*
1072 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1073 * ref. count) or NULL if the whole root's subtree has been visited.
1074 *
1075 * helper function to be used by mem_cgroup_iter
1076 */
1077static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
1078 struct mem_cgroup *last_visited)
1079{
Tejun Heo492eb212013-08-08 20:11:25 -04001080 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001081
Tejun Heobd8815a2013-08-08 20:11:27 -04001082 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001083skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001084 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001085
1086 /*
1087 * Even if we found a group we have to make sure it is
1088 * alive. css && !memcg means that the groups should be
1089 * skipped and we should continue the tree walk.
1090 * last_visited css is safe to use because it is
1091 * protected by css_get and the tree walk is rcu safe.
1092 */
Tejun Heo492eb212013-08-08 20:11:25 -04001093 if (next_css) {
1094 struct mem_cgroup *mem = mem_cgroup_from_css(next_css);
1095
Michal Hocko16248d82013-04-29 15:07:19 -07001096 if (css_tryget(&mem->css))
1097 return mem;
1098 else {
Tejun Heo492eb212013-08-08 20:11:25 -04001099 prev_css = next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001100 goto skip_node;
1101 }
1102 }
1103
1104 return NULL;
1105}
1106
Johannes Weiner519ebea2013-07-03 15:04:51 -07001107static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1108{
1109 /*
1110 * When a group in the hierarchy below root is destroyed, the
1111 * hierarchy iterator can no longer be trusted since it might
1112 * have pointed to the destroyed group. Invalidate it.
1113 */
1114 atomic_inc(&root->dead_count);
1115}
1116
1117static struct mem_cgroup *
1118mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1119 struct mem_cgroup *root,
1120 int *sequence)
1121{
1122 struct mem_cgroup *position = NULL;
1123 /*
1124 * A cgroup destruction happens in two stages: offlining and
1125 * release. They are separated by a RCU grace period.
1126 *
1127 * If the iterator is valid, we may still race with an
1128 * offlining. The RCU lock ensures the object won't be
1129 * released, tryget will fail if we lost the race.
1130 */
1131 *sequence = atomic_read(&root->dead_count);
1132 if (iter->last_dead_count == *sequence) {
1133 smp_rmb();
1134 position = iter->last_visited;
1135 if (position && !css_tryget(&position->css))
1136 position = NULL;
1137 }
1138 return position;
1139}
1140
1141static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1142 struct mem_cgroup *last_visited,
1143 struct mem_cgroup *new_position,
1144 int sequence)
1145{
1146 if (last_visited)
1147 css_put(&last_visited->css);
1148 /*
1149 * We store the sequence count from the time @last_visited was
1150 * loaded successfully instead of rereading it here so that we
1151 * don't lose destruction events in between. We could have
1152 * raced with the destruction of @new_position after all.
1153 */
1154 iter->last_visited = new_position;
1155 smp_wmb();
1156 iter->last_dead_count = sequence;
1157}
1158
Johannes Weiner56600482012-01-12 17:17:59 -08001159/**
1160 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1161 * @root: hierarchy root
1162 * @prev: previously returned memcg, NULL on first invocation
1163 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1164 *
1165 * Returns references to children of the hierarchy below @root, or
1166 * @root itself, or %NULL after a full round-trip.
1167 *
1168 * Caller must pass the return value in @prev on subsequent
1169 * invocations for reference counting, or use mem_cgroup_iter_break()
1170 * to cancel a hierarchy walk before the round-trip is complete.
1171 *
1172 * Reclaimers can specify a zone and a priority level in @reclaim to
1173 * divide up the memcgs in the hierarchy among all concurrent
1174 * reclaimers operating on the same zone and priority.
1175 */
1176struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
1177 struct mem_cgroup *prev,
1178 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001179{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001180 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001181 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001182
Johannes Weiner56600482012-01-12 17:17:59 -08001183 if (mem_cgroup_disabled())
1184 return NULL;
1185
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001186 if (!root)
1187 root = root_mem_cgroup;
1188
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001189 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001190 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001191
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001192 if (!root->use_hierarchy && root != root_mem_cgroup) {
1193 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001194 goto out_css_put;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001195 return root;
1196 }
1197
Michal Hocko542f85f2013-04-29 15:07:15 -07001198 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001199 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001200 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001201 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001202
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001203 if (reclaim) {
1204 int nid = zone_to_nid(reclaim->zone);
1205 int zid = zone_idx(reclaim->zone);
1206 struct mem_cgroup_per_zone *mz;
1207
1208 mz = mem_cgroup_zoneinfo(root, nid, zid);
1209 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001210 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001211 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001212 goto out_unlock;
1213 }
Michal Hocko5f578162013-04-29 15:07:17 -07001214
Johannes Weiner519ebea2013-07-03 15:04:51 -07001215 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001216 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001217
Michal Hocko16248d82013-04-29 15:07:19 -07001218 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001219
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001220 if (reclaim) {
Johannes Weiner519ebea2013-07-03 15:04:51 -07001221 mem_cgroup_iter_update(iter, last_visited, memcg, seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001222
Michal Hocko19f39402013-04-29 15:07:18 -07001223 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001224 iter->generation++;
1225 else if (!prev && memcg)
1226 reclaim->generation = iter->generation;
1227 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001228
Michal Hocko19f39402013-04-29 15:07:18 -07001229 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001230 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001231 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001232out_unlock:
1233 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001234out_css_put:
1235 if (prev && prev != root)
1236 css_put(&prev->css);
1237
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001238 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001239}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001240
Johannes Weiner56600482012-01-12 17:17:59 -08001241/**
1242 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1243 * @root: hierarchy root
1244 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1245 */
1246void mem_cgroup_iter_break(struct mem_cgroup *root,
1247 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001248{
1249 if (!root)
1250 root = root_mem_cgroup;
1251 if (prev && prev != root)
1252 css_put(&prev->css);
1253}
1254
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001255/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001256 * Iteration constructs for visiting all cgroups (under a tree). If
1257 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1258 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001259 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001260#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001261 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001262 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001263 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001264
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001265#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001266 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001267 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001268 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001269
David Rientjes68ae5642012-12-12 13:51:57 -08001270void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001271{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001272 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001273
Ying Han456f9982011-05-26 16:25:38 -07001274 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001275 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1276 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001277 goto out;
1278
1279 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001280 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001281 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1282 break;
1283 case PGMAJFAULT:
1284 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001285 break;
1286 default:
1287 BUG();
1288 }
1289out:
1290 rcu_read_unlock();
1291}
David Rientjes68ae5642012-12-12 13:51:57 -08001292EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001293
Johannes Weiner925b7672012-01-12 17:18:15 -08001294/**
1295 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1296 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001297 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001298 *
1299 * Returns the lru list vector holding pages for the given @zone and
1300 * @mem. This can be the global zone lruvec, if the memory controller
1301 * is disabled.
1302 */
1303struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1304 struct mem_cgroup *memcg)
1305{
1306 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001307 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001308
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001309 if (mem_cgroup_disabled()) {
1310 lruvec = &zone->lruvec;
1311 goto out;
1312 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001313
1314 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001315 lruvec = &mz->lruvec;
1316out:
1317 /*
1318 * Since a node can be onlined after the mem_cgroup was created,
1319 * we have to be prepared to initialize lruvec->zone here;
1320 * and if offlined then reonlined, we need to reinitialize it.
1321 */
1322 if (unlikely(lruvec->zone != zone))
1323 lruvec->zone = zone;
1324 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001325}
1326
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001327/*
1328 * Following LRU functions are allowed to be used without PCG_LOCK.
1329 * Operations are called by routine of global LRU independently from memcg.
1330 * What we have to take care of here is validness of pc->mem_cgroup.
1331 *
1332 * Changes to pc->mem_cgroup happens when
1333 * 1. charge
1334 * 2. moving account
1335 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1336 * It is added to LRU before charge.
1337 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1338 * When moving account, the page is not on LRU. It's isolated.
1339 */
1340
Johannes Weiner925b7672012-01-12 17:18:15 -08001341/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001342 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001343 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001344 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001345 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001346struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001347{
1348 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001349 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001350 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001351 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001352
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001353 if (mem_cgroup_disabled()) {
1354 lruvec = &zone->lruvec;
1355 goto out;
1356 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001357
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001358 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001359 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001360
1361 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001362 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001363 * an uncharged page off lru does nothing to secure
1364 * its former mem_cgroup from sudden removal.
1365 *
1366 * Our caller holds lru_lock, and PageCgroupUsed is updated
1367 * under page_cgroup lock: between them, they make all uses
1368 * of pc->mem_cgroup safe.
1369 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001370 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001371 pc->mem_cgroup = memcg = root_mem_cgroup;
1372
Johannes Weiner925b7672012-01-12 17:18:15 -08001373 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001374 lruvec = &mz->lruvec;
1375out:
1376 /*
1377 * Since a node can be onlined after the mem_cgroup was created,
1378 * we have to be prepared to initialize lruvec->zone here;
1379 * and if offlined then reonlined, we need to reinitialize it.
1380 */
1381 if (unlikely(lruvec->zone != zone))
1382 lruvec->zone = zone;
1383 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001384}
1385
1386/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001387 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1388 * @lruvec: mem_cgroup per zone lru vector
1389 * @lru: index of lru list the page is sitting on
1390 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001391 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001392 * This function must be called when a page is added to or removed from an
1393 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001394 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001395void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1396 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001397{
1398 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001399 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001400
1401 if (mem_cgroup_disabled())
1402 return;
1403
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001404 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1405 lru_size = mz->lru_size + lru;
1406 *lru_size += nr_pages;
1407 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001408}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001409
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001410/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001411 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001412 * hierarchy subtree
1413 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001414bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1415 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001416{
Johannes Weiner91c637342012-05-29 15:06:24 -07001417 if (root_memcg == memcg)
1418 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001419 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001420 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001421 return css_is_ancestor(&memcg->css, &root_memcg->css);
1422}
1423
1424static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1425 struct mem_cgroup *memcg)
1426{
1427 bool ret;
1428
Johannes Weiner91c637342012-05-29 15:06:24 -07001429 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001430 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001431 rcu_read_unlock();
1432 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001433}
1434
David Rientjesffbdccf2013-07-03 15:01:23 -07001435bool task_in_mem_cgroup(struct task_struct *task,
1436 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001437{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001438 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001439 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001440 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001441
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001442 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001443 if (p) {
1444 curr = try_get_mem_cgroup_from_mm(p->mm);
1445 task_unlock(p);
1446 } else {
1447 /*
1448 * All threads may have already detached their mm's, but the oom
1449 * killer still needs to detect if they have already been oom
1450 * killed to prevent needlessly killing additional tasks.
1451 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001452 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001453 curr = mem_cgroup_from_task(task);
1454 if (curr)
1455 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001456 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001457 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001458 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001459 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001460 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001461 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001462 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001463 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1464 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001465 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001466 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001467 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001468 return ret;
1469}
1470
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001471int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001472{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001473 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001474 unsigned long inactive;
1475 unsigned long active;
1476 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001477
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001478 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1479 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001480
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001481 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1482 if (gb)
1483 inactive_ratio = int_sqrt(10 * gb);
1484 else
1485 inactive_ratio = 1;
1486
Johannes Weiner9b272972011-11-02 13:38:23 -07001487 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001488}
1489
Balbir Singh6d61ef42009-01-07 18:08:06 -08001490#define mem_cgroup_from_res_counter(counter, member) \
1491 container_of(counter, struct mem_cgroup, member)
1492
Johannes Weiner19942822011-02-01 15:52:43 -08001493/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001494 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001495 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001496 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001497 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001498 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001499 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001500static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001501{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001502 unsigned long long margin;
1503
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001504 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001505 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001506 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001507 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001508}
1509
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001510int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001511{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001512 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001513 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001514 return vm_swappiness;
1515
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001516 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001517}
1518
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001519/*
1520 * memcg->moving_account is used for checking possibility that some thread is
1521 * calling move_account(). When a thread on CPU-A starts moving pages under
1522 * a memcg, other threads should check memcg->moving_account under
1523 * rcu_read_lock(), like this:
1524 *
1525 * CPU-A CPU-B
1526 * rcu_read_lock()
1527 * memcg->moving_account+1 if (memcg->mocing_account)
1528 * take heavy locks.
1529 * synchronize_rcu() update something.
1530 * rcu_read_unlock()
1531 * start move here.
1532 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001533
1534/* for quick checking without looking up memcg */
1535atomic_t memcg_moving __read_mostly;
1536
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001537static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001538{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001539 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001540 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001541 synchronize_rcu();
1542}
1543
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001544static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001545{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001546 /*
1547 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1548 * We check NULL in callee rather than caller.
1549 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001550 if (memcg) {
1551 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001552 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001553 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001554}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001555
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001556/*
1557 * 2 routines for checking "mem" is under move_account() or not.
1558 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001559 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1560 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001561 * pc->mem_cgroup may be overwritten.
1562 *
1563 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1564 * under hierarchy of moving cgroups. This is for
1565 * waiting at hith-memory prressure caused by "move".
1566 */
1567
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001568static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001569{
1570 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001571 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001572}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001573
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001574static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001575{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001576 struct mem_cgroup *from;
1577 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001578 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001579 /*
1580 * Unlike task_move routines, we access mc.to, mc.from not under
1581 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1582 */
1583 spin_lock(&mc.lock);
1584 from = mc.from;
1585 to = mc.to;
1586 if (!from)
1587 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001588
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001589 ret = mem_cgroup_same_or_subtree(memcg, from)
1590 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001591unlock:
1592 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001593 return ret;
1594}
1595
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001596static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001597{
1598 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001599 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001600 DEFINE_WAIT(wait);
1601 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1602 /* moving charge context might have finished. */
1603 if (mc.moving_task)
1604 schedule();
1605 finish_wait(&mc.waitq, &wait);
1606 return true;
1607 }
1608 }
1609 return false;
1610}
1611
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001612/*
1613 * Take this lock when
1614 * - a code tries to modify page's memcg while it's USED.
1615 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001616 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001617 */
1618static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1619 unsigned long *flags)
1620{
1621 spin_lock_irqsave(&memcg->move_lock, *flags);
1622}
1623
1624static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1625 unsigned long *flags)
1626{
1627 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1628}
1629
Sha Zhengju58cf1882013-02-22 16:32:05 -08001630#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001631/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001632 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001633 * @memcg: The memory cgroup that went over limit
1634 * @p: Task that is going to be killed
1635 *
1636 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1637 * enabled
1638 */
1639void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1640{
1641 struct cgroup *task_cgrp;
1642 struct cgroup *mem_cgrp;
1643 /*
1644 * Need a buffer in BSS, can't rely on allocations. The code relies
1645 * on the assumption that OOM is serialized for memory controller.
1646 * If this assumption is broken, revisit this code.
1647 */
1648 static char memcg_name[PATH_MAX];
1649 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001650 struct mem_cgroup *iter;
1651 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001652
Sha Zhengju58cf1882013-02-22 16:32:05 -08001653 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001654 return;
1655
Balbir Singhe2224322009-04-02 16:57:39 -07001656 rcu_read_lock();
1657
1658 mem_cgrp = memcg->css.cgroup;
1659 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1660
1661 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1662 if (ret < 0) {
1663 /*
1664 * Unfortunately, we are unable to convert to a useful name
1665 * But we'll still print out the usage information
1666 */
1667 rcu_read_unlock();
1668 goto done;
1669 }
1670 rcu_read_unlock();
1671
Andrew Mortond0451972013-02-22 16:32:06 -08001672 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001673
1674 rcu_read_lock();
1675 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1676 if (ret < 0) {
1677 rcu_read_unlock();
1678 goto done;
1679 }
1680 rcu_read_unlock();
1681
1682 /*
1683 * Continues from above, so we don't need an KERN_ level
1684 */
Andrew Mortond0451972013-02-22 16:32:06 -08001685 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001686done:
1687
Andrew Mortond0451972013-02-22 16:32:06 -08001688 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001689 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1690 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1691 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001692 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001693 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1694 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1695 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001696 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001697 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1698 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1699 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001700
1701 for_each_mem_cgroup_tree(iter, memcg) {
1702 pr_info("Memory cgroup stats");
1703
1704 rcu_read_lock();
1705 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1706 if (!ret)
1707 pr_cont(" for %s", memcg_name);
1708 rcu_read_unlock();
1709 pr_cont(":");
1710
1711 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1712 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1713 continue;
1714 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1715 K(mem_cgroup_read_stat(iter, i)));
1716 }
1717
1718 for (i = 0; i < NR_LRU_LISTS; i++)
1719 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1720 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1721
1722 pr_cont("\n");
1723 }
Balbir Singhe2224322009-04-02 16:57:39 -07001724}
1725
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001726/*
1727 * This function returns the number of memcg under hierarchy tree. Returns
1728 * 1(self count) if no children.
1729 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001730static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001731{
1732 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001733 struct mem_cgroup *iter;
1734
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001735 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001736 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001737 return num;
1738}
1739
Balbir Singh6d61ef42009-01-07 18:08:06 -08001740/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001741 * Return the memory (and swap, if configured) limit for a memcg.
1742 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001743static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001744{
1745 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001746
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001747 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001748
David Rientjesa63d83f2010-08-09 17:19:46 -07001749 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001750 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001751 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001752 if (mem_cgroup_swappiness(memcg)) {
1753 u64 memsw;
1754
1755 limit += total_swap_pages << PAGE_SHIFT;
1756 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1757
1758 /*
1759 * If memsw is finite and limits the amount of swap space
1760 * available to this memcg, return that limit.
1761 */
1762 limit = min(limit, memsw);
1763 }
1764
1765 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001766}
1767
David Rientjes19965462012-12-11 16:00:26 -08001768static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1769 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001770{
1771 struct mem_cgroup *iter;
1772 unsigned long chosen_points = 0;
1773 unsigned long totalpages;
1774 unsigned int points = 0;
1775 struct task_struct *chosen = NULL;
1776
David Rientjes876aafb2012-07-31 16:43:48 -07001777 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001778 * If current has a pending SIGKILL or is exiting, then automatically
1779 * select it. The goal is to allow it to allocate so that it may
1780 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001781 */
David Rientjes465adcf2013-04-29 15:08:45 -07001782 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001783 set_thread_flag(TIF_MEMDIE);
1784 return;
1785 }
1786
1787 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001788 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1789 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001790 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001791 struct task_struct *task;
1792
Tejun Heo72ec7022013-08-08 20:11:26 -04001793 css_task_iter_start(&iter->css, &it);
1794 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001795 switch (oom_scan_process_thread(task, totalpages, NULL,
1796 false)) {
1797 case OOM_SCAN_SELECT:
1798 if (chosen)
1799 put_task_struct(chosen);
1800 chosen = task;
1801 chosen_points = ULONG_MAX;
1802 get_task_struct(chosen);
1803 /* fall through */
1804 case OOM_SCAN_CONTINUE:
1805 continue;
1806 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001807 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001808 mem_cgroup_iter_break(memcg, iter);
1809 if (chosen)
1810 put_task_struct(chosen);
1811 return;
1812 case OOM_SCAN_OK:
1813 break;
1814 };
1815 points = oom_badness(task, memcg, NULL, totalpages);
1816 if (points > chosen_points) {
1817 if (chosen)
1818 put_task_struct(chosen);
1819 chosen = task;
1820 chosen_points = points;
1821 get_task_struct(chosen);
1822 }
1823 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001824 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001825 }
1826
1827 if (!chosen)
1828 return;
1829 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001830 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1831 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001832}
1833
Johannes Weiner56600482012-01-12 17:17:59 -08001834static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1835 gfp_t gfp_mask,
1836 unsigned long flags)
1837{
1838 unsigned long total = 0;
1839 bool noswap = false;
1840 int loop;
1841
1842 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1843 noswap = true;
1844 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1845 noswap = true;
1846
1847 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1848 if (loop)
1849 drain_all_stock_async(memcg);
1850 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1851 /*
1852 * Allow limit shrinkers, which are triggered directly
1853 * by userspace, to catch signals and stop reclaim
1854 * after minimal progress, regardless of the margin.
1855 */
1856 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1857 break;
1858 if (mem_cgroup_margin(memcg))
1859 break;
1860 /*
1861 * If nothing was reclaimed after two attempts, there
1862 * may be no reclaimable pages in this hierarchy.
1863 */
1864 if (loop && !total)
1865 break;
1866 }
1867 return total;
1868}
1869
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001870/**
1871 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001872 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001873 * @nid: the node ID to be checked.
1874 * @noswap : specify true here if the user wants flle only information.
1875 *
1876 * This function returns whether the specified memcg contains any
1877 * reclaimable pages on a node. Returns true if there are any reclaimable
1878 * pages in the node.
1879 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001880static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001881 int nid, bool noswap)
1882{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001883 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001884 return true;
1885 if (noswap || !total_swap_pages)
1886 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001887 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001888 return true;
1889 return false;
1890
1891}
Ying Han889976d2011-05-26 16:25:33 -07001892#if MAX_NUMNODES > 1
1893
1894/*
1895 * Always updating the nodemask is not very good - even if we have an empty
1896 * list or the wrong list here, we can start from some node and traverse all
1897 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1898 *
1899 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001900static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001901{
1902 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001903 /*
1904 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1905 * pagein/pageout changes since the last update.
1906 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001907 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001908 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001909 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001910 return;
1911
Ying Han889976d2011-05-26 16:25:33 -07001912 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001913 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001914
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001915 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001916
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001917 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1918 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001919 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001920
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001921 atomic_set(&memcg->numainfo_events, 0);
1922 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001923}
1924
1925/*
1926 * Selecting a node where we start reclaim from. Because what we need is just
1927 * reducing usage counter, start from anywhere is O,K. Considering
1928 * memory reclaim from current node, there are pros. and cons.
1929 *
1930 * Freeing memory from current node means freeing memory from a node which
1931 * we'll use or we've used. So, it may make LRU bad. And if several threads
1932 * hit limits, it will see a contention on a node. But freeing from remote
1933 * node means more costs for memory reclaim because of memory latency.
1934 *
1935 * Now, we use round-robin. Better algorithm is welcomed.
1936 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001937int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001938{
1939 int node;
1940
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001941 mem_cgroup_may_update_nodemask(memcg);
1942 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001943
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001944 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001945 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001946 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001947 /*
1948 * We call this when we hit limit, not when pages are added to LRU.
1949 * No LRU may hold pages because all pages are UNEVICTABLE or
1950 * memcg is too small and all pages are not on LRU. In that case,
1951 * we use curret node.
1952 */
1953 if (unlikely(node == MAX_NUMNODES))
1954 node = numa_node_id();
1955
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001956 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001957 return node;
1958}
1959
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001960/*
1961 * Check all nodes whether it contains reclaimable pages or not.
1962 * For quick scan, we make use of scan_nodes. This will allow us to skip
1963 * unused nodes. But scan_nodes is lazily updated and may not cotain
1964 * enough new information. We need to do double check.
1965 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001966static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001967{
1968 int nid;
1969
1970 /*
1971 * quick check...making use of scan_node.
1972 * We can skip unused nodes.
1973 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001974 if (!nodes_empty(memcg->scan_nodes)) {
1975 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001976 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001977 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001978
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001979 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001980 return true;
1981 }
1982 }
1983 /*
1984 * Check rest of nodes.
1985 */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001986 for_each_node_state(nid, N_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001987 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001988 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001989 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001990 return true;
1991 }
1992 return false;
1993}
1994
Ying Han889976d2011-05-26 16:25:33 -07001995#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001996int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001997{
1998 return 0;
1999}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002000
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002001static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002002{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002003 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002004}
Ying Han889976d2011-05-26 16:25:33 -07002005#endif
2006
Michal Hocko3b387222013-09-12 15:13:21 -07002007/*
2008 * A group is eligible for the soft limit reclaim if it is
2009 * a) is over its soft limit
2010 * b) any parent up the hierarchy is over its soft limit
2011 */
2012bool mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002013{
Michal Hocko3b387222013-09-12 15:13:21 -07002014 struct mem_cgroup *parent = memcg;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002015
Michal Hocko3b387222013-09-12 15:13:21 -07002016 if (res_counter_soft_limit_excess(&memcg->res))
2017 return true;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002018
Michal Hocko3b387222013-09-12 15:13:21 -07002019 /*
2020 * If any parent up the hierarchy is over its soft limit then we
2021 * have to obey and reclaim from this group as well.
2022 */
2023 while((parent = parent_mem_cgroup(parent))) {
2024 if (res_counter_soft_limit_excess(&parent->res))
2025 return true;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002026 }
Michal Hocko3b387222013-09-12 15:13:21 -07002027
2028 return false;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002029}
2030
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002031/*
2032 * Check OOM-Killer is already running under our hierarchy.
2033 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07002034 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002035 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002036static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002037{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002038 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002039
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002040 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002041 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002042 /*
2043 * this subtree of our hierarchy is already locked
2044 * so we cannot give a lock.
2045 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002046 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002047 mem_cgroup_iter_break(memcg, iter);
2048 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002049 } else
2050 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002051 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002052
Michal Hocko79dfdac2011-07-26 16:08:23 -07002053 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07002054 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002055
2056 /*
2057 * OK, we failed to lock the whole subtree so we have to clean up
2058 * what we set up to the failing subtree
2059 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002060 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002061 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002062 mem_cgroup_iter_break(memcg, iter);
2063 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002064 }
2065 iter->oom_lock = false;
2066 }
Johannes Weiner23751be2011-08-25 15:59:16 -07002067 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002068}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002069
Michal Hocko79dfdac2011-07-26 16:08:23 -07002070/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07002071 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07002072 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002073static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002074{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002075 struct mem_cgroup *iter;
2076
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002077 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002078 iter->oom_lock = false;
2079 return 0;
2080}
2081
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002082static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002083{
2084 struct mem_cgroup *iter;
2085
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002086 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002087 atomic_inc(&iter->under_oom);
2088}
2089
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002090static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002091{
2092 struct mem_cgroup *iter;
2093
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002094 /*
2095 * When a new child is created while the hierarchy is under oom,
2096 * mem_cgroup_oom_lock() may not be called. We have to use
2097 * atomic_add_unless() here.
2098 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002099 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002100 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002101}
2102
Michal Hocko1af8efe2011-07-26 16:08:24 -07002103static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002104static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2105
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002106struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002107 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002108 wait_queue_t wait;
2109};
2110
2111static int memcg_oom_wake_function(wait_queue_t *wait,
2112 unsigned mode, int sync, void *arg)
2113{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002114 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2115 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002116 struct oom_wait_info *oom_wait_info;
2117
2118 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002119 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002120
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002121 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002122 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002123 * Then we can use css_is_ancestor without taking care of RCU.
2124 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002125 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2126 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002127 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002128 return autoremove_wake_function(wait, mode, sync, arg);
2129}
2130
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002131static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002132{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002133 /* for filtering, pass "memcg" as argument. */
2134 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002135}
2136
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002137static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002138{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002139 if (memcg && atomic_read(&memcg->under_oom))
2140 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002141}
2142
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002143/*
2144 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
2145 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002146static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
2147 int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002148{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002149 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002150 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002151
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002152 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002153 owait.wait.flags = 0;
2154 owait.wait.func = memcg_oom_wake_function;
2155 owait.wait.private = current;
2156 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002157 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002158 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002159
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002160 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07002161 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002162 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002163 /*
2164 * Even if signal_pending(), we can't quit charge() loop without
2165 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
2166 * under OOM is always welcomed, use TASK_KILLABLE here.
2167 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002168 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002169 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002170 need_to_kill = false;
2171 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002172 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002173 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002174
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002175 if (need_to_kill) {
2176 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07002177 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002178 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002179 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002180 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002181 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07002182 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002183 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002184 mem_cgroup_oom_unlock(memcg);
2185 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002186 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002187
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002188 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002189
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002190 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
2191 return false;
2192 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07002193 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002194 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002195}
2196
Balbir Singhd69b0422009-06-17 16:26:34 -07002197/*
2198 * Currently used to update mapped file statistics, but the routine can be
2199 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002200 *
2201 * Notes: Race condition
2202 *
2203 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2204 * it tends to be costly. But considering some conditions, we doesn't need
2205 * to do so _always_.
2206 *
2207 * Considering "charge", lock_page_cgroup() is not required because all
2208 * file-stat operations happen after a page is attached to radix-tree. There
2209 * are no race with "charge".
2210 *
2211 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2212 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2213 * if there are race with "uncharge". Statistics itself is properly handled
2214 * by flags.
2215 *
2216 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002217 * small, we check mm->moving_account and detect there are possibility of race
2218 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002219 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002220
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002221void __mem_cgroup_begin_update_page_stat(struct page *page,
2222 bool *locked, unsigned long *flags)
2223{
2224 struct mem_cgroup *memcg;
2225 struct page_cgroup *pc;
2226
2227 pc = lookup_page_cgroup(page);
2228again:
2229 memcg = pc->mem_cgroup;
2230 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2231 return;
2232 /*
2233 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002234 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002235 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002236 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002237 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002238 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002239 return;
2240
2241 move_lock_mem_cgroup(memcg, flags);
2242 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2243 move_unlock_mem_cgroup(memcg, flags);
2244 goto again;
2245 }
2246 *locked = true;
2247}
2248
2249void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2250{
2251 struct page_cgroup *pc = lookup_page_cgroup(page);
2252
2253 /*
2254 * It's guaranteed that pc->mem_cgroup never changes while
2255 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002256 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002257 */
2258 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2259}
2260
Greg Thelen2a7106f2011-01-13 15:47:37 -08002261void mem_cgroup_update_page_stat(struct page *page,
2262 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002263{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002264 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002265 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002266 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002267
Johannes Weinercfa44942012-01-12 17:18:38 -08002268 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002269 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002270
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002271 memcg = pc->mem_cgroup;
2272 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002273 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002274
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002275 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08002276 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08002277 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002278 break;
2279 default:
2280 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002281 }
Balbir Singhd69b0422009-06-17 16:26:34 -07002282
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002283 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002284}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002285
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002286/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002287 * size of first charge trial. "32" comes from vmscan.c's magic value.
2288 * TODO: maybe necessary to use big numbers in big irons.
2289 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002290#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002291struct memcg_stock_pcp {
2292 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002293 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002294 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002295 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002296#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002297};
2298static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002299static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002300
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002301/**
2302 * consume_stock: Try to consume stocked charge on this cpu.
2303 * @memcg: memcg to consume from.
2304 * @nr_pages: how many pages to charge.
2305 *
2306 * The charges will only happen if @memcg matches the current cpu's memcg
2307 * stock, and at least @nr_pages are available in that stock. Failure to
2308 * service an allocation will refill the stock.
2309 *
2310 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002311 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002312static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002313{
2314 struct memcg_stock_pcp *stock;
2315 bool ret = true;
2316
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002317 if (nr_pages > CHARGE_BATCH)
2318 return false;
2319
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002320 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002321 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2322 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002323 else /* need to call res_counter_charge */
2324 ret = false;
2325 put_cpu_var(memcg_stock);
2326 return ret;
2327}
2328
2329/*
2330 * Returns stocks cached in percpu to res_counter and reset cached information.
2331 */
2332static void drain_stock(struct memcg_stock_pcp *stock)
2333{
2334 struct mem_cgroup *old = stock->cached;
2335
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002336 if (stock->nr_pages) {
2337 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2338
2339 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002340 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002341 res_counter_uncharge(&old->memsw, bytes);
2342 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002343 }
2344 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002345}
2346
2347/*
2348 * This must be called under preempt disabled or must be called by
2349 * a thread which is pinned to local cpu.
2350 */
2351static void drain_local_stock(struct work_struct *dummy)
2352{
2353 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2354 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002355 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002356}
2357
Michal Hockoe4777492013-02-22 16:35:40 -08002358static void __init memcg_stock_init(void)
2359{
2360 int cpu;
2361
2362 for_each_possible_cpu(cpu) {
2363 struct memcg_stock_pcp *stock =
2364 &per_cpu(memcg_stock, cpu);
2365 INIT_WORK(&stock->work, drain_local_stock);
2366 }
2367}
2368
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369/*
2370 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002371 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002372 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002373static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002374{
2375 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2376
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002377 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002378 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002379 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002380 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002381 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002382 put_cpu_var(memcg_stock);
2383}
2384
2385/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002386 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002387 * of the hierarchy under it. sync flag says whether we should block
2388 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002389 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002390static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002391{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002392 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002393
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002394 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002395 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002396 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002397 for_each_online_cpu(cpu) {
2398 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002399 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002400
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002401 memcg = stock->cached;
2402 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002403 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002404 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002405 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002406 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2407 if (cpu == curcpu)
2408 drain_local_stock(&stock->work);
2409 else
2410 schedule_work_on(cpu, &stock->work);
2411 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002412 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002413 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002414
2415 if (!sync)
2416 goto out;
2417
2418 for_each_online_cpu(cpu) {
2419 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002420 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002421 flush_work(&stock->work);
2422 }
2423out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002424 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002425}
2426
2427/*
2428 * Tries to drain stocked charges in other cpus. This function is asynchronous
2429 * and just put a work per cpu for draining localy on each cpu. Caller can
2430 * expects some charges will be back to res_counter later but cannot wait for
2431 * it.
2432 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002433static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002434{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002435 /*
2436 * If someone calls draining, avoid adding more kworker runs.
2437 */
2438 if (!mutex_trylock(&percpu_charge_mutex))
2439 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002440 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002441 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002442}
2443
2444/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002445static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002446{
2447 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002448 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002449 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002450 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002451}
2452
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002453/*
2454 * This function drains percpu counter value from DEAD cpu and
2455 * move it to local cpu. Note that this function can be preempted.
2456 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002457static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002458{
2459 int i;
2460
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002461 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002462 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002463 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002464
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002465 per_cpu(memcg->stat->count[i], cpu) = 0;
2466 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002467 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002468 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002469 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002470
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002471 per_cpu(memcg->stat->events[i], cpu) = 0;
2472 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002473 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002474 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002475}
2476
Paul Gortmaker0db06282013-06-19 14:53:51 -04002477static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002478 unsigned long action,
2479 void *hcpu)
2480{
2481 int cpu = (unsigned long)hcpu;
2482 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002483 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002484
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002485 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002486 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002487
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002488 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002489 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002490
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002491 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002492 mem_cgroup_drain_pcp_counter(iter, cpu);
2493
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002494 stock = &per_cpu(memcg_stock, cpu);
2495 drain_stock(stock);
2496 return NOTIFY_OK;
2497}
2498
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002499
2500/* See __mem_cgroup_try_charge() for details */
2501enum {
2502 CHARGE_OK, /* success */
2503 CHARGE_RETRY, /* need to retry but retry is not bad */
2504 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2505 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2506 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2507};
2508
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002509static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002510 unsigned int nr_pages, unsigned int min_pages,
2511 bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002512{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002513 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002514 struct mem_cgroup *mem_over_limit;
2515 struct res_counter *fail_res;
2516 unsigned long flags = 0;
2517 int ret;
2518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002519 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002520
2521 if (likely(!ret)) {
2522 if (!do_swap_account)
2523 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002524 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002525 if (likely(!ret))
2526 return CHARGE_OK;
2527
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002528 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002529 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2530 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2531 } else
2532 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002533 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002534 * Never reclaim on behalf of optional batching, retry with a
2535 * single page instead.
2536 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002537 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002538 return CHARGE_RETRY;
2539
2540 if (!(gfp_mask & __GFP_WAIT))
2541 return CHARGE_WOULDBLOCK;
2542
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002543 if (gfp_mask & __GFP_NORETRY)
2544 return CHARGE_NOMEM;
2545
Johannes Weiner56600482012-01-12 17:17:59 -08002546 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002547 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002548 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002549 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002550 * Even though the limit is exceeded at this point, reclaim
2551 * may have been able to free some pages. Retry the charge
2552 * before killing the task.
2553 *
2554 * Only for regular pages, though: huge pages are rather
2555 * unlikely to succeed so close to the limit, and we fall back
2556 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002557 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002558 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002559 return CHARGE_RETRY;
2560
2561 /*
2562 * At task move, charge accounts can be doubly counted. So, it's
2563 * better to wait until the end of task_move if something is going on.
2564 */
2565 if (mem_cgroup_wait_acct_move(mem_over_limit))
2566 return CHARGE_RETRY;
2567
2568 /* If we don't need to call oom-killer at el, return immediately */
2569 if (!oom_check)
2570 return CHARGE_NOMEM;
2571 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002572 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002573 return CHARGE_OOM_DIE;
2574
2575 return CHARGE_RETRY;
2576}
2577
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002578/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002579 * __mem_cgroup_try_charge() does
2580 * 1. detect memcg to be charged against from passed *mm and *ptr,
2581 * 2. update res_counter
2582 * 3. call memory reclaim if necessary.
2583 *
2584 * In some special case, if the task is fatal, fatal_signal_pending() or
2585 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2586 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2587 * as possible without any hazards. 2: all pages should have a valid
2588 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2589 * pointer, that is treated as a charge to root_mem_cgroup.
2590 *
2591 * So __mem_cgroup_try_charge() will return
2592 * 0 ... on success, filling *ptr with a valid memcg pointer.
2593 * -ENOMEM ... charge failure because of resource limits.
2594 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2595 *
2596 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2597 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002598 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002599static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002600 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002601 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002602 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002603 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002604{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002605 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002606 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002607 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002608 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002609
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002610 /*
2611 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2612 * in system level. So, allow to go ahead dying process in addition to
2613 * MEMDIE process.
2614 */
2615 if (unlikely(test_thread_flag(TIF_MEMDIE)
2616 || fatal_signal_pending(current)))
2617 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002618
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002619 /*
Hugh Dickins3be912772008-02-07 00:14:19 -08002620 * We always charge the cgroup the mm_struct belongs to.
2621 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002622 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002623 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002624 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002625 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002626 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002627again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002628 if (*ptr) { /* css should be a valid one */
2629 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002630 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002631 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002632 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002633 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002634 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002635 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002636 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002637
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002638 rcu_read_lock();
2639 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002640 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002641 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002642 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002643 * race with swapoff. Then, we have small risk of mis-accouning.
2644 * But such kind of mis-account by race always happens because
2645 * we don't have cgroup_mutex(). It's overkill and we allo that
2646 * small race, here.
2647 * (*) swapoff at el will charge against mm-struct not against
2648 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002649 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002650 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002651 if (!memcg)
2652 memcg = root_mem_cgroup;
2653 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002654 rcu_read_unlock();
2655 goto done;
2656 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002657 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002658 /*
2659 * It seems dagerous to access memcg without css_get().
2660 * But considering how consume_stok works, it's not
2661 * necessary. If consume_stock success, some charges
2662 * from this memcg are cached on this cpu. So, we
2663 * don't need to call css_get()/css_tryget() before
2664 * calling consume_stock().
2665 */
2666 rcu_read_unlock();
2667 goto done;
2668 }
2669 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002670 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002671 rcu_read_unlock();
2672 goto again;
2673 }
2674 rcu_read_unlock();
2675 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002676
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002677 do {
2678 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002679
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002680 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002681 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002682 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002683 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002684 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002685
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002686 oom_check = false;
2687 if (oom && !nr_oom_retries) {
2688 oom_check = true;
2689 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2690 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002691
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002692 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages,
2693 oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002694 switch (ret) {
2695 case CHARGE_OK:
2696 break;
2697 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002698 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002699 css_put(&memcg->css);
2700 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002701 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002702 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002703 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002704 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002705 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002706 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002707 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002708 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002709 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002710 /* If oom, we never return -ENOMEM */
2711 nr_oom_retries--;
2712 break;
2713 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002715 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002716 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002717 } while (ret != CHARGE_OK);
2718
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002719 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002720 refill_stock(memcg, batch - nr_pages);
2721 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002722done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002723 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002724 return 0;
2725nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002726 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002727 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002728bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002729 *ptr = root_mem_cgroup;
2730 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002731}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002732
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002733/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002734 * Somemtimes we have to undo a charge we got by try_charge().
2735 * This function is for that and do uncharge, put css's refcnt.
2736 * gotten by try_charge().
2737 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002738static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002739 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002740{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002741 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002742 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002743
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002744 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002745 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002746 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002747 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002748}
2749
2750/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002751 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2752 * This is useful when moving usage to parent cgroup.
2753 */
2754static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2755 unsigned int nr_pages)
2756{
2757 unsigned long bytes = nr_pages * PAGE_SIZE;
2758
2759 if (mem_cgroup_is_root(memcg))
2760 return;
2761
2762 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2763 if (do_swap_account)
2764 res_counter_uncharge_until(&memcg->memsw,
2765 memcg->memsw.parent, bytes);
2766}
2767
2768/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002769 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002770 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2771 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2772 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002773 */
2774static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2775{
2776 struct cgroup_subsys_state *css;
2777
2778 /* ID 0 is unused ID */
2779 if (!id)
2780 return NULL;
2781 css = css_lookup(&mem_cgroup_subsys, id);
2782 if (!css)
2783 return NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -07002784 return mem_cgroup_from_css(css);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002785}
2786
Wu Fengguange42d9d52009-12-16 12:19:59 +01002787struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002788{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002789 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002790 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002791 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002792 swp_entry_t ent;
2793
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002794 VM_BUG_ON(!PageLocked(page));
2795
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002796 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002797 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002798 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002799 memcg = pc->mem_cgroup;
2800 if (memcg && !css_tryget(&memcg->css))
2801 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002802 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002803 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002804 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002805 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002806 memcg = mem_cgroup_lookup(id);
2807 if (memcg && !css_tryget(&memcg->css))
2808 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002809 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002810 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002811 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002812 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002813}
2814
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002815static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002816 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002817 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002818 enum charge_type ctype,
2819 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002820{
Johannes Weinerce587e62012-04-24 20:22:33 +02002821 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002822 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002823 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002824 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002825 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002826
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002827 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002828 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002829 /*
2830 * we don't need page_cgroup_lock about tail pages, becase they are not
2831 * accessed by any other context at this point.
2832 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002833
2834 /*
2835 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2836 * may already be on some other mem_cgroup's LRU. Take care of it.
2837 */
2838 if (lrucare) {
2839 zone = page_zone(page);
2840 spin_lock_irq(&zone->lru_lock);
2841 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002842 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002843 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002844 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002845 was_on_lru = true;
2846 }
2847 }
2848
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002849 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002850 /*
2851 * We access a page_cgroup asynchronously without lock_page_cgroup().
2852 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2853 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2854 * before USED bit, we need memory barrier here.
2855 * See mem_cgroup_add_lru_list(), etc.
2856 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002857 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002858 SetPageCgroupUsed(pc);
Hugh Dickins3be912772008-02-07 00:14:19 -08002859
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002860 if (lrucare) {
2861 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002862 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002863 VM_BUG_ON(PageLRU(page));
2864 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002865 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002866 }
2867 spin_unlock_irq(&zone->lru_lock);
2868 }
2869
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002870 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002871 anon = true;
2872 else
2873 anon = false;
2874
David Rientjesb070e652013-05-07 16:18:09 -07002875 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002876 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002877
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002878 /*
2879 * "charge_statistics" updated event counter. Then, check it.
2880 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2881 * if they exceeds softlimit.
2882 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002883 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002884}
2885
Glauber Costa7cf27982012-12-18 14:22:55 -08002886static DEFINE_MUTEX(set_limit_mutex);
2887
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002888#ifdef CONFIG_MEMCG_KMEM
2889static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2890{
2891 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
2892 (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK);
2893}
2894
Glauber Costa1f458cb2012-12-18 14:22:50 -08002895/*
2896 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2897 * in the memcg_cache_params struct.
2898 */
2899static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2900{
2901 struct kmem_cache *cachep;
2902
2903 VM_BUG_ON(p->is_root_cache);
2904 cachep = p->root_cache;
2905 return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)];
2906}
2907
Glauber Costa749c5412012-12-18 14:23:01 -08002908#ifdef CONFIG_SLABINFO
Tejun Heo182446d2013-08-08 20:11:24 -04002909static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css,
2910 struct cftype *cft, struct seq_file *m)
Glauber Costa749c5412012-12-18 14:23:01 -08002911{
Tejun Heo182446d2013-08-08 20:11:24 -04002912 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa749c5412012-12-18 14:23:01 -08002913 struct memcg_cache_params *params;
2914
2915 if (!memcg_can_account_kmem(memcg))
2916 return -EIO;
2917
2918 print_slabinfo_header(m);
2919
2920 mutex_lock(&memcg->slab_caches_mutex);
2921 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2922 cache_show(memcg_params_to_cache(params), m);
2923 mutex_unlock(&memcg->slab_caches_mutex);
2924
2925 return 0;
2926}
2927#endif
2928
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002929static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2930{
2931 struct res_counter *fail_res;
2932 struct mem_cgroup *_memcg;
2933 int ret = 0;
2934 bool may_oom;
2935
2936 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2937 if (ret)
2938 return ret;
2939
2940 /*
2941 * Conditions under which we can wait for the oom_killer. Those are
2942 * the same conditions tested by the core page allocator
2943 */
2944 may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY);
2945
2946 _memcg = memcg;
2947 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
2948 &_memcg, may_oom);
2949
2950 if (ret == -EINTR) {
2951 /*
2952 * __mem_cgroup_try_charge() chosed to bypass to root due to
2953 * OOM kill or fatal signal. Since our only options are to
2954 * either fail the allocation or charge it to this cgroup, do
2955 * it as a temporary condition. But we can't fail. From a
2956 * kmem/slab perspective, the cache has already been selected,
2957 * by mem_cgroup_kmem_get_cache(), so it is too late to change
2958 * our minds.
2959 *
2960 * This condition will only trigger if the task entered
2961 * memcg_charge_kmem in a sane state, but was OOM-killed during
2962 * __mem_cgroup_try_charge() above. Tasks that were already
2963 * dying when the allocation triggers should have been already
2964 * directed to the root cgroup in memcontrol.h
2965 */
2966 res_counter_charge_nofail(&memcg->res, size, &fail_res);
2967 if (do_swap_account)
2968 res_counter_charge_nofail(&memcg->memsw, size,
2969 &fail_res);
2970 ret = 0;
2971 } else if (ret)
2972 res_counter_uncharge(&memcg->kmem, size);
2973
2974 return ret;
2975}
2976
2977static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
2978{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002979 res_counter_uncharge(&memcg->res, size);
2980 if (do_swap_account)
2981 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08002982
2983 /* Not down to 0 */
2984 if (res_counter_uncharge(&memcg->kmem, size))
2985 return;
2986
Li Zefan10d5ebf2013-07-08 16:00:33 -07002987 /*
2988 * Releases a reference taken in kmem_cgroup_css_offline in case
2989 * this last uncharge is racing with the offlining code or it is
2990 * outliving the memcg existence.
2991 *
2992 * The memory barrier imposed by test&clear is paired with the
2993 * explicit one in memcg_kmem_mark_dead().
2994 */
Glauber Costa7de37682012-12-18 14:22:07 -08002995 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07002996 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002997}
2998
Glauber Costa2633d7a2012-12-18 14:22:34 -08002999void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
3000{
3001 if (!memcg)
3002 return;
3003
3004 mutex_lock(&memcg->slab_caches_mutex);
3005 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
3006 mutex_unlock(&memcg->slab_caches_mutex);
3007}
3008
3009/*
3010 * helper for acessing a memcg's index. It will be used as an index in the
3011 * child cache array in kmem_cache, and also to derive its name. This function
3012 * will return -1 when this is not a kmem-limited memcg.
3013 */
3014int memcg_cache_id(struct mem_cgroup *memcg)
3015{
3016 return memcg ? memcg->kmemcg_id : -1;
3017}
3018
Glauber Costa55007d82012-12-18 14:22:38 -08003019/*
3020 * This ends up being protected by the set_limit mutex, during normal
3021 * operation, because that is its main call site.
3022 *
3023 * But when we create a new cache, we can call this as well if its parent
3024 * is kmem-limited. That will have to hold set_limit_mutex as well.
3025 */
3026int memcg_update_cache_sizes(struct mem_cgroup *memcg)
3027{
3028 int num, ret;
3029
3030 num = ida_simple_get(&kmem_limited_groups,
3031 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
3032 if (num < 0)
3033 return num;
3034 /*
3035 * After this point, kmem_accounted (that we test atomically in
3036 * the beginning of this conditional), is no longer 0. This
3037 * guarantees only one process will set the following boolean
3038 * to true. We don't need test_and_set because we're protected
3039 * by the set_limit_mutex anyway.
3040 */
3041 memcg_kmem_set_activated(memcg);
3042
3043 ret = memcg_update_all_caches(num+1);
3044 if (ret) {
3045 ida_simple_remove(&kmem_limited_groups, num);
3046 memcg_kmem_clear_activated(memcg);
3047 return ret;
3048 }
3049
3050 memcg->kmemcg_id = num;
3051 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
3052 mutex_init(&memcg->slab_caches_mutex);
3053 return 0;
3054}
3055
3056static size_t memcg_caches_array_size(int num_groups)
3057{
3058 ssize_t size;
3059 if (num_groups <= 0)
3060 return 0;
3061
3062 size = 2 * num_groups;
3063 if (size < MEMCG_CACHES_MIN_SIZE)
3064 size = MEMCG_CACHES_MIN_SIZE;
3065 else if (size > MEMCG_CACHES_MAX_SIZE)
3066 size = MEMCG_CACHES_MAX_SIZE;
3067
3068 return size;
3069}
3070
3071/*
3072 * We should update the current array size iff all caches updates succeed. This
3073 * can only be done from the slab side. The slab mutex needs to be held when
3074 * calling this.
3075 */
3076void memcg_update_array_size(int num)
3077{
3078 if (num > memcg_limited_groups_array_size)
3079 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3080}
3081
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003082static void kmem_cache_destroy_work_func(struct work_struct *w);
3083
Glauber Costa55007d82012-12-18 14:22:38 -08003084int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3085{
3086 struct memcg_cache_params *cur_params = s->memcg_params;
3087
3088 VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache);
3089
3090 if (num_groups > memcg_limited_groups_array_size) {
3091 int i;
3092 ssize_t size = memcg_caches_array_size(num_groups);
3093
3094 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003095 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003096
3097 s->memcg_params = kzalloc(size, GFP_KERNEL);
3098 if (!s->memcg_params) {
3099 s->memcg_params = cur_params;
3100 return -ENOMEM;
3101 }
3102
3103 s->memcg_params->is_root_cache = true;
3104
3105 /*
3106 * There is the chance it will be bigger than
3107 * memcg_limited_groups_array_size, if we failed an allocation
3108 * in a cache, in which case all caches updated before it, will
3109 * have a bigger array.
3110 *
3111 * But if that is the case, the data after
3112 * memcg_limited_groups_array_size is certainly unused
3113 */
3114 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3115 if (!cur_params->memcg_caches[i])
3116 continue;
3117 s->memcg_params->memcg_caches[i] =
3118 cur_params->memcg_caches[i];
3119 }
3120
3121 /*
3122 * Ideally, we would wait until all caches succeed, and only
3123 * then free the old one. But this is not worth the extra
3124 * pointer per-cache we'd have to have for this.
3125 *
3126 * It is not a big deal if some caches are left with a size
3127 * bigger than the others. And all updates will reset this
3128 * anyway.
3129 */
3130 kfree(cur_params);
3131 }
3132 return 0;
3133}
3134
Glauber Costa943a4512012-12-18 14:23:03 -08003135int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
3136 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003137{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003138 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003139
3140 if (!memcg_kmem_enabled())
3141 return 0;
3142
Andrey Vagin90c7a792013-09-11 14:22:18 -07003143 if (!memcg) {
3144 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003145 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003146 } else
3147 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003148
Glauber Costa2633d7a2012-12-18 14:22:34 -08003149 s->memcg_params = kzalloc(size, GFP_KERNEL);
3150 if (!s->memcg_params)
3151 return -ENOMEM;
3152
Glauber Costa943a4512012-12-18 14:23:03 -08003153 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003154 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003155 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003156 INIT_WORK(&s->memcg_params->destroy,
3157 kmem_cache_destroy_work_func);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003158 } else
3159 s->memcg_params->is_root_cache = true;
3160
Glauber Costa2633d7a2012-12-18 14:22:34 -08003161 return 0;
3162}
3163
3164void memcg_release_cache(struct kmem_cache *s)
3165{
Glauber Costad7f25f82012-12-18 14:22:40 -08003166 struct kmem_cache *root;
3167 struct mem_cgroup *memcg;
3168 int id;
3169
3170 /*
3171 * This happens, for instance, when a root cache goes away before we
3172 * add any memcg.
3173 */
3174 if (!s->memcg_params)
3175 return;
3176
3177 if (s->memcg_params->is_root_cache)
3178 goto out;
3179
3180 memcg = s->memcg_params->memcg;
3181 id = memcg_cache_id(memcg);
3182
3183 root = s->memcg_params->root_cache;
3184 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08003185
3186 mutex_lock(&memcg->slab_caches_mutex);
3187 list_del(&s->memcg_params->list);
3188 mutex_unlock(&memcg->slab_caches_mutex);
3189
Li Zefan20f05312013-07-08 16:00:31 -07003190 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003191out:
Glauber Costa2633d7a2012-12-18 14:22:34 -08003192 kfree(s->memcg_params);
3193}
3194
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003195/*
3196 * During the creation a new cache, we need to disable our accounting mechanism
3197 * altogether. This is true even if we are not creating, but rather just
3198 * enqueing new caches to be created.
3199 *
3200 * This is because that process will trigger allocations; some visible, like
3201 * explicit kmallocs to auxiliary data structures, name strings and internal
3202 * cache structures; some well concealed, like INIT_WORK() that can allocate
3203 * objects during debug.
3204 *
3205 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3206 * to it. This may not be a bounded recursion: since the first cache creation
3207 * failed to complete (waiting on the allocation), we'll just try to create the
3208 * cache again, failing at the same point.
3209 *
3210 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3211 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3212 * inside the following two functions.
3213 */
3214static inline void memcg_stop_kmem_account(void)
3215{
3216 VM_BUG_ON(!current->mm);
3217 current->memcg_kmem_skip_account++;
3218}
3219
3220static inline void memcg_resume_kmem_account(void)
3221{
3222 VM_BUG_ON(!current->mm);
3223 current->memcg_kmem_skip_account--;
3224}
3225
Glauber Costa1f458cb2012-12-18 14:22:50 -08003226static void kmem_cache_destroy_work_func(struct work_struct *w)
3227{
3228 struct kmem_cache *cachep;
3229 struct memcg_cache_params *p;
3230
3231 p = container_of(w, struct memcg_cache_params, destroy);
3232
3233 cachep = memcg_params_to_cache(p);
3234
Glauber Costa22933152012-12-18 14:22:59 -08003235 /*
3236 * If we get down to 0 after shrink, we could delete right away.
3237 * However, memcg_release_pages() already puts us back in the workqueue
3238 * in that case. If we proceed deleting, we'll get a dangling
3239 * reference, and removing the object from the workqueue in that case
3240 * is unnecessary complication. We are not a fast path.
3241 *
3242 * Note that this case is fundamentally different from racing with
3243 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3244 * kmem_cache_shrink, not only we would be reinserting a dead cache
3245 * into the queue, but doing so from inside the worker racing to
3246 * destroy it.
3247 *
3248 * So if we aren't down to zero, we'll just schedule a worker and try
3249 * again
3250 */
3251 if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
3252 kmem_cache_shrink(cachep);
3253 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
3254 return;
3255 } else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003256 kmem_cache_destroy(cachep);
3257}
3258
3259void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3260{
3261 if (!cachep->memcg_params->dead)
3262 return;
3263
3264 /*
Glauber Costa22933152012-12-18 14:22:59 -08003265 * There are many ways in which we can get here.
3266 *
3267 * We can get to a memory-pressure situation while the delayed work is
3268 * still pending to run. The vmscan shrinkers can then release all
3269 * cache memory and get us to destruction. If this is the case, we'll
3270 * be executed twice, which is a bug (the second time will execute over
3271 * bogus data). In this case, cancelling the work should be fine.
3272 *
3273 * But we can also get here from the worker itself, if
3274 * kmem_cache_shrink is enough to shake all the remaining objects and
3275 * get the page count to 0. In this case, we'll deadlock if we try to
3276 * cancel the work (the worker runs with an internal lock held, which
3277 * is the same lock we would hold for cancel_work_sync().)
3278 *
3279 * Since we can't possibly know who got us here, just refrain from
3280 * running if there is already work pending
3281 */
3282 if (work_pending(&cachep->memcg_params->destroy))
3283 return;
3284 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003285 * We have to defer the actual destroying to a workqueue, because
3286 * we might currently be in a context that cannot sleep.
3287 */
3288 schedule_work(&cachep->memcg_params->destroy);
3289}
3290
Glauber Costad7f25f82012-12-18 14:22:40 -08003291/*
3292 * This lock protects updaters, not readers. We want readers to be as fast as
3293 * they can, and they will either see NULL or a valid cache value. Our model
3294 * allow them to see NULL, in which case the root memcg will be selected.
3295 *
3296 * We need this lock because multiple allocations to the same cache from a non
3297 * will span more than one worker. Only one of them can create the cache.
3298 */
3299static DEFINE_MUTEX(memcg_cache_mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003300
3301/*
3302 * Called with memcg_cache_mutex held
3303 */
3304static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
3305 struct kmem_cache *s)
3306{
3307 struct kmem_cache *new;
3308 static char *tmp_name = NULL;
3309
3310 lockdep_assert_held(&memcg_cache_mutex);
3311
3312 /*
3313 * kmem_cache_create_memcg duplicates the given name and
3314 * cgroup_name for this name requires RCU context.
3315 * This static temporary buffer is used to prevent from
3316 * pointless shortliving allocation.
3317 */
3318 if (!tmp_name) {
3319 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3320 if (!tmp_name)
3321 return NULL;
3322 }
3323
3324 rcu_read_lock();
3325 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3326 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3327 rcu_read_unlock();
3328
3329 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3330 (s->flags & ~SLAB_PANIC), s->ctor, s);
3331
3332 if (new)
3333 new->allocflags |= __GFP_KMEMCG;
3334
3335 return new;
3336}
3337
Glauber Costad7f25f82012-12-18 14:22:40 -08003338static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3339 struct kmem_cache *cachep)
3340{
3341 struct kmem_cache *new_cachep;
3342 int idx;
3343
3344 BUG_ON(!memcg_can_account_kmem(memcg));
3345
3346 idx = memcg_cache_id(memcg);
3347
3348 mutex_lock(&memcg_cache_mutex);
3349 new_cachep = cachep->memcg_params->memcg_caches[idx];
Li Zefan20f05312013-07-08 16:00:31 -07003350 if (new_cachep) {
3351 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003352 goto out;
Li Zefan20f05312013-07-08 16:00:31 -07003353 }
Glauber Costad7f25f82012-12-18 14:22:40 -08003354
3355 new_cachep = kmem_cache_dup(memcg, cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003356 if (new_cachep == NULL) {
3357 new_cachep = cachep;
Li Zefan20f05312013-07-08 16:00:31 -07003358 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003359 goto out;
3360 }
3361
Glauber Costa1f458cb2012-12-18 14:22:50 -08003362 atomic_set(&new_cachep->memcg_params->nr_pages , 0);
Glauber Costad7f25f82012-12-18 14:22:40 -08003363
3364 cachep->memcg_params->memcg_caches[idx] = new_cachep;
3365 /*
3366 * the readers won't lock, make sure everybody sees the updated value,
3367 * so they won't put stuff in the queue again for no reason
3368 */
3369 wmb();
3370out:
3371 mutex_unlock(&memcg_cache_mutex);
3372 return new_cachep;
3373}
3374
Glauber Costa7cf27982012-12-18 14:22:55 -08003375void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3376{
3377 struct kmem_cache *c;
3378 int i;
3379
3380 if (!s->memcg_params)
3381 return;
3382 if (!s->memcg_params->is_root_cache)
3383 return;
3384
3385 /*
3386 * If the cache is being destroyed, we trust that there is no one else
3387 * requesting objects from it. Even if there are, the sanity checks in
3388 * kmem_cache_destroy should caught this ill-case.
3389 *
3390 * Still, we don't want anyone else freeing memcg_caches under our
3391 * noses, which can happen if a new memcg comes to life. As usual,
3392 * we'll take the set_limit_mutex to protect ourselves against this.
3393 */
3394 mutex_lock(&set_limit_mutex);
3395 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3396 c = s->memcg_params->memcg_caches[i];
3397 if (!c)
3398 continue;
3399
3400 /*
3401 * We will now manually delete the caches, so to avoid races
3402 * we need to cancel all pending destruction workers and
3403 * proceed with destruction ourselves.
3404 *
3405 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3406 * and that could spawn the workers again: it is likely that
3407 * the cache still have active pages until this very moment.
3408 * This would lead us back to mem_cgroup_destroy_cache.
3409 *
3410 * But that will not execute at all if the "dead" flag is not
3411 * set, so flip it down to guarantee we are in control.
3412 */
3413 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003414 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003415 kmem_cache_destroy(c);
3416 }
3417 mutex_unlock(&set_limit_mutex);
3418}
3419
Glauber Costad7f25f82012-12-18 14:22:40 -08003420struct create_work {
3421 struct mem_cgroup *memcg;
3422 struct kmem_cache *cachep;
3423 struct work_struct work;
3424};
3425
Glauber Costa1f458cb2012-12-18 14:22:50 -08003426static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3427{
3428 struct kmem_cache *cachep;
3429 struct memcg_cache_params *params;
3430
3431 if (!memcg_kmem_is_active(memcg))
3432 return;
3433
3434 mutex_lock(&memcg->slab_caches_mutex);
3435 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3436 cachep = memcg_params_to_cache(params);
3437 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003438 schedule_work(&cachep->memcg_params->destroy);
3439 }
3440 mutex_unlock(&memcg->slab_caches_mutex);
3441}
3442
Glauber Costad7f25f82012-12-18 14:22:40 -08003443static void memcg_create_cache_work_func(struct work_struct *w)
3444{
3445 struct create_work *cw;
3446
3447 cw = container_of(w, struct create_work, work);
3448 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003449 kfree(cw);
3450}
3451
3452/*
3453 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003454 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003455static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3456 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003457{
3458 struct create_work *cw;
3459
3460 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003461 if (cw == NULL) {
3462 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003463 return;
3464 }
3465
3466 cw->memcg = memcg;
3467 cw->cachep = cachep;
3468
3469 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3470 schedule_work(&cw->work);
3471}
3472
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003473static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3474 struct kmem_cache *cachep)
3475{
3476 /*
3477 * We need to stop accounting when we kmalloc, because if the
3478 * corresponding kmalloc cache is not yet created, the first allocation
3479 * in __memcg_create_cache_enqueue will recurse.
3480 *
3481 * However, it is better to enclose the whole function. Depending on
3482 * the debugging options enabled, INIT_WORK(), for instance, can
3483 * trigger an allocation. This too, will make us recurse. Because at
3484 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3485 * the safest choice is to do it like this, wrapping the whole function.
3486 */
3487 memcg_stop_kmem_account();
3488 __memcg_create_cache_enqueue(memcg, cachep);
3489 memcg_resume_kmem_account();
3490}
Glauber Costad7f25f82012-12-18 14:22:40 -08003491/*
3492 * Return the kmem_cache we're supposed to use for a slab allocation.
3493 * We try to use the current memcg's version of the cache.
3494 *
3495 * If the cache does not exist yet, if we are the first user of it,
3496 * we either create it immediately, if possible, or create it asynchronously
3497 * in a workqueue.
3498 * In the latter case, we will let the current allocation go through with
3499 * the original cache.
3500 *
3501 * Can't be called in interrupt context or from kernel threads.
3502 * This function needs to be called with rcu_read_lock() held.
3503 */
3504struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3505 gfp_t gfp)
3506{
3507 struct mem_cgroup *memcg;
3508 int idx;
3509
3510 VM_BUG_ON(!cachep->memcg_params);
3511 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3512
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003513 if (!current->mm || current->memcg_kmem_skip_account)
3514 return cachep;
3515
Glauber Costad7f25f82012-12-18 14:22:40 -08003516 rcu_read_lock();
3517 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003518
3519 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003520 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003521
3522 idx = memcg_cache_id(memcg);
3523
3524 /*
3525 * barrier to mare sure we're always seeing the up to date value. The
3526 * code updating memcg_caches will issue a write barrier to match this.
3527 */
3528 read_barrier_depends();
Li Zefanca0dde92013-04-29 15:08:57 -07003529 if (likely(cachep->memcg_params->memcg_caches[idx])) {
3530 cachep = cachep->memcg_params->memcg_caches[idx];
3531 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003532 }
3533
Li Zefanca0dde92013-04-29 15:08:57 -07003534 /* The corresponding put will be done in the workqueue. */
3535 if (!css_tryget(&memcg->css))
3536 goto out;
3537 rcu_read_unlock();
3538
3539 /*
3540 * If we are in a safe context (can wait, and not in interrupt
3541 * context), we could be be predictable and return right away.
3542 * This would guarantee that the allocation being performed
3543 * already belongs in the new cache.
3544 *
3545 * However, there are some clashes that can arrive from locking.
3546 * For instance, because we acquire the slab_mutex while doing
3547 * kmem_cache_dup, this means no further allocation could happen
3548 * with the slab_mutex held.
3549 *
3550 * Also, because cache creation issue get_online_cpus(), this
3551 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3552 * that ends up reversed during cpu hotplug. (cpuset allocates
3553 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3554 * better to defer everything.
3555 */
3556 memcg_create_cache_enqueue(memcg, cachep);
3557 return cachep;
3558out:
3559 rcu_read_unlock();
3560 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003561}
3562EXPORT_SYMBOL(__memcg_kmem_get_cache);
3563
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003564/*
3565 * We need to verify if the allocation against current->mm->owner's memcg is
3566 * possible for the given order. But the page is not allocated yet, so we'll
3567 * need a further commit step to do the final arrangements.
3568 *
3569 * It is possible for the task to switch cgroups in this mean time, so at
3570 * commit time, we can't rely on task conversion any longer. We'll then use
3571 * the handle argument to return to the caller which cgroup we should commit
3572 * against. We could also return the memcg directly and avoid the pointer
3573 * passing, but a boolean return value gives better semantics considering
3574 * the compiled-out case as well.
3575 *
3576 * Returning true means the allocation is possible.
3577 */
3578bool
3579__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3580{
3581 struct mem_cgroup *memcg;
3582 int ret;
3583
3584 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003585
3586 /*
3587 * Disabling accounting is only relevant for some specific memcg
3588 * internal allocations. Therefore we would initially not have such
3589 * check here, since direct calls to the page allocator that are marked
3590 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3591 * concerned with cache allocations, and by having this test at
3592 * memcg_kmem_get_cache, we are already able to relay the allocation to
3593 * the root cache and bypass the memcg cache altogether.
3594 *
3595 * There is one exception, though: the SLUB allocator does not create
3596 * large order caches, but rather service large kmallocs directly from
3597 * the page allocator. Therefore, the following sequence when backed by
3598 * the SLUB allocator:
3599 *
3600 * memcg_stop_kmem_account();
3601 * kmalloc(<large_number>)
3602 * memcg_resume_kmem_account();
3603 *
3604 * would effectively ignore the fact that we should skip accounting,
3605 * since it will drive us directly to this function without passing
3606 * through the cache selector memcg_kmem_get_cache. Such large
3607 * allocations are extremely rare but can happen, for instance, for the
3608 * cache arrays. We bring this test here.
3609 */
3610 if (!current->mm || current->memcg_kmem_skip_account)
3611 return true;
3612
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003613 memcg = try_get_mem_cgroup_from_mm(current->mm);
3614
3615 /*
3616 * very rare case described in mem_cgroup_from_task. Unfortunately there
3617 * isn't much we can do without complicating this too much, and it would
3618 * be gfp-dependent anyway. Just let it go
3619 */
3620 if (unlikely(!memcg))
3621 return true;
3622
3623 if (!memcg_can_account_kmem(memcg)) {
3624 css_put(&memcg->css);
3625 return true;
3626 }
3627
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003628 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3629 if (!ret)
3630 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003631
3632 css_put(&memcg->css);
3633 return (ret == 0);
3634}
3635
3636void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3637 int order)
3638{
3639 struct page_cgroup *pc;
3640
3641 VM_BUG_ON(mem_cgroup_is_root(memcg));
3642
3643 /* The page allocation failed. Revert */
3644 if (!page) {
3645 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003646 return;
3647 }
3648
3649 pc = lookup_page_cgroup(page);
3650 lock_page_cgroup(pc);
3651 pc->mem_cgroup = memcg;
3652 SetPageCgroupUsed(pc);
3653 unlock_page_cgroup(pc);
3654}
3655
3656void __memcg_kmem_uncharge_pages(struct page *page, int order)
3657{
3658 struct mem_cgroup *memcg = NULL;
3659 struct page_cgroup *pc;
3660
3661
3662 pc = lookup_page_cgroup(page);
3663 /*
3664 * Fast unlocked return. Theoretically might have changed, have to
3665 * check again after locking.
3666 */
3667 if (!PageCgroupUsed(pc))
3668 return;
3669
3670 lock_page_cgroup(pc);
3671 if (PageCgroupUsed(pc)) {
3672 memcg = pc->mem_cgroup;
3673 ClearPageCgroupUsed(pc);
3674 }
3675 unlock_page_cgroup(pc);
3676
3677 /*
3678 * We trust that only if there is a memcg associated with the page, it
3679 * is a valid allocation
3680 */
3681 if (!memcg)
3682 return;
3683
3684 VM_BUG_ON(mem_cgroup_is_root(memcg));
3685 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003686}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003687#else
3688static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3689{
3690}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003691#endif /* CONFIG_MEMCG_KMEM */
3692
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003693#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3694
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003695#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003696/*
3697 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003698 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3699 * charge/uncharge will be never happen and move_account() is done under
3700 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003701 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003702void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003703{
3704 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003705 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003706 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003707 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003708
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003709 if (mem_cgroup_disabled())
3710 return;
David Rientjesb070e652013-05-07 16:18:09 -07003711
3712 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003713 for (i = 1; i < HPAGE_PMD_NR; i++) {
3714 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003715 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003716 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003717 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3718 }
David Rientjesb070e652013-05-07 16:18:09 -07003719 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3720 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003721}
Hugh Dickins12d27102012-01-12 17:19:52 -08003722#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003723
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003724/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003725 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003726 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003727 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003728 * @pc: page_cgroup of the page.
3729 * @from: mem_cgroup which the page is moved from.
3730 * @to: mem_cgroup which the page is moved to. @from != @to.
3731 *
3732 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003733 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003734 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003735 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003736 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3737 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003738 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003739static int mem_cgroup_move_account(struct page *page,
3740 unsigned int nr_pages,
3741 struct page_cgroup *pc,
3742 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003743 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003744{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003745 unsigned long flags;
3746 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003747 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003748
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003749 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07003750 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07003751 /*
3752 * The page is isolated from LRU. So, collapse function
3753 * will not handle this page. But page splitting can happen.
3754 * Do this check under compound_page_lock(). The caller should
3755 * hold it.
3756 */
3757 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003758 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003759 goto out;
3760
3761 lock_page_cgroup(pc);
3762
3763 ret = -EINVAL;
3764 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3765 goto unlock;
3766
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003767 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003768
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07003769 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003770 /* Update mapped_file data for mem_cgroup */
3771 preempt_disable();
3772 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3773 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3774 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07003775 }
David Rientjesb070e652013-05-07 16:18:09 -07003776 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003777
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003778 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003779 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003780 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003781 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003782 ret = 0;
3783unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003784 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003785 /*
3786 * check events
3787 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003788 memcg_check_events(to, page);
3789 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003790out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003791 return ret;
3792}
3793
Michal Hocko2ef37d32012-10-26 13:37:30 +02003794/**
3795 * mem_cgroup_move_parent - moves page to the parent group
3796 * @page: the page to move
3797 * @pc: page_cgroup of the page
3798 * @child: page's cgroup
3799 *
3800 * move charges to its parent or the root cgroup if the group has no
3801 * parent (aka use_hierarchy==0).
3802 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3803 * mem_cgroup_move_account fails) the failure is always temporary and
3804 * it signals a race with a page removal/uncharge or migration. In the
3805 * first case the page is on the way out and it will vanish from the LRU
3806 * on the next attempt and the call should be retried later.
3807 * Isolation from the LRU fails only if page has been isolated from
3808 * the LRU since we looked at it and that usually means either global
3809 * reclaim or migration going on. The page will either get back to the
3810 * LRU or vanish.
3811 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3812 * (!PageCgroupUsed) or moved to a different group. The page will
3813 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003814 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003815static int mem_cgroup_move_parent(struct page *page,
3816 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003817 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003818{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003819 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003820 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003821 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003822 int ret;
3823
Michal Hockod8423012012-10-26 13:37:29 +02003824 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003825
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003826 ret = -EBUSY;
3827 if (!get_page_unless_zero(page))
3828 goto out;
3829 if (isolate_lru_page(page))
3830 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003831
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003832 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003833
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003834 parent = parent_mem_cgroup(child);
3835 /*
3836 * If no parent, move charges to root cgroup.
3837 */
3838 if (!parent)
3839 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003840
Michal Hocko2ef37d32012-10-26 13:37:30 +02003841 if (nr_pages > 1) {
3842 VM_BUG_ON(!PageTransHuge(page));
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003843 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003844 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003845
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003846 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003847 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003848 if (!ret)
3849 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003850
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003851 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003852 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003853 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003854put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003855 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003856out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003857 return ret;
3858}
3859
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003860/*
3861 * Charge the memory controller for page usage.
3862 * Return
3863 * 0 if the charge was successful
3864 * < 0 if the cgroup is over its limit
3865 */
3866static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003867 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003868{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003869 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003870 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003871 bool oom = true;
3872 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003873
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003874 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003875 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003876 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08003877 /*
3878 * Never OOM-kill a process for a huge page. The
3879 * fault handler will fall back to regular pages.
3880 */
3881 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003882 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003883
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003884 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003885 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003886 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003887 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003888 return 0;
3889}
3890
3891int mem_cgroup_newpage_charge(struct page *page,
3892 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003893{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003894 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003895 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003896 VM_BUG_ON(page_mapped(page));
3897 VM_BUG_ON(page->mapping && !PageAnon(page));
3898 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003899 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003900 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003901}
3902
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003903/*
3904 * While swap-in, try_charge -> commit or cancel, the page is locked.
3905 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003906 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003907 * "commit()" or removed by "cancel()"
3908 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003909static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3910 struct page *page,
3911 gfp_t mask,
3912 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003913{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003914 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003915 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003916 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003917
Johannes Weiner90deb782012-07-31 16:45:47 -07003918 pc = lookup_page_cgroup(page);
3919 /*
3920 * Every swap fault against a single page tries to charge the
3921 * page, bail as early as possible. shmem_unuse() encounters
3922 * already charged pages, too. The USED bit is protected by
3923 * the page lock, which serializes swap cache removal, which
3924 * in turn serializes uncharging.
3925 */
3926 if (PageCgroupUsed(pc))
3927 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003928 if (!do_swap_account)
3929 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003930 memcg = try_get_mem_cgroup_from_page(page);
3931 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003932 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08003933 *memcgp = memcg;
3934 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003935 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003936 if (ret == -EINTR)
3937 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003938 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003939charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003940 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
3941 if (ret == -EINTR)
3942 ret = 0;
3943 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003944}
3945
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003946int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
3947 gfp_t gfp_mask, struct mem_cgroup **memcgp)
3948{
3949 *memcgp = NULL;
3950 if (mem_cgroup_disabled())
3951 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003952 /*
3953 * A racing thread's fault, or swapoff, may have already
3954 * updated the pte, and even removed page from swap cache: in
3955 * those cases unuse_pte()'s pte_same() test will fail; but
3956 * there's also a KSM case which does need to charge the page.
3957 */
3958 if (!PageSwapCache(page)) {
3959 int ret;
3960
3961 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
3962 if (ret == -EINTR)
3963 ret = 0;
3964 return ret;
3965 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003966 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
3967}
3968
Johannes Weiner827a03d2012-07-31 16:45:36 -07003969void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
3970{
3971 if (mem_cgroup_disabled())
3972 return;
3973 if (!memcg)
3974 return;
3975 __mem_cgroup_cancel_charge(memcg, 1);
3976}
3977
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003978static void
Johannes Weiner72835c82012-01-12 17:18:32 -08003979__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07003980 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003981{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003982 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003983 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08003984 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003985 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07003986
Johannes Weinerce587e62012-04-24 20:22:33 +02003987 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003988 /*
3989 * Now swap is on-memory. This means this page may be
3990 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003991 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
3992 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
3993 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003994 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003995 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003996 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07003997 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003998 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003999}
4000
Johannes Weiner72835c82012-01-12 17:18:32 -08004001void mem_cgroup_commit_charge_swapin(struct page *page,
4002 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004003{
Johannes Weiner72835c82012-01-12 17:18:32 -08004004 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004005 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004006}
4007
Johannes Weiner827a03d2012-07-31 16:45:36 -07004008int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4009 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004010{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004011 struct mem_cgroup *memcg = NULL;
4012 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4013 int ret;
4014
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004015 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004016 return 0;
4017 if (PageCompound(page))
4018 return 0;
4019
Johannes Weiner827a03d2012-07-31 16:45:36 -07004020 if (!PageSwapCache(page))
4021 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4022 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004023 ret = __mem_cgroup_try_charge_swapin(mm, page,
4024 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004025 if (!ret)
4026 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4027 }
4028 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004029}
4030
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004031static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004032 unsigned int nr_pages,
4033 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004034{
4035 struct memcg_batch_info *batch = NULL;
4036 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004037
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004038 /* If swapout, usage of swap doesn't decrease */
4039 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4040 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004041
4042 batch = &current->memcg_batch;
4043 /*
4044 * In usual, we do css_get() when we remember memcg pointer.
4045 * But in this case, we keep res->usage until end of a series of
4046 * uncharges. Then, it's ok to ignore memcg's refcnt.
4047 */
4048 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004049 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004050 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004051 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004052 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004053 * the same cgroup and we have chance to coalesce uncharges.
4054 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4055 * because we want to do uncharge as soon as possible.
4056 */
4057
4058 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4059 goto direct_uncharge;
4060
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004061 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004062 goto direct_uncharge;
4063
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004064 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004065 * In typical case, batch->memcg == mem. This means we can
4066 * merge a series of uncharges to an uncharge of res_counter.
4067 * If not, we uncharge res_counter ony by one.
4068 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004069 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004070 goto direct_uncharge;
4071 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004072 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004073 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004074 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004075 return;
4076direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004077 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004078 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004079 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4080 if (unlikely(batch->memcg != memcg))
4081 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004082}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004083
Balbir Singh8697d332008-02-07 00:13:59 -08004084/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004085 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004086 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004087static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004088__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4089 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004090{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004091 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004092 unsigned int nr_pages = 1;
4093 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004094 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004095
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004096 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004097 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004098
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004099 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004100 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004101 VM_BUG_ON(!PageTransHuge(page));
4102 }
Balbir Singh8697d332008-02-07 00:13:59 -08004103 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004104 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004105 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004106 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004107 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004108 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004109
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004110 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004111
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004112 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004113
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004114 if (!PageCgroupUsed(pc))
4115 goto unlock_out;
4116
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004117 anon = PageAnon(page);
4118
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004119 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004120 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004121 /*
4122 * Generally PageAnon tells if it's the anon statistics to be
4123 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4124 * used before page reached the stage of being marked PageAnon.
4125 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004126 anon = true;
4127 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004128 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004129 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004130 if (page_mapped(page))
4131 goto unlock_out;
4132 /*
4133 * Pages under migration may not be uncharged. But
4134 * end_migration() /must/ be the one uncharging the
4135 * unused post-migration page and so it has to call
4136 * here with the migration bit still set. See the
4137 * res_counter handling below.
4138 */
4139 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004140 goto unlock_out;
4141 break;
4142 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4143 if (!PageAnon(page)) { /* Shared memory */
4144 if (page->mapping && !page_is_file_cache(page))
4145 goto unlock_out;
4146 } else if (page_mapped(page)) /* Anon */
4147 goto unlock_out;
4148 break;
4149 default:
4150 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004151 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004152
David Rientjesb070e652013-05-07 16:18:09 -07004153 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004154
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004155 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004156 /*
4157 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4158 * freed from LRU. This is safe because uncharged page is expected not
4159 * to be reused (freed soon). Exception is SwapCache, it's handled by
4160 * special functions.
4161 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004162
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004163 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004164 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004165 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004166 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004167 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004168 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004169 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004170 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004171 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004172 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004173 /*
4174 * Migration does not charge the res_counter for the
4175 * replacement page, so leave it alone when phasing out the
4176 * page that is unused after the migration.
4177 */
4178 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004179 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004180
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004181 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004182
4183unlock_out:
4184 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004185 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004186}
4187
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004188void mem_cgroup_uncharge_page(struct page *page)
4189{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004190 /* early check. */
4191 if (page_mapped(page))
4192 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08004193 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004194 /*
4195 * If the page is in swap cache, uncharge should be deferred
4196 * to the swap path, which also properly accounts swap usage
4197 * and handles memcg lifetime.
4198 *
4199 * Note that this check is not stable and reclaim may add the
4200 * page to swap cache at any time after this. However, if the
4201 * page is not in swap cache by the time page->mapcount hits
4202 * 0, there won't be any page table references to the swap
4203 * slot, and reclaim will free it and not actually write the
4204 * page to disk.
4205 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004206 if (PageSwapCache(page))
4207 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004208 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004209}
4210
4211void mem_cgroup_uncharge_cache_page(struct page *page)
4212{
4213 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07004214 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07004215 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004216}
4217
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004218/*
4219 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4220 * In that cases, pages are freed continuously and we can expect pages
4221 * are in the same memcg. All these calls itself limits the number of
4222 * pages freed at once, then uncharge_start/end() is called properly.
4223 * This may be called prural(2) times in a context,
4224 */
4225
4226void mem_cgroup_uncharge_start(void)
4227{
4228 current->memcg_batch.do_batch++;
4229 /* We can do nest. */
4230 if (current->memcg_batch.do_batch == 1) {
4231 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004232 current->memcg_batch.nr_pages = 0;
4233 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004234 }
4235}
4236
4237void mem_cgroup_uncharge_end(void)
4238{
4239 struct memcg_batch_info *batch = &current->memcg_batch;
4240
4241 if (!batch->do_batch)
4242 return;
4243
4244 batch->do_batch--;
4245 if (batch->do_batch) /* If stacked, do nothing. */
4246 return;
4247
4248 if (!batch->memcg)
4249 return;
4250 /*
4251 * This "batch->memcg" is valid without any css_get/put etc...
4252 * bacause we hide charges behind us.
4253 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004254 if (batch->nr_pages)
4255 res_counter_uncharge(&batch->memcg->res,
4256 batch->nr_pages * PAGE_SIZE);
4257 if (batch->memsw_nr_pages)
4258 res_counter_uncharge(&batch->memcg->memsw,
4259 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004260 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004261 /* forget this pointer (for sanity check) */
4262 batch->memcg = NULL;
4263}
4264
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004265#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004266/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004267 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004268 * memcg information is recorded to swap_cgroup of "ent"
4269 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004270void
4271mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004272{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004273 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004274 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004275
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004276 if (!swapout) /* this was a swap cache but the swap is unused ! */
4277 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4278
Johannes Weiner0030f532012-07-31 16:45:25 -07004279 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004280
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004281 /*
4282 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004283 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004284 */
4285 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004286 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004287}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004288#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004289
Andrew Mortonc255a452012-07-31 16:43:02 -07004290#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004291/*
4292 * called from swap_entry_free(). remove record in swap_cgroup and
4293 * uncharge "memsw" account.
4294 */
4295void mem_cgroup_uncharge_swap(swp_entry_t ent)
4296{
4297 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004298 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004299
4300 if (!do_swap_account)
4301 return;
4302
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004303 id = swap_cgroup_record(ent, 0);
4304 rcu_read_lock();
4305 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004306 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004307 /*
4308 * We uncharge this because swap is freed.
4309 * This memcg can be obsolete one. We avoid calling css_tryget
4310 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004311 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004312 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004313 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004314 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004315 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004316 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004317}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004318
4319/**
4320 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4321 * @entry: swap entry to be moved
4322 * @from: mem_cgroup which the entry is moved from
4323 * @to: mem_cgroup which the entry is moved to
4324 *
4325 * It succeeds only when the swap_cgroup's record for this entry is the same
4326 * as the mem_cgroup's id of @from.
4327 *
4328 * Returns 0 on success, -EINVAL on failure.
4329 *
4330 * The caller must have charged to @to, IOW, called res_counter_charge() about
4331 * both res and memsw, and called css_get().
4332 */
4333static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004334 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004335{
4336 unsigned short old_id, new_id;
4337
4338 old_id = css_id(&from->css);
4339 new_id = css_id(&to->css);
4340
4341 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004342 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004343 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004344 /*
4345 * This function is only called from task migration context now.
4346 * It postpones res_counter and refcount handling till the end
4347 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004348 * improvement. But we cannot postpone css_get(to) because if
4349 * the process that has been moved to @to does swap-in, the
4350 * refcount of @to might be decreased to 0.
4351 *
4352 * We are in attach() phase, so the cgroup is guaranteed to be
4353 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004354 */
Li Zefan40503772013-07-08 16:00:34 -07004355 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004356 return 0;
4357 }
4358 return -EINVAL;
4359}
4360#else
4361static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004362 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004363{
4364 return -EINVAL;
4365}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004366#endif
4367
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004368/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004369 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4370 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004371 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004372void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4373 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004374{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004375 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004376 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004377 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004378 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004379
Johannes Weiner72835c82012-01-12 17:18:32 -08004380 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004381
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004382 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004383 return;
Balbir Singh40779602008-04-04 14:29:59 -07004384
Mel Gormanb32967f2012-11-19 12:35:47 +00004385 if (PageTransHuge(page))
4386 nr_pages <<= compound_order(page);
4387
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004388 pc = lookup_page_cgroup(page);
4389 lock_page_cgroup(pc);
4390 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004391 memcg = pc->mem_cgroup;
4392 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004393 /*
4394 * At migrating an anonymous page, its mapcount goes down
4395 * to 0 and uncharge() will be called. But, even if it's fully
4396 * unmapped, migration may fail and this page has to be
4397 * charged again. We set MIGRATION flag here and delay uncharge
4398 * until end_migration() is called
4399 *
4400 * Corner Case Thinking
4401 * A)
4402 * When the old page was mapped as Anon and it's unmap-and-freed
4403 * while migration was ongoing.
4404 * If unmap finds the old page, uncharge() of it will be delayed
4405 * until end_migration(). If unmap finds a new page, it's
4406 * uncharged when it make mapcount to be 1->0. If unmap code
4407 * finds swap_migration_entry, the new page will not be mapped
4408 * and end_migration() will find it(mapcount==0).
4409 *
4410 * B)
4411 * When the old page was mapped but migraion fails, the kernel
4412 * remaps it. A charge for it is kept by MIGRATION flag even
4413 * if mapcount goes down to 0. We can do remap successfully
4414 * without charging it again.
4415 *
4416 * C)
4417 * The "old" page is under lock_page() until the end of
4418 * migration, so, the old page itself will not be swapped-out.
4419 * If the new page is swapped out before end_migraton, our
4420 * hook to usual swap-out path will catch the event.
4421 */
4422 if (PageAnon(page))
4423 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004424 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004425 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004426 /*
4427 * If the page is not charged at this point,
4428 * we return here.
4429 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004430 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004431 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004432
Johannes Weiner72835c82012-01-12 17:18:32 -08004433 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004434 /*
4435 * We charge new page before it's used/mapped. So, even if unlock_page()
4436 * is called before end_migration, we can catch all events on this new
4437 * page. In the case new page is migrated but not remapped, new page's
4438 * mapcount will be finally 0 and we call uncharge in end_migration().
4439 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004440 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004441 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004442 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004443 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004444 /*
4445 * The page is committed to the memcg, but it's not actually
4446 * charged to the res_counter since we plan on replacing the
4447 * old one and only one page is going to be left afterwards.
4448 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004449 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004450}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004451
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004452/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004453void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004454 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004455{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004456 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004457 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004458 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004459
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004460 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004461 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004462
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004463 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004464 used = oldpage;
4465 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004466 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004467 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004468 unused = oldpage;
4469 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004470 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004471 __mem_cgroup_uncharge_common(unused,
4472 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4473 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4474 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004475 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004476 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004477 * We disallowed uncharge of pages under migration because mapcount
4478 * of the page goes down to zero, temporarly.
4479 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004480 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004481 pc = lookup_page_cgroup(oldpage);
4482 lock_page_cgroup(pc);
4483 ClearPageCgroupMigration(pc);
4484 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004485
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004486 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004487 * If a page is a file cache, radix-tree replacement is very atomic
4488 * and we can skip this check. When it was an Anon page, its mapcount
4489 * goes down to 0. But because we added MIGRATION flage, it's not
4490 * uncharged yet. There are several case but page->mapcount check
4491 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4492 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004493 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004494 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004495 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004496}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004497
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004498/*
4499 * At replace page cache, newpage is not under any memcg but it's on
4500 * LRU. So, this function doesn't touch res_counter but handles LRU
4501 * in correct way. Both pages are locked so we cannot race with uncharge.
4502 */
4503void mem_cgroup_replace_page_cache(struct page *oldpage,
4504 struct page *newpage)
4505{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004506 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004507 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004508 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004509
4510 if (mem_cgroup_disabled())
4511 return;
4512
4513 pc = lookup_page_cgroup(oldpage);
4514 /* fix accounting on old pages */
4515 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004516 if (PageCgroupUsed(pc)) {
4517 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004518 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004519 ClearPageCgroupUsed(pc);
4520 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004521 unlock_page_cgroup(pc);
4522
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004523 /*
4524 * When called from shmem_replace_page(), in some cases the
4525 * oldpage has already been charged, and in some cases not.
4526 */
4527 if (!memcg)
4528 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004529 /*
4530 * Even if newpage->mapping was NULL before starting replacement,
4531 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4532 * LRU while we overwrite pc->mem_cgroup.
4533 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004534 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004535}
4536
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004537#ifdef CONFIG_DEBUG_VM
4538static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4539{
4540 struct page_cgroup *pc;
4541
4542 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004543 /*
4544 * Can be NULL while feeding pages into the page allocator for
4545 * the first time, i.e. during boot or memory hotplug;
4546 * or when mem_cgroup_disabled().
4547 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004548 if (likely(pc) && PageCgroupUsed(pc))
4549 return pc;
4550 return NULL;
4551}
4552
4553bool mem_cgroup_bad_page_check(struct page *page)
4554{
4555 if (mem_cgroup_disabled())
4556 return false;
4557
4558 return lookup_page_cgroup_used(page) != NULL;
4559}
4560
4561void mem_cgroup_print_bad_page(struct page *page)
4562{
4563 struct page_cgroup *pc;
4564
4565 pc = lookup_page_cgroup_used(page);
4566 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004567 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4568 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004569 }
4570}
4571#endif
4572
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004573static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004574 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004575{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004576 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004577 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004578 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004579 int children = mem_cgroup_count_children(memcg);
4580 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004581 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004582
4583 /*
4584 * For keeping hierarchical_reclaim simple, how long we should retry
4585 * is depends on callers. We set our retry-count to be function
4586 * of # of children which we should visit in this loop.
4587 */
4588 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4589
4590 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004591
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004592 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004593 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004594 if (signal_pending(current)) {
4595 ret = -EINTR;
4596 break;
4597 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004598 /*
4599 * Rather than hide all in some function, I do this in
4600 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004601 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004602 */
4603 mutex_lock(&set_limit_mutex);
4604 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4605 if (memswlimit < val) {
4606 ret = -EINVAL;
4607 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004608 break;
4609 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004610
4611 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4612 if (memlimit < val)
4613 enlarge = 1;
4614
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004615 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004616 if (!ret) {
4617 if (memswlimit == val)
4618 memcg->memsw_is_minimum = true;
4619 else
4620 memcg->memsw_is_minimum = false;
4621 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004622 mutex_unlock(&set_limit_mutex);
4623
4624 if (!ret)
4625 break;
4626
Johannes Weiner56600482012-01-12 17:17:59 -08004627 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4628 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004629 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4630 /* Usage is reduced ? */
4631 if (curusage >= oldusage)
4632 retry_count--;
4633 else
4634 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004635 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004636 if (!ret && enlarge)
4637 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004638
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004639 return ret;
4640}
4641
Li Zefan338c8432009-06-17 16:27:15 -07004642static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4643 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004644{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004645 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004646 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004647 int children = mem_cgroup_count_children(memcg);
4648 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004649 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004650
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004651 /* see mem_cgroup_resize_res_limit */
4652 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
4653 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004654 while (retry_count) {
4655 if (signal_pending(current)) {
4656 ret = -EINTR;
4657 break;
4658 }
4659 /*
4660 * Rather than hide all in some function, I do this in
4661 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004662 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004663 */
4664 mutex_lock(&set_limit_mutex);
4665 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4666 if (memlimit > val) {
4667 ret = -EINVAL;
4668 mutex_unlock(&set_limit_mutex);
4669 break;
4670 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004671 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4672 if (memswlimit < val)
4673 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004674 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004675 if (!ret) {
4676 if (memlimit == val)
4677 memcg->memsw_is_minimum = true;
4678 else
4679 memcg->memsw_is_minimum = false;
4680 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004681 mutex_unlock(&set_limit_mutex);
4682
4683 if (!ret)
4684 break;
4685
Johannes Weiner56600482012-01-12 17:17:59 -08004686 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4687 MEM_CGROUP_RECLAIM_NOSWAP |
4688 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004689 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004690 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004691 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004692 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004693 else
4694 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004695 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004696 if (!ret && enlarge)
4697 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004698 return ret;
4699}
4700
Michal Hocko2ef37d32012-10-26 13:37:30 +02004701/**
4702 * mem_cgroup_force_empty_list - clears LRU of a group
4703 * @memcg: group to clear
4704 * @node: NUMA node
4705 * @zid: zone id
4706 * @lru: lru to to clear
4707 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004708 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004709 * reclaim the pages page themselves - pages are moved to the parent (or root)
4710 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004711 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004712static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004713 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004714{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004715 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004716 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004717 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004718 struct page *busy;
4719 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004720
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004721 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004722 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4723 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004724
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004725 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004726 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004727 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004728 struct page *page;
4729
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004730 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004731 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004732 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004733 break;
4734 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004735 page = list_entry(list->prev, struct page, lru);
4736 if (busy == page) {
4737 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004738 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004739 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004740 continue;
4741 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004742 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004743
Johannes Weiner925b7672012-01-12 17:18:15 -08004744 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004745
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004746 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004747 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004748 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004749 cond_resched();
4750 } else
4751 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004752 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004753}
4754
4755/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004756 * make mem_cgroup's charge to be 0 if there is no task by moving
4757 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004758 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004759 *
4760 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004761 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004762static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004763{
Michal Hockoc26251f2012-10-26 13:37:28 +02004764 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004765 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004766
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004767 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004768 /* This is for making all *used* pages to be on LRU. */
4769 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004770 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004771 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004772 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004773 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab932012-03-21 16:34:19 -07004774 enum lru_list lru;
4775 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004776 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab932012-03-21 16:34:19 -07004777 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004778 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004779 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004780 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004781 mem_cgroup_end_move(memcg);
4782 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004783 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004784
Michal Hocko2ef37d32012-10-26 13:37:30 +02004785 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004786 * Kernel memory may not necessarily be trackable to a specific
4787 * process. So they are not migrated, and therefore we can't
4788 * expect their value to drop to 0 here.
4789 * Having res filled up with kmem only is enough.
4790 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004791 * This is a safety check because mem_cgroup_force_empty_list
4792 * could have raced with mem_cgroup_replace_page_cache callers
4793 * so the lru seemed empty but the page could have been added
4794 * right after the check. RES_USAGE should be safe as we always
4795 * charge before adding to the LRU.
4796 */
Glauber Costabea207c2012-12-18 14:22:11 -08004797 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4798 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4799 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004800}
4801
4802/*
Glauber Costab5f99b52013-02-22 16:34:53 -08004803 * This mainly exists for tests during the setting of set of use_hierarchy.
4804 * Since this is the very setting we are changing, the current hierarchy value
4805 * is meaningless
4806 */
4807static inline bool __memcg_has_children(struct mem_cgroup *memcg)
4808{
Tejun Heo492eb212013-08-08 20:11:25 -04004809 struct cgroup_subsys_state *pos;
Glauber Costab5f99b52013-02-22 16:34:53 -08004810
4811 /* bounce at first found */
Tejun Heo492eb212013-08-08 20:11:25 -04004812 css_for_each_child(pos, &memcg->css)
Glauber Costab5f99b52013-02-22 16:34:53 -08004813 return true;
4814 return false;
4815}
4816
4817/*
Glauber Costa09998212013-02-22 16:34:55 -08004818 * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed
4819 * to be already dead (as in mem_cgroup_force_empty, for instance). This is
Glauber Costab5f99b52013-02-22 16:34:53 -08004820 * from mem_cgroup_count_children(), in the sense that we don't really care how
4821 * many children we have; we only need to know if we have any. It also counts
4822 * any memcg without hierarchy as infertile.
4823 */
4824static inline bool memcg_has_children(struct mem_cgroup *memcg)
4825{
4826 return memcg->use_hierarchy && __memcg_has_children(memcg);
4827}
4828
4829/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004830 * Reclaims as many pages from the given memcg as possible and moves
4831 * the rest to the parent.
4832 *
4833 * Caller is responsible for holding css reference for memcg.
4834 */
4835static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4836{
4837 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4838 struct cgroup *cgrp = memcg->css.cgroup;
4839
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004840 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004841 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4842 return -EBUSY;
4843
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004844 /* we call try-to-free pages for make this cgroup empty */
4845 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004846 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004847 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004848 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004849
Michal Hockoc26251f2012-10-26 13:37:28 +02004850 if (signal_pending(current))
4851 return -EINTR;
4852
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004853 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07004854 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004855 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004856 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004857 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02004858 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004859 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004860
4861 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004862 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02004863 mem_cgroup_reparent_charges(memcg);
4864
4865 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004866}
4867
Tejun Heo182446d2013-08-08 20:11:24 -04004868static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
4869 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004870{
Tejun Heo182446d2013-08-08 20:11:24 -04004871 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02004872
Michal Hockod8423012012-10-26 13:37:29 +02004873 if (mem_cgroup_is_root(memcg))
4874 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07004875 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004876}
4877
Tejun Heo182446d2013-08-08 20:11:24 -04004878static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
4879 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004880{
Tejun Heo182446d2013-08-08 20:11:24 -04004881 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004882}
4883
Tejun Heo182446d2013-08-08 20:11:24 -04004884static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
4885 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08004886{
4887 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04004888 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04004889 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08004890
Glauber Costa09998212013-02-22 16:34:55 -08004891 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07004892
4893 if (memcg->use_hierarchy == val)
4894 goto out;
4895
Balbir Singh18f59ea2009-01-07 18:08:07 -08004896 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02004897 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08004898 * in the child subtrees. If it is unset, then the change can
4899 * occur, provided the current cgroup has no children.
4900 *
4901 * For the root cgroup, parent_mem is NULL, we allow value to be
4902 * set if there are no children.
4903 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004904 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08004905 (val == 1 || val == 0)) {
Glauber Costab5f99b52013-02-22 16:34:53 -08004906 if (!__memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004907 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004908 else
4909 retval = -EBUSY;
4910 } else
4911 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07004912
4913out:
Glauber Costa09998212013-02-22 16:34:55 -08004914 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004915
4916 return retval;
4917}
4918
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004919
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004920static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004921 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004922{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004923 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004924 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004925
Johannes Weiner7a159cc2011-03-23 16:42:38 -07004926 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004927 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004928 val += mem_cgroup_read_stat(iter, idx);
4929
4930 if (val < 0) /* race ? */
4931 val = 0;
4932 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004933}
4934
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004935static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004936{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004937 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004938
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004939 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004940 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00004941 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004942 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00004943 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004944 }
4945
David Rientjesb070e652013-05-07 16:18:09 -07004946 /*
4947 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
4948 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
4949 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004950 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
4951 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004952
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004953 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07004954 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004955
4956 return val << PAGE_SHIFT;
4957}
4958
Tejun Heo182446d2013-08-08 20:11:24 -04004959static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css,
4960 struct cftype *cft, struct file *file,
4961 char __user *buf, size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004962{
Tejun Heo182446d2013-08-08 20:11:24 -04004963 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heoaf36f902012-04-01 12:09:55 -07004964 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004965 u64 val;
Glauber Costa86ae53e2012-12-18 14:21:45 -08004966 int name, len;
4967 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004968
4969 type = MEMFILE_TYPE(cft->private);
4970 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07004971
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004972 switch (type) {
4973 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004974 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004975 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004976 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004977 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004978 break;
4979 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004980 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004981 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08004982 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004983 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004984 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08004985 case _KMEM:
4986 val = res_counter_read_u64(&memcg->kmem, name);
4987 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004988 default:
4989 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004990 }
Tejun Heoaf36f902012-04-01 12:09:55 -07004991
4992 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
4993 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004994}
Glauber Costa510fc4e2012-12-18 14:21:47 -08004995
Tejun Heo182446d2013-08-08 20:11:24 -04004996static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val)
Glauber Costa510fc4e2012-12-18 14:21:47 -08004997{
4998 int ret = -EINVAL;
4999#ifdef CONFIG_MEMCG_KMEM
Tejun Heo182446d2013-08-08 20:11:24 -04005000 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005001 /*
5002 * For simplicity, we won't allow this to be disabled. It also can't
5003 * be changed if the cgroup has children already, or if tasks had
5004 * already joined.
5005 *
5006 * If tasks join before we set the limit, a person looking at
5007 * kmem.usage_in_bytes will have no way to determine when it took
5008 * place, which makes the value quite meaningless.
5009 *
5010 * After it first became limited, changes in the value of the limit are
5011 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005012 */
Glauber Costa09998212013-02-22 16:34:55 -08005013 mutex_lock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005014 mutex_lock(&set_limit_mutex);
5015 if (!memcg->kmem_account_flags && val != RESOURCE_MAX) {
Tejun Heo182446d2013-08-08 20:11:24 -04005016 if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005017 ret = -EBUSY;
5018 goto out;
5019 }
5020 ret = res_counter_set_limit(&memcg->kmem, val);
5021 VM_BUG_ON(ret);
5022
Glauber Costa55007d82012-12-18 14:22:38 -08005023 ret = memcg_update_cache_sizes(memcg);
5024 if (ret) {
5025 res_counter_set_limit(&memcg->kmem, RESOURCE_MAX);
5026 goto out;
5027 }
Glauber Costa692e89a2013-02-22 16:34:56 -08005028 static_key_slow_inc(&memcg_kmem_enabled_key);
5029 /*
5030 * setting the active bit after the inc will guarantee no one
5031 * starts accounting before all call sites are patched
5032 */
5033 memcg_kmem_set_active(memcg);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005034 } else
5035 ret = res_counter_set_limit(&memcg->kmem, val);
5036out:
5037 mutex_unlock(&set_limit_mutex);
Glauber Costa09998212013-02-22 16:34:55 -08005038 mutex_unlock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005039#endif
5040 return ret;
5041}
5042
Hugh Dickins6d0439902013-02-22 16:35:50 -08005043#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -08005044static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005045{
Glauber Costa55007d82012-12-18 14:22:38 -08005046 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005047 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5048 if (!parent)
Glauber Costa55007d82012-12-18 14:22:38 -08005049 goto out;
5050
Glauber Costa510fc4e2012-12-18 14:21:47 -08005051 memcg->kmem_account_flags = parent->kmem_account_flags;
Glauber Costaa8964b92012-12-18 14:22:09 -08005052 /*
5053 * When that happen, we need to disable the static branch only on those
5054 * memcgs that enabled it. To achieve this, we would be forced to
5055 * complicate the code by keeping track of which memcgs were the ones
5056 * that actually enabled limits, and which ones got it from its
5057 * parents.
5058 *
5059 * It is a lot simpler just to do static_key_slow_inc() on every child
5060 * that is accounted.
5061 */
Glauber Costa55007d82012-12-18 14:22:38 -08005062 if (!memcg_kmem_is_active(memcg))
5063 goto out;
5064
5065 /*
Li Zefan10d5ebf2013-07-08 16:00:33 -07005066 * __mem_cgroup_free() will issue static_key_slow_dec() because this
5067 * memcg is active already. If the later initialization fails then the
5068 * cgroup core triggers the cleanup so we do not have to do it here.
Glauber Costa55007d82012-12-18 14:22:38 -08005069 */
Glauber Costa55007d82012-12-18 14:22:38 -08005070 static_key_slow_inc(&memcg_kmem_enabled_key);
5071
5072 mutex_lock(&set_limit_mutex);
Glauber Costa425c5982013-07-08 16:00:01 -07005073 memcg_stop_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005074 ret = memcg_update_cache_sizes(memcg);
Glauber Costa425c5982013-07-08 16:00:01 -07005075 memcg_resume_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005076 mutex_unlock(&set_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005077out:
5078 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005079}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005080#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005081
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005082/*
5083 * The user of this function is...
5084 * RES_LIMIT.
5085 */
Tejun Heo182446d2013-08-08 20:11:24 -04005086static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Paul Menage856c13a2008-07-25 01:47:04 -07005087 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005088{
Tejun Heo182446d2013-08-08 20:11:24 -04005089 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005090 enum res_type type;
5091 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005092 unsigned long long val;
5093 int ret;
5094
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005095 type = MEMFILE_TYPE(cft->private);
5096 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005097
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005098 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005099 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005100 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5101 ret = -EINVAL;
5102 break;
5103 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005104 /* This function does all necessary parse...reuse it */
5105 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005106 if (ret)
5107 break;
5108 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005109 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005110 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005111 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005112 else if (type == _KMEM)
Tejun Heo182446d2013-08-08 20:11:24 -04005113 ret = memcg_update_kmem_limit(css, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005114 else
5115 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005116 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005117 case RES_SOFT_LIMIT:
5118 ret = res_counter_memparse_write_strategy(buffer, &val);
5119 if (ret)
5120 break;
5121 /*
5122 * For memsw, soft limits are hard to implement in terms
5123 * of semantics, for now, we support soft limits for
5124 * control without swap
5125 */
5126 if (type == _MEM)
5127 ret = res_counter_set_soft_limit(&memcg->res, val);
5128 else
5129 ret = -EINVAL;
5130 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005131 default:
5132 ret = -EINVAL; /* should be BUG() ? */
5133 break;
5134 }
5135 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005136}
5137
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005138static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5139 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5140{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005141 unsigned long long min_limit, min_memsw_limit, tmp;
5142
5143 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5144 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005145 if (!memcg->use_hierarchy)
5146 goto out;
5147
Tejun Heo63876982013-08-08 20:11:23 -04005148 while (css_parent(&memcg->css)) {
5149 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005150 if (!memcg->use_hierarchy)
5151 break;
5152 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5153 min_limit = min(min_limit, tmp);
5154 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5155 min_memsw_limit = min(min_memsw_limit, tmp);
5156 }
5157out:
5158 *mem_limit = min_limit;
5159 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005160}
5161
Tejun Heo182446d2013-08-08 20:11:24 -04005162static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005163{
Tejun Heo182446d2013-08-08 20:11:24 -04005164 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005165 int name;
5166 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005167
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005168 type = MEMFILE_TYPE(event);
5169 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005170
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005171 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005172 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005173 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005174 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005175 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005176 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005177 else if (type == _KMEM)
5178 res_counter_reset_max(&memcg->kmem);
5179 else
5180 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005181 break;
5182 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005183 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005184 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005185 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005186 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005187 else if (type == _KMEM)
5188 res_counter_reset_failcnt(&memcg->kmem);
5189 else
5190 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005191 break;
5192 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005193
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005194 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005195}
5196
Tejun Heo182446d2013-08-08 20:11:24 -04005197static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005198 struct cftype *cft)
5199{
Tejun Heo182446d2013-08-08 20:11:24 -04005200 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005201}
5202
Daisuke Nishimura02491442010-03-10 15:22:17 -08005203#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005204static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005205 struct cftype *cft, u64 val)
5206{
Tejun Heo182446d2013-08-08 20:11:24 -04005207 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005208
5209 if (val >= (1 << NR_MOVE_TYPE))
5210 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005211
Glauber Costaee5e8472013-02-22 16:34:50 -08005212 /*
5213 * No kind of locking is needed in here, because ->can_attach() will
5214 * check this value once in the beginning of the process, and then carry
5215 * on with stale data. This means that changes to this value will only
5216 * affect task migrations starting after the change.
5217 */
5218 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005219 return 0;
5220}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005221#else
Tejun Heo182446d2013-08-08 20:11:24 -04005222static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005223 struct cftype *cft, u64 val)
5224{
5225 return -ENOSYS;
5226}
5227#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005228
Ying Han406eb0c2011-05-26 16:25:37 -07005229#ifdef CONFIG_NUMA
Tejun Heo182446d2013-08-08 20:11:24 -04005230static int memcg_numa_stat_show(struct cgroup_subsys_state *css,
5231 struct cftype *cft, struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07005232{
5233 int nid;
5234 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
5235 unsigned long node_nr;
Tejun Heo182446d2013-08-08 20:11:24 -04005236 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Ying Han406eb0c2011-05-26 16:25:37 -07005237
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005238 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005239 seq_printf(m, "total=%lu", total_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005240 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005241 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005242 seq_printf(m, " N%d=%lu", nid, node_nr);
5243 }
5244 seq_putc(m, '\n');
5245
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005246 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005247 seq_printf(m, "file=%lu", file_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005248 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005249 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005250 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005251 seq_printf(m, " N%d=%lu", nid, node_nr);
5252 }
5253 seq_putc(m, '\n');
5254
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005255 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005256 seq_printf(m, "anon=%lu", anon_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005257 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005258 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005259 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005260 seq_printf(m, " N%d=%lu", nid, node_nr);
5261 }
5262 seq_putc(m, '\n');
5263
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005264 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005265 seq_printf(m, "unevictable=%lu", unevictable_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005266 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005267 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005268 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005269 seq_printf(m, " N%d=%lu", nid, node_nr);
5270 }
5271 seq_putc(m, '\n');
5272 return 0;
5273}
5274#endif /* CONFIG_NUMA */
5275
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005276static inline void mem_cgroup_lru_names_not_uptodate(void)
5277{
5278 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5279}
5280
Tejun Heo182446d2013-08-08 20:11:24 -04005281static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005282 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005283{
Tejun Heo182446d2013-08-08 20:11:24 -04005284 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005285 struct mem_cgroup *mi;
5286 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005287
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005288 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005289 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005290 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005291 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5292 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005293 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005294
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005295 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5296 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5297 mem_cgroup_read_events(memcg, i));
5298
5299 for (i = 0; i < NR_LRU_LISTS; i++)
5300 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5301 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5302
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005303 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005304 {
5305 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005306 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005307 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005308 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005309 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5310 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005311 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005312
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005313 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5314 long long val = 0;
5315
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005316 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005317 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005318 for_each_mem_cgroup_tree(mi, memcg)
5319 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5320 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5321 }
5322
5323 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5324 unsigned long long val = 0;
5325
5326 for_each_mem_cgroup_tree(mi, memcg)
5327 val += mem_cgroup_read_events(mi, i);
5328 seq_printf(m, "total_%s %llu\n",
5329 mem_cgroup_events_names[i], val);
5330 }
5331
5332 for (i = 0; i < NR_LRU_LISTS; i++) {
5333 unsigned long long val = 0;
5334
5335 for_each_mem_cgroup_tree(mi, memcg)
5336 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5337 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005338 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005339
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005340#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005341 {
5342 int nid, zid;
5343 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005344 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005345 unsigned long recent_rotated[2] = {0, 0};
5346 unsigned long recent_scanned[2] = {0, 0};
5347
5348 for_each_online_node(nid)
5349 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005350 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005351 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005352
Hugh Dickins89abfab2012-05-29 15:06:53 -07005353 recent_rotated[0] += rstat->recent_rotated[0];
5354 recent_rotated[1] += rstat->recent_rotated[1];
5355 recent_scanned[0] += rstat->recent_scanned[0];
5356 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005357 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005358 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5359 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5360 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5361 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005362 }
5363#endif
5364
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005365 return 0;
5366}
5367
Tejun Heo182446d2013-08-08 20:11:24 -04005368static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5369 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005370{
Tejun Heo182446d2013-08-08 20:11:24 -04005371 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005372
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005373 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005374}
5375
Tejun Heo182446d2013-08-08 20:11:24 -04005376static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5377 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005378{
Tejun Heo182446d2013-08-08 20:11:24 -04005379 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005380 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005381
Tejun Heo63876982013-08-08 20:11:23 -04005382 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005383 return -EINVAL;
5384
Glauber Costa09998212013-02-22 16:34:55 -08005385 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005386
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005387 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005388 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005389 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005390 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005391 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005392
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005393 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005394
Glauber Costa09998212013-02-22 16:34:55 -08005395 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005396
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005397 return 0;
5398}
5399
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005400static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5401{
5402 struct mem_cgroup_threshold_ary *t;
5403 u64 usage;
5404 int i;
5405
5406 rcu_read_lock();
5407 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005408 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005409 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005410 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005411
5412 if (!t)
5413 goto unlock;
5414
5415 usage = mem_cgroup_usage(memcg, swap);
5416
5417 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005418 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005419 * If it's not true, a threshold was crossed after last
5420 * call of __mem_cgroup_threshold().
5421 */
Phil Carmody5407a562010-05-26 14:42:42 -07005422 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005423
5424 /*
5425 * Iterate backward over array of thresholds starting from
5426 * current_threshold and check if a threshold is crossed.
5427 * If none of thresholds below usage is crossed, we read
5428 * only one element of the array here.
5429 */
5430 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5431 eventfd_signal(t->entries[i].eventfd, 1);
5432
5433 /* i = current_threshold + 1 */
5434 i++;
5435
5436 /*
5437 * Iterate forward over array of thresholds starting from
5438 * current_threshold+1 and check if a threshold is crossed.
5439 * If none of thresholds above usage is crossed, we read
5440 * only one element of the array here.
5441 */
5442 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5443 eventfd_signal(t->entries[i].eventfd, 1);
5444
5445 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005446 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005447unlock:
5448 rcu_read_unlock();
5449}
5450
5451static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5452{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005453 while (memcg) {
5454 __mem_cgroup_threshold(memcg, false);
5455 if (do_swap_account)
5456 __mem_cgroup_threshold(memcg, true);
5457
5458 memcg = parent_mem_cgroup(memcg);
5459 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005460}
5461
5462static int compare_thresholds(const void *a, const void *b)
5463{
5464 const struct mem_cgroup_threshold *_a = a;
5465 const struct mem_cgroup_threshold *_b = b;
5466
Greg Thelen2bff24a2013-09-11 14:23:08 -07005467 if (_a->threshold > _b->threshold)
5468 return 1;
5469
5470 if (_a->threshold < _b->threshold)
5471 return -1;
5472
5473 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005474}
5475
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005476static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005477{
5478 struct mem_cgroup_eventfd_list *ev;
5479
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005480 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005481 eventfd_signal(ev->eventfd, 1);
5482 return 0;
5483}
5484
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005485static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005486{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005487 struct mem_cgroup *iter;
5488
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005489 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005490 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005491}
5492
Tejun Heo81eeaf02013-08-08 20:11:26 -04005493static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005494 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005495{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005496 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005497 struct mem_cgroup_thresholds *thresholds;
5498 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005499 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005500 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005501 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005502
5503 ret = res_counter_memparse_write_strategy(args, &threshold);
5504 if (ret)
5505 return ret;
5506
5507 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005508
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005509 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005510 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005511 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005512 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005513 else
5514 BUG();
5515
5516 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5517
5518 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005519 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005520 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5521
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005522 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005523
5524 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005525 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005526 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005527 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005528 ret = -ENOMEM;
5529 goto unlock;
5530 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005531 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005532
5533 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005534 if (thresholds->primary) {
5535 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005536 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005537 }
5538
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005539 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005540 new->entries[size - 1].eventfd = eventfd;
5541 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005542
5543 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005544 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005545 compare_thresholds, NULL);
5546
5547 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005548 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005549 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005550 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005551 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005552 * new->current_threshold will not be used until
5553 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005554 * it here.
5555 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005556 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005557 } else
5558 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005559 }
5560
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005561 /* Free old spare buffer and save old primary buffer as spare */
5562 kfree(thresholds->spare);
5563 thresholds->spare = thresholds->primary;
5564
5565 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005566
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005567 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005568 synchronize_rcu();
5569
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005570unlock:
5571 mutex_unlock(&memcg->thresholds_lock);
5572
5573 return ret;
5574}
5575
Tejun Heo81eeaf02013-08-08 20:11:26 -04005576static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005577 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005578{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005579 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005580 struct mem_cgroup_thresholds *thresholds;
5581 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005582 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005583 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005584 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005585
5586 mutex_lock(&memcg->thresholds_lock);
5587 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005588 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005589 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005590 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005591 else
5592 BUG();
5593
Anton Vorontsov371528c2012-02-24 05:14:46 +04005594 if (!thresholds->primary)
5595 goto unlock;
5596
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005597 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5598
5599 /* Check if a threshold crossed before removing */
5600 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5601
5602 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005603 size = 0;
5604 for (i = 0; i < thresholds->primary->size; i++) {
5605 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005606 size++;
5607 }
5608
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005609 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005610
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005611 /* Set thresholds array to NULL if we don't have thresholds */
5612 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005613 kfree(new);
5614 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005615 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005616 }
5617
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005618 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005619
5620 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005621 new->current_threshold = -1;
5622 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5623 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005624 continue;
5625
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005626 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005627 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005628 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005629 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005630 * until rcu_assign_pointer(), so it's safe to increment
5631 * it here.
5632 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005633 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005634 }
5635 j++;
5636 }
5637
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005638swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005639 /* Swap primary and spare array */
5640 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005641 /* If all events are unregistered, free the spare array */
5642 if (!new) {
5643 kfree(thresholds->spare);
5644 thresholds->spare = NULL;
5645 }
5646
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005647 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005648
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005649 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005650 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005651unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005652 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005653}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005654
Tejun Heo81eeaf02013-08-08 20:11:26 -04005655static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005656 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5657{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005658 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005659 struct mem_cgroup_eventfd_list *event;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005660 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005661
5662 BUG_ON(type != _OOM_TYPE);
5663 event = kmalloc(sizeof(*event), GFP_KERNEL);
5664 if (!event)
5665 return -ENOMEM;
5666
Michal Hocko1af8efe2011-07-26 16:08:24 -07005667 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005668
5669 event->eventfd = eventfd;
5670 list_add(&event->list, &memcg->oom_notify);
5671
5672 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005673 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005674 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005675 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005676
5677 return 0;
5678}
5679
Tejun Heo81eeaf02013-08-08 20:11:26 -04005680static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005681 struct cftype *cft, struct eventfd_ctx *eventfd)
5682{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005683 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005684 struct mem_cgroup_eventfd_list *ev, *tmp;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005685 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005686
5687 BUG_ON(type != _OOM_TYPE);
5688
Michal Hocko1af8efe2011-07-26 16:08:24 -07005689 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005690
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005691 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005692 if (ev->eventfd == eventfd) {
5693 list_del(&ev->list);
5694 kfree(ev);
5695 }
5696 }
5697
Michal Hocko1af8efe2011-07-26 16:08:24 -07005698 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005699}
5700
Tejun Heo182446d2013-08-08 20:11:24 -04005701static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005702 struct cftype *cft, struct cgroup_map_cb *cb)
5703{
Tejun Heo182446d2013-08-08 20:11:24 -04005704 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005705
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005706 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005707
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005708 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005709 cb->fill(cb, "under_oom", 1);
5710 else
5711 cb->fill(cb, "under_oom", 0);
5712 return 0;
5713}
5714
Tejun Heo182446d2013-08-08 20:11:24 -04005715static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005716 struct cftype *cft, u64 val)
5717{
Tejun Heo182446d2013-08-08 20:11:24 -04005718 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005719 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005720
5721 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005722 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005723 return -EINVAL;
5724
Glauber Costa09998212013-02-22 16:34:55 -08005725 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005726 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005727 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005728 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005729 return -EINVAL;
5730 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005731 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005732 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005733 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005734 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005735 return 0;
5736}
5737
Andrew Mortonc255a452012-07-31 16:43:02 -07005738#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005739static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005740{
Glauber Costa55007d82012-12-18 14:22:38 -08005741 int ret;
5742
Glauber Costa2633d7a2012-12-18 14:22:34 -08005743 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005744 ret = memcg_propagate_kmem(memcg);
5745 if (ret)
5746 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005747
Glauber Costa1d62e432012-04-09 19:36:33 -03005748 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005749}
Glauber Costae5671df2011-12-11 21:47:01 +00005750
Li Zefan10d5ebf2013-07-08 16:00:33 -07005751static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005752{
Glauber Costa1d62e432012-04-09 19:36:33 -03005753 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005754}
5755
5756static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5757{
5758 if (!memcg_kmem_is_active(memcg))
5759 return;
5760
5761 /*
5762 * kmem charges can outlive the cgroup. In the case of slab
5763 * pages, for instance, a page contain objects from various
5764 * processes. As we prevent from taking a reference for every
5765 * such allocation we have to be careful when doing uncharge
5766 * (see memcg_uncharge_kmem) and here during offlining.
5767 *
5768 * The idea is that that only the _last_ uncharge which sees
5769 * the dead memcg will drop the last reference. An additional
5770 * reference is taken here before the group is marked dead
5771 * which is then paired with css_put during uncharge resp. here.
5772 *
5773 * Although this might sound strange as this path is called from
5774 * css_offline() when the referencemight have dropped down to 0
5775 * and shouldn't be incremented anymore (css_tryget would fail)
5776 * we do not have other options because of the kmem allocations
5777 * lifetime.
5778 */
5779 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005780
5781 memcg_kmem_mark_dead(memcg);
5782
5783 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5784 return;
5785
Glauber Costa7de37682012-12-18 14:22:07 -08005786 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005787 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005788}
Glauber Costae5671df2011-12-11 21:47:01 +00005789#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005790static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005791{
5792 return 0;
5793}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005794
Li Zefan10d5ebf2013-07-08 16:00:33 -07005795static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5796{
5797}
5798
5799static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005800{
5801}
Glauber Costae5671df2011-12-11 21:47:01 +00005802#endif
5803
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005804static struct cftype mem_cgroup_files[] = {
5805 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005806 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005807 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07005808 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005809 .register_event = mem_cgroup_usage_register_event,
5810 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005811 },
5812 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005813 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005814 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005815 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005816 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005817 },
5818 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005819 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005820 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07005821 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005822 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005823 },
5824 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005825 .name = "soft_limit_in_bytes",
5826 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
5827 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005828 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07005829 },
5830 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005831 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005832 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005833 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005834 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005835 },
Balbir Singh8697d332008-02-07 00:13:59 -08005836 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005837 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005838 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005839 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005840 {
5841 .name = "force_empty",
5842 .trigger = mem_cgroup_force_empty_write,
5843 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005844 {
5845 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07005846 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08005847 .write_u64 = mem_cgroup_hierarchy_write,
5848 .read_u64 = mem_cgroup_hierarchy_read,
5849 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005850 {
5851 .name = "swappiness",
5852 .read_u64 = mem_cgroup_swappiness_read,
5853 .write_u64 = mem_cgroup_swappiness_write,
5854 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005855 {
5856 .name = "move_charge_at_immigrate",
5857 .read_u64 = mem_cgroup_move_charge_read,
5858 .write_u64 = mem_cgroup_move_charge_write,
5859 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005860 {
5861 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005862 .read_map = mem_cgroup_oom_control_read,
5863 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005864 .register_event = mem_cgroup_oom_register_event,
5865 .unregister_event = mem_cgroup_oom_unregister_event,
5866 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
5867 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005868 {
5869 .name = "pressure_level",
5870 .register_event = vmpressure_register_event,
5871 .unregister_event = vmpressure_unregister_event,
5872 },
Ying Han406eb0c2011-05-26 16:25:37 -07005873#ifdef CONFIG_NUMA
5874 {
5875 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005876 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07005877 },
5878#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005879#ifdef CONFIG_MEMCG_KMEM
5880 {
5881 .name = "kmem.limit_in_bytes",
5882 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
5883 .write_string = mem_cgroup_write,
5884 .read = mem_cgroup_read,
5885 },
5886 {
5887 .name = "kmem.usage_in_bytes",
5888 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
5889 .read = mem_cgroup_read,
5890 },
5891 {
5892 .name = "kmem.failcnt",
5893 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
5894 .trigger = mem_cgroup_reset,
5895 .read = mem_cgroup_read,
5896 },
5897 {
5898 .name = "kmem.max_usage_in_bytes",
5899 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
5900 .trigger = mem_cgroup_reset,
5901 .read = mem_cgroup_read,
5902 },
Glauber Costa749c5412012-12-18 14:23:01 -08005903#ifdef CONFIG_SLABINFO
5904 {
5905 .name = "kmem.slabinfo",
5906 .read_seq_string = mem_cgroup_slabinfo_read,
5907 },
5908#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005909#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07005910 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07005911};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005912
Michal Hocko2d110852013-02-22 16:34:43 -08005913#ifdef CONFIG_MEMCG_SWAP
5914static struct cftype memsw_cgroup_files[] = {
5915 {
5916 .name = "memsw.usage_in_bytes",
5917 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5918 .read = mem_cgroup_read,
5919 .register_event = mem_cgroup_usage_register_event,
5920 .unregister_event = mem_cgroup_usage_unregister_event,
5921 },
5922 {
5923 .name = "memsw.max_usage_in_bytes",
5924 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5925 .trigger = mem_cgroup_reset,
5926 .read = mem_cgroup_read,
5927 },
5928 {
5929 .name = "memsw.limit_in_bytes",
5930 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5931 .write_string = mem_cgroup_write,
5932 .read = mem_cgroup_read,
5933 },
5934 {
5935 .name = "memsw.failcnt",
5936 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5937 .trigger = mem_cgroup_reset,
5938 .read = mem_cgroup_read,
5939 },
5940 { }, /* terminate */
5941};
5942#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005943static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005944{
5945 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005946 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005947 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005948 /*
5949 * This routine is called against possible nodes.
5950 * But it's BUG to call kmalloc() against offline node.
5951 *
5952 * TODO: this routine can waste much memory for nodes which will
5953 * never be onlined. It's better to use memory hotplug callback
5954 * function.
5955 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005956 if (!node_state(node, N_NORMAL_MEMORY))
5957 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08005958 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005959 if (!pn)
5960 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005961
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005962 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5963 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08005964 lruvec_init(&mz->lruvec);
Balbir Singhf64c3f52009-09-23 15:56:37 -07005965 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07005966 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005967 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005968 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005969 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005970 return 0;
5971}
5972
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005973static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005974{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005975 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005976}
5977
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005978static struct mem_cgroup *mem_cgroup_alloc(void)
5979{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005980 struct mem_cgroup *memcg;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08005981 size_t size = memcg_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005982
Glauber Costa45cf7eb2013-02-22 16:34:49 -08005983 /* Can be very big if nr_node_ids is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08005984 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005985 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005986 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005987 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005988
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005989 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005990 return NULL;
5991
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005992 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
5993 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08005994 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005995 spin_lock_init(&memcg->pcp_counter_lock);
5996 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08005997
5998out_free:
5999 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006000 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006001 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006002 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006003 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006004}
6005
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006006/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006007 * At destroying mem_cgroup, references from swap_cgroup can remain.
6008 * (scanning all at force_empty is too costly...)
6009 *
6010 * Instead of clearing all references at force_empty, we remember
6011 * the number of reference from swap_cgroup and free mem_cgroup when
6012 * it goes down to 0.
6013 *
6014 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006015 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006016
6017static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006018{
Glauber Costac8b2a362012-12-18 14:22:13 -08006019 int node;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006020 size_t size = memcg_size();
Hugh Dickins59927fb2012-03-15 15:17:07 -07006021
Glauber Costac8b2a362012-12-18 14:22:13 -08006022 mem_cgroup_remove_from_trees(memcg);
6023 free_css_id(&mem_cgroup_subsys, &memcg->css);
6024
6025 for_each_node(node)
6026 free_mem_cgroup_per_zone_info(memcg, node);
6027
6028 free_percpu(memcg->stat);
6029
Glauber Costa3f134612012-05-29 15:07:11 -07006030 /*
6031 * We need to make sure that (at least for now), the jump label
6032 * destruction code runs outside of the cgroup lock. This is because
6033 * get_online_cpus(), which is called from the static_branch update,
6034 * can't be called inside the cgroup_lock. cpusets are the ones
6035 * enforcing this dependency, so if they ever change, we might as well.
6036 *
6037 * schedule_work() will guarantee this happens. Be careful if you need
6038 * to move this code around, and make sure it is outside
6039 * the cgroup_lock.
6040 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006041 disarm_static_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006042 if (size < PAGE_SIZE)
6043 kfree(memcg);
6044 else
6045 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006046}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006047
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006048/*
6049 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6050 */
Glauber Costae1aab162011-12-11 21:47:03 +00006051struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006052{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006053 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006054 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006055 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006056}
Glauber Costae1aab162011-12-11 21:47:03 +00006057EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006058
Michal Hocko8787a1d2013-02-22 16:35:39 -08006059static void __init mem_cgroup_soft_limit_tree_init(void)
Balbir Singhf64c3f52009-09-23 15:56:37 -07006060{
6061 struct mem_cgroup_tree_per_node *rtpn;
6062 struct mem_cgroup_tree_per_zone *rtpz;
6063 int tmp, node, zone;
6064
Bob Liu3ed28fa2012-01-12 17:19:04 -08006065 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07006066 tmp = node;
6067 if (!node_state(node, N_NORMAL_MEMORY))
6068 tmp = -1;
6069 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
Michal Hocko8787a1d2013-02-22 16:35:39 -08006070 BUG_ON(!rtpn);
Balbir Singhf64c3f52009-09-23 15:56:37 -07006071
6072 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6073
6074 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6075 rtpz = &rtpn->rb_tree_per_zone[zone];
6076 rtpz->rb_root = RB_ROOT;
6077 spin_lock_init(&rtpz->lock);
6078 }
6079 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07006080}
6081
Li Zefan0eb253e2009-01-15 13:51:25 -08006082static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006083mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006084{
Glauber Costad142e3e2013-02-22 16:34:52 -08006085 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006086 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006087 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006088
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006089 memcg = mem_cgroup_alloc();
6090 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006091 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006092
Bob Liu3ed28fa2012-01-12 17:19:04 -08006093 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006094 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006095 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006096
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006097 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006098 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006099 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006100 res_counter_init(&memcg->res, NULL);
6101 res_counter_init(&memcg->memsw, NULL);
6102 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006103 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006104
Glauber Costad142e3e2013-02-22 16:34:52 -08006105 memcg->last_scanned_node = MAX_NUMNODES;
6106 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006107 memcg->move_charge_at_immigrate = 0;
6108 mutex_init(&memcg->thresholds_lock);
6109 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006110 vmpressure_init(&memcg->vmpressure);
Glauber Costad142e3e2013-02-22 16:34:52 -08006111
6112 return &memcg->css;
6113
6114free_out:
6115 __mem_cgroup_free(memcg);
6116 return ERR_PTR(error);
6117}
6118
6119static int
Tejun Heoeb954192013-08-08 20:11:23 -04006120mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006121{
Tejun Heoeb954192013-08-08 20:11:23 -04006122 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6123 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006124 int error = 0;
6125
Tejun Heo63876982013-08-08 20:11:23 -04006126 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006127 return 0;
6128
Glauber Costa09998212013-02-22 16:34:55 -08006129 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006130
6131 memcg->use_hierarchy = parent->use_hierarchy;
6132 memcg->oom_kill_disable = parent->oom_kill_disable;
6133 memcg->swappiness = mem_cgroup_swappiness(parent);
6134
6135 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006136 res_counter_init(&memcg->res, &parent->res);
6137 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006138 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006139
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006140 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006141 * No need to take a reference to the parent because cgroup
6142 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006143 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006144 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006145 res_counter_init(&memcg->res, NULL);
6146 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006147 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006148 /*
6149 * Deeper hierachy with use_hierarchy == false doesn't make
6150 * much sense so let cgroup subsystem know about this
6151 * unfortunate state in our controller.
6152 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006153 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006154 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006155 }
Glauber Costacbe128e32012-04-09 19:36:34 -03006156
6157 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
Glauber Costa09998212013-02-22 16:34:55 -08006158 mutex_unlock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006159 return error;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006160}
6161
Michal Hocko5f578162013-04-29 15:07:17 -07006162/*
6163 * Announce all parents that a group from their hierarchy is gone.
6164 */
6165static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6166{
6167 struct mem_cgroup *parent = memcg;
6168
6169 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006170 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006171
6172 /*
6173 * if the root memcg is not hierarchical we have to check it
6174 * explicitely.
6175 */
6176 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006177 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006178}
6179
Tejun Heoeb954192013-08-08 20:11:23 -04006180static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006181{
Tejun Heoeb954192013-08-08 20:11:23 -04006182 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006183
Li Zefan10d5ebf2013-07-08 16:00:33 -07006184 kmem_cgroup_css_offline(memcg);
6185
Michal Hocko5f578162013-04-29 15:07:17 -07006186 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006187 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006188 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006189 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006190}
6191
Tejun Heoeb954192013-08-08 20:11:23 -04006192static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006193{
Tejun Heoeb954192013-08-08 20:11:23 -04006194 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006195
Li Zefan10d5ebf2013-07-08 16:00:33 -07006196 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006197 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006198}
6199
Daisuke Nishimura02491442010-03-10 15:22:17 -08006200#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006201/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006202#define PRECHARGE_COUNT_AT_ONCE 256
6203static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006204{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006205 int ret = 0;
6206 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006207 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006208
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006209 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006210 mc.precharge += count;
6211 /* we don't need css_get for root */
6212 return ret;
6213 }
6214 /* try to charge at once */
6215 if (count > 1) {
6216 struct res_counter *dummy;
6217 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006218 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006219 * by cgroup_lock_live_cgroup() that it is not removed and we
6220 * are still under the same cgroup_mutex. So we can postpone
6221 * css_get().
6222 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006223 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006224 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006225 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006226 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006227 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006228 goto one_by_one;
6229 }
6230 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006231 return ret;
6232 }
6233one_by_one:
6234 /* fall back to one by one charge */
6235 while (count--) {
6236 if (signal_pending(current)) {
6237 ret = -EINTR;
6238 break;
6239 }
6240 if (!batch_count--) {
6241 batch_count = PRECHARGE_COUNT_AT_ONCE;
6242 cond_resched();
6243 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006244 ret = __mem_cgroup_try_charge(NULL,
6245 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006246 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006247 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006248 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006249 mc.precharge++;
6250 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006251 return ret;
6252}
6253
6254/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006255 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006256 * @vma: the vma the pte to be checked belongs
6257 * @addr: the address corresponding to the pte to be checked
6258 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006259 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006260 *
6261 * Returns
6262 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6263 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6264 * move charge. if @target is not NULL, the page is stored in target->page
6265 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006266 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6267 * target for charge migration. if @target is not NULL, the entry is stored
6268 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006269 *
6270 * Called with pte lock held.
6271 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006272union mc_target {
6273 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006274 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006275};
6276
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006277enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006278 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006279 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006280 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006281};
6282
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006283static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6284 unsigned long addr, pte_t ptent)
6285{
6286 struct page *page = vm_normal_page(vma, addr, ptent);
6287
6288 if (!page || !page_mapped(page))
6289 return NULL;
6290 if (PageAnon(page)) {
6291 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006292 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006293 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006294 } else if (!move_file())
6295 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006296 return NULL;
6297 if (!get_page_unless_zero(page))
6298 return NULL;
6299
6300 return page;
6301}
6302
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006303#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006304static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6305 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6306{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006307 struct page *page = NULL;
6308 swp_entry_t ent = pte_to_swp_entry(ptent);
6309
6310 if (!move_anon() || non_swap_entry(ent))
6311 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006312 /*
6313 * Because lookup_swap_cache() updates some statistics counter,
6314 * we call find_get_page() with swapper_space directly.
6315 */
Shaohua Li33806f02013-02-22 16:34:37 -08006316 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006317 if (do_swap_account)
6318 entry->val = ent.val;
6319
6320 return page;
6321}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006322#else
6323static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6324 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6325{
6326 return NULL;
6327}
6328#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006329
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006330static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6331 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6332{
6333 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006334 struct address_space *mapping;
6335 pgoff_t pgoff;
6336
6337 if (!vma->vm_file) /* anonymous vma */
6338 return NULL;
6339 if (!move_file())
6340 return NULL;
6341
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006342 mapping = vma->vm_file->f_mapping;
6343 if (pte_none(ptent))
6344 pgoff = linear_page_index(vma, addr);
6345 else /* pte_file(ptent) is true */
6346 pgoff = pte_to_pgoff(ptent);
6347
6348 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006349 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006350
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006351#ifdef CONFIG_SWAP
6352 /* shmem/tmpfs may report page out on swap: account for that too. */
6353 if (radix_tree_exceptional_entry(page)) {
6354 swp_entry_t swap = radix_to_swp_entry(page);
6355 if (do_swap_account)
6356 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006357 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006358 }
6359#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006360 return page;
6361}
6362
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006363static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006364 unsigned long addr, pte_t ptent, union mc_target *target)
6365{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006366 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006367 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006368 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006369 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006370
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006371 if (pte_present(ptent))
6372 page = mc_handle_present_pte(vma, addr, ptent);
6373 else if (is_swap_pte(ptent))
6374 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006375 else if (pte_none(ptent) || pte_file(ptent))
6376 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006377
6378 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006379 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006380 if (page) {
6381 pc = lookup_page_cgroup(page);
6382 /*
6383 * Do only loose check w/o page_cgroup lock.
6384 * mem_cgroup_move_account() checks the pc is valid or not under
6385 * the lock.
6386 */
6387 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6388 ret = MC_TARGET_PAGE;
6389 if (target)
6390 target->page = page;
6391 }
6392 if (!ret || !target)
6393 put_page(page);
6394 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006395 /* There is a swap entry and a page doesn't exist or isn't charged */
6396 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08006397 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006398 ret = MC_TARGET_SWAP;
6399 if (target)
6400 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006401 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006402 return ret;
6403}
6404
Naoya Horiguchi12724852012-03-21 16:34:28 -07006405#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6406/*
6407 * We don't consider swapping or file mapped pages because THP does not
6408 * support them for now.
6409 * Caller should make sure that pmd_trans_huge(pmd) is true.
6410 */
6411static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6412 unsigned long addr, pmd_t pmd, union mc_target *target)
6413{
6414 struct page *page = NULL;
6415 struct page_cgroup *pc;
6416 enum mc_target_type ret = MC_TARGET_NONE;
6417
6418 page = pmd_page(pmd);
6419 VM_BUG_ON(!page || !PageHead(page));
6420 if (!move_anon())
6421 return ret;
6422 pc = lookup_page_cgroup(page);
6423 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6424 ret = MC_TARGET_PAGE;
6425 if (target) {
6426 get_page(page);
6427 target->page = page;
6428 }
6429 }
6430 return ret;
6431}
6432#else
6433static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6434 unsigned long addr, pmd_t pmd, union mc_target *target)
6435{
6436 return MC_TARGET_NONE;
6437}
6438#endif
6439
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006440static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6441 unsigned long addr, unsigned long end,
6442 struct mm_walk *walk)
6443{
6444 struct vm_area_struct *vma = walk->private;
6445 pte_t *pte;
6446 spinlock_t *ptl;
6447
Naoya Horiguchi12724852012-03-21 16:34:28 -07006448 if (pmd_trans_huge_lock(pmd, vma) == 1) {
6449 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6450 mc.precharge += HPAGE_PMD_NR;
6451 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006452 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006453 }
Dave Hansen03319322011-03-22 16:32:56 -07006454
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006455 if (pmd_trans_unstable(pmd))
6456 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006457 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6458 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006459 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006460 mc.precharge++; /* increment precharge temporarily */
6461 pte_unmap_unlock(pte - 1, ptl);
6462 cond_resched();
6463
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006464 return 0;
6465}
6466
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006467static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6468{
6469 unsigned long precharge;
6470 struct vm_area_struct *vma;
6471
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006472 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006473 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6474 struct mm_walk mem_cgroup_count_precharge_walk = {
6475 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6476 .mm = mm,
6477 .private = vma,
6478 };
6479 if (is_vm_hugetlb_page(vma))
6480 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006481 walk_page_range(vma->vm_start, vma->vm_end,
6482 &mem_cgroup_count_precharge_walk);
6483 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006484 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006485
6486 precharge = mc.precharge;
6487 mc.precharge = 0;
6488
6489 return precharge;
6490}
6491
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006492static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6493{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006494 unsigned long precharge = mem_cgroup_count_precharge(mm);
6495
6496 VM_BUG_ON(mc.moving_task);
6497 mc.moving_task = current;
6498 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006499}
6500
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006501/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6502static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006503{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006504 struct mem_cgroup *from = mc.from;
6505 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006506 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006507
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006508 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006509 if (mc.precharge) {
6510 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6511 mc.precharge = 0;
6512 }
6513 /*
6514 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6515 * we must uncharge here.
6516 */
6517 if (mc.moved_charge) {
6518 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6519 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006520 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006521 /* we must fixup refcnts and charges */
6522 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006523 /* uncharge swap account from the old cgroup */
6524 if (!mem_cgroup_is_root(mc.from))
6525 res_counter_uncharge(&mc.from->memsw,
6526 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006527
6528 for (i = 0; i < mc.moved_swap; i++)
6529 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006530
6531 if (!mem_cgroup_is_root(mc.to)) {
6532 /*
6533 * we charged both to->res and to->memsw, so we should
6534 * uncharge to->res.
6535 */
6536 res_counter_uncharge(&mc.to->res,
6537 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006538 }
Li Zefan40503772013-07-08 16:00:34 -07006539 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006540 mc.moved_swap = 0;
6541 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006542 memcg_oom_recover(from);
6543 memcg_oom_recover(to);
6544 wake_up_all(&mc.waitq);
6545}
6546
6547static void mem_cgroup_clear_mc(void)
6548{
6549 struct mem_cgroup *from = mc.from;
6550
6551 /*
6552 * we must clear moving_task before waking up waiters at the end of
6553 * task migration.
6554 */
6555 mc.moving_task = NULL;
6556 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006557 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006558 mc.from = NULL;
6559 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006560 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006561 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006562}
6563
Tejun Heoeb954192013-08-08 20:11:23 -04006564static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006565 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006566{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006567 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006568 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04006569 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08006570 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006571
Glauber Costaee5e8472013-02-22 16:34:50 -08006572 /*
6573 * We are now commited to this value whatever it is. Changes in this
6574 * tunable will only affect upcoming migrations, not the current one.
6575 * So we need to save it, and keep it going.
6576 */
6577 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6578 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006579 struct mm_struct *mm;
6580 struct mem_cgroup *from = mem_cgroup_from_task(p);
6581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006582 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006583
6584 mm = get_task_mm(p);
6585 if (!mm)
6586 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006587 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006588 if (mm->owner == p) {
6589 VM_BUG_ON(mc.from);
6590 VM_BUG_ON(mc.to);
6591 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006592 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006593 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006594 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006595 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006596 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006597 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006598 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006599 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006600 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006601
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006602 ret = mem_cgroup_precharge_mc(mm);
6603 if (ret)
6604 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006605 }
6606 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006607 }
6608 return ret;
6609}
6610
Tejun Heoeb954192013-08-08 20:11:23 -04006611static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006612 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006613{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006614 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006615}
6616
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006617static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6618 unsigned long addr, unsigned long end,
6619 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006620{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006621 int ret = 0;
6622 struct vm_area_struct *vma = walk->private;
6623 pte_t *pte;
6624 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006625 enum mc_target_type target_type;
6626 union mc_target target;
6627 struct page *page;
6628 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006629
Naoya Horiguchi12724852012-03-21 16:34:28 -07006630 /*
6631 * We don't take compound_lock() here but no race with splitting thp
6632 * happens because:
6633 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6634 * under splitting, which means there's no concurrent thp split,
6635 * - if another thread runs into split_huge_page() just after we
6636 * entered this if-block, the thread must wait for page table lock
6637 * to be unlocked in __split_huge_page_splitting(), where the main
6638 * part of thp split is not executed yet.
6639 */
6640 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006641 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006642 spin_unlock(&vma->vm_mm->page_table_lock);
6643 return 0;
6644 }
6645 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6646 if (target_type == MC_TARGET_PAGE) {
6647 page = target.page;
6648 if (!isolate_lru_page(page)) {
6649 pc = lookup_page_cgroup(page);
6650 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006651 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006652 mc.precharge -= HPAGE_PMD_NR;
6653 mc.moved_charge += HPAGE_PMD_NR;
6654 }
6655 putback_lru_page(page);
6656 }
6657 put_page(page);
6658 }
6659 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006660 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006661 }
6662
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006663 if (pmd_trans_unstable(pmd))
6664 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006665retry:
6666 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6667 for (; addr != end; addr += PAGE_SIZE) {
6668 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006669 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006670
6671 if (!mc.precharge)
6672 break;
6673
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006674 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006675 case MC_TARGET_PAGE:
6676 page = target.page;
6677 if (isolate_lru_page(page))
6678 goto put;
6679 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006680 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006681 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006682 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006683 /* we uncharge from mc.from later. */
6684 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006685 }
6686 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006687put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006688 put_page(page);
6689 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006690 case MC_TARGET_SWAP:
6691 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006692 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006693 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006694 /* we fixup refcnts and charges later. */
6695 mc.moved_swap++;
6696 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006697 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006698 default:
6699 break;
6700 }
6701 }
6702 pte_unmap_unlock(pte - 1, ptl);
6703 cond_resched();
6704
6705 if (addr != end) {
6706 /*
6707 * We have consumed all precharges we got in can_attach().
6708 * We try charge one by one, but don't do any additional
6709 * charges to mc.to if we have failed in charge once in attach()
6710 * phase.
6711 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006712 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006713 if (!ret)
6714 goto retry;
6715 }
6716
6717 return ret;
6718}
6719
6720static void mem_cgroup_move_charge(struct mm_struct *mm)
6721{
6722 struct vm_area_struct *vma;
6723
6724 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006725retry:
6726 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6727 /*
6728 * Someone who are holding the mmap_sem might be waiting in
6729 * waitq. So we cancel all extra charges, wake up all waiters,
6730 * and retry. Because we cancel precharges, we might not be able
6731 * to move enough charges, but moving charge is a best-effort
6732 * feature anyway, so it wouldn't be a big problem.
6733 */
6734 __mem_cgroup_clear_mc();
6735 cond_resched();
6736 goto retry;
6737 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006738 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6739 int ret;
6740 struct mm_walk mem_cgroup_move_charge_walk = {
6741 .pmd_entry = mem_cgroup_move_charge_pte_range,
6742 .mm = mm,
6743 .private = vma,
6744 };
6745 if (is_vm_hugetlb_page(vma))
6746 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006747 ret = walk_page_range(vma->vm_start, vma->vm_end,
6748 &mem_cgroup_move_charge_walk);
6749 if (ret)
6750 /*
6751 * means we have consumed all precharges and failed in
6752 * doing additional charge. Just abandon here.
6753 */
6754 break;
6755 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006756 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006757}
6758
Tejun Heoeb954192013-08-08 20:11:23 -04006759static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006760 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006761{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006762 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006763 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006764
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006765 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006766 if (mc.to)
6767 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006768 mmput(mm);
6769 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006770 if (mc.to)
6771 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006772}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006773#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04006774static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006775 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006776{
6777 return 0;
6778}
Tejun Heoeb954192013-08-08 20:11:23 -04006779static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006780 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006781{
6782}
Tejun Heoeb954192013-08-08 20:11:23 -04006783static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef52012-01-31 13:47:36 +08006784 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006785{
6786}
6787#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006788
Tejun Heof00baae2013-04-15 13:41:15 -07006789/*
6790 * Cgroup retains root cgroups across [un]mount cycles making it necessary
6791 * to verify sane_behavior flag on each mount attempt.
6792 */
Tejun Heoeb954192013-08-08 20:11:23 -04006793static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006794{
6795 /*
6796 * use_hierarchy is forced with sane_behavior. cgroup core
6797 * guarantees that @root doesn't have any children, so turning it
6798 * on for the root memcg is enough.
6799 */
Tejun Heoeb954192013-08-08 20:11:23 -04006800 if (cgroup_sane_behavior(root_css->cgroup))
6801 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07006802}
6803
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006804struct cgroup_subsys mem_cgroup_subsys = {
6805 .name = "memory",
6806 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08006807 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006808 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006809 .css_offline = mem_cgroup_css_offline,
6810 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006811 .can_attach = mem_cgroup_can_attach,
6812 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08006813 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006814 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07006815 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006816 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006817 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006818};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006819
Andrew Mortonc255a452012-07-31 16:43:02 -07006820#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08006821static int __init enable_swap_account(char *s)
6822{
Michal Hockoa2c89902011-05-24 17:12:50 -07006823 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006824 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07006825 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006826 really_do_swap_account = 0;
6827 return 1;
6828}
Michal Hockoa2c89902011-05-24 17:12:50 -07006829__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006830
Michal Hocko2d110852013-02-22 16:34:43 -08006831static void __init memsw_file_init(void)
6832{
Michal Hocko6acc8b02013-02-22 16:34:45 -08006833 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08006834}
Michal Hocko6acc8b02013-02-22 16:34:45 -08006835
6836static void __init enable_swap_cgroup(void)
6837{
6838 if (!mem_cgroup_disabled() && really_do_swap_account) {
6839 do_swap_account = 1;
6840 memsw_file_init();
6841 }
6842}
6843
Michal Hocko2d110852013-02-22 16:34:43 -08006844#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08006845static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08006846{
6847}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006848#endif
Michal Hocko2d110852013-02-22 16:34:43 -08006849
6850/*
Michal Hocko10813122013-02-22 16:35:41 -08006851 * subsys_initcall() for memory controller.
6852 *
6853 * Some parts like hotcpu_notifier() have to be initialized from this context
6854 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
6855 * everything that doesn't depend on a specific mem_cgroup structure should
6856 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08006857 */
6858static int __init mem_cgroup_init(void)
6859{
6860 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08006861 enable_swap_cgroup();
Michal Hocko8787a1d2013-02-22 16:35:39 -08006862 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08006863 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08006864 return 0;
6865}
6866subsys_initcall(mem_cgroup_init);