blob: d12ca6f3c2934bc83bd83c467ed672c7f352d172 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080017 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
21 *
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
26 */
27
28#include <linux/res_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080031#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080032#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080033#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
48#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080049#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080050#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070051#include <linux/vmalloc.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070054#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080057#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000058#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070059#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000060#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080061
Balbir Singh8697d332008-02-07 00:13:59 -080062#include <asm/uaccess.h>
63
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070064#include <trace/events/vmscan.h>
65
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070066struct cgroup_subsys mem_cgroup_subsys __read_mostly;
David Rientjes68ae5642012-12-12 13:51:57 -080067EXPORT_SYMBOL(mem_cgroup_subsys);
68
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070069#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070070static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080071
Andrew Mortonc255a452012-07-31 16:43:02 -070072#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070073/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080074int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080075
76/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070077#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080078static int really_do_swap_account __initdata = 1;
79#else
80static int really_do_swap_account __initdata = 0;
81#endif
82
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080083#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070084#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#endif
86
87
Balbir Singh8cdea7c2008-02-07 00:13:50 -080088/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080089 * Statistics for memory cgroup.
90 */
91enum mem_cgroup_stat_index {
92 /*
93 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
94 */
David Rientjesb070e652013-05-07 16:18:09 -070095 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
96 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
97 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
98 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
99 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800100 MEM_CGROUP_STAT_NSTATS,
101};
102
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700103static const char * const mem_cgroup_stat_names[] = {
104 "cache",
105 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700106 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107 "mapped_file",
108 "swap",
109};
110
Johannes Weinere9f89742011-03-23 16:42:37 -0700111enum mem_cgroup_events_index {
112 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
113 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700114 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
115 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700116 MEM_CGROUP_EVENTS_NSTATS,
117};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700118
119static const char * const mem_cgroup_events_names[] = {
120 "pgpgin",
121 "pgpgout",
122 "pgfault",
123 "pgmajfault",
124};
125
Sha Zhengju58cf1882013-02-22 16:32:05 -0800126static const char * const mem_cgroup_lru_names[] = {
127 "inactive_anon",
128 "active_anon",
129 "inactive_file",
130 "active_file",
131 "unevictable",
132};
133
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700134/*
135 * Per memcg event counter is incremented at every pagein/pageout. With THP,
136 * it will be incremated by the number of pages. This counter is used for
137 * for trigger some periodic events. This is straightforward and better
138 * than using jiffies etc. to handle periodic memcg event.
139 */
140enum mem_cgroup_events_target {
141 MEM_CGROUP_TARGET_THRESH,
142 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700143 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700144 MEM_CGROUP_NTARGETS,
145};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700146#define THRESHOLDS_EVENTS_TARGET 128
147#define SOFTLIMIT_EVENTS_TARGET 1024
148#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700149
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800150struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700151 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700152 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700153 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700154 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800155};
156
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800157struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700158 /*
159 * last scanned hierarchy member. Valid only if last_dead_count
160 * matches memcg->dead_count of the hierarchy root group.
161 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700162 struct mem_cgroup *last_visited;
Michal Hocko5f578162013-04-29 15:07:17 -0700163 unsigned long last_dead_count;
164
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800165 /* scan generation, increased every round-trip */
166 unsigned int generation;
167};
168
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800169/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800170 * per-zone information in memory controller.
171 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800172struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800173 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700174 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800175
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800176 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
177
Balbir Singhf64c3f52009-09-23 15:56:37 -0700178 struct rb_node tree_node; /* RB tree node */
179 unsigned long long usage_in_excess;/* Set to the value by which */
180 /* the soft limit is exceeded*/
181 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700182 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700183 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800184};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800185
186struct mem_cgroup_per_node {
187 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
188};
189
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 -0700486static inline
487struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
488{
489 return container_of(s, struct mem_cgroup, css);
490}
491
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700492/* Some nice accessors for the vmpressure. */
493struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
494{
495 if (!memcg)
496 memcg = root_mem_cgroup;
497 return &memcg->vmpressure;
498}
499
500struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
501{
502 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
503}
504
505struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
506{
507 return &mem_cgroup_from_css(css)->vmpressure;
508}
509
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700510static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
511{
512 return (memcg == root_mem_cgroup);
513}
514
Glauber Costae1aab162011-12-11 21:47:03 +0000515/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700516#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000517
Glauber Costae1aab162011-12-11 21:47:03 +0000518void sock_update_memcg(struct sock *sk)
519{
Glauber Costa376be5f2012-01-20 04:57:14 +0000520 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000521 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700522 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000523
524 BUG_ON(!sk->sk_prot->proto_cgroup);
525
Glauber Costaf3f511e2012-01-05 20:16:39 +0000526 /* Socket cloning can throw us here with sk_cgrp already
527 * filled. It won't however, necessarily happen from
528 * process context. So the test for root memcg given
529 * the current task's memcg won't help us in this case.
530 *
531 * Respecting the original socket's memcg is a better
532 * decision in this case.
533 */
534 if (sk->sk_cgrp) {
535 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700536 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000537 return;
538 }
539
Glauber Costae1aab162011-12-11 21:47:03 +0000540 rcu_read_lock();
541 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700542 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700543 if (!mem_cgroup_is_root(memcg) &&
544 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700545 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000546 }
547 rcu_read_unlock();
548 }
549}
550EXPORT_SYMBOL(sock_update_memcg);
551
552void sock_release_memcg(struct sock *sk)
553{
Glauber Costa376be5f2012-01-20 04:57:14 +0000554 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000555 struct mem_cgroup *memcg;
556 WARN_ON(!sk->sk_cgrp->memcg);
557 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700558 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000559 }
560}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000561
562struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
563{
564 if (!memcg || mem_cgroup_is_root(memcg))
565 return NULL;
566
567 return &memcg->tcp_mem.cg_proto;
568}
569EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000570
Glauber Costa3f134612012-05-29 15:07:11 -0700571static void disarm_sock_keys(struct mem_cgroup *memcg)
572{
573 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
574 return;
575 static_key_slow_dec(&memcg_socket_limit_enabled);
576}
577#else
578static void disarm_sock_keys(struct mem_cgroup *memcg)
579{
580}
581#endif
582
Glauber Costaa8964b92012-12-18 14:22:09 -0800583#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800584/*
585 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
586 * There are two main reasons for not using the css_id for this:
587 * 1) this works better in sparse environments, where we have a lot of memcgs,
588 * but only a few kmem-limited. Or also, if we have, for instance, 200
589 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
590 * 200 entry array for that.
591 *
592 * 2) In order not to violate the cgroup API, we would like to do all memory
593 * allocation in ->create(). At that point, we haven't yet allocated the
594 * css_id. Having a separate index prevents us from messing with the cgroup
595 * core for this
596 *
597 * The current size of the caches array is stored in
598 * memcg_limited_groups_array_size. It will double each time we have to
599 * increase it.
600 */
601static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800602int memcg_limited_groups_array_size;
603
Glauber Costa55007d82012-12-18 14:22:38 -0800604/*
605 * MIN_SIZE is different than 1, because we would like to avoid going through
606 * the alloc/free process all the time. In a small machine, 4 kmem-limited
607 * cgroups is a reasonable guess. In the future, it could be a parameter or
608 * tunable, but that is strictly not necessary.
609 *
610 * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get
611 * this constant directly from cgroup, but it is understandable that this is
612 * better kept as an internal representation in cgroup.c. In any case, the
613 * css_id space is not getting any smaller, and we don't have to necessarily
614 * increase ours as well if it increases.
615 */
616#define MEMCG_CACHES_MIN_SIZE 4
617#define MEMCG_CACHES_MAX_SIZE 65535
618
Glauber Costad7f25f82012-12-18 14:22:40 -0800619/*
620 * A lot of the calls to the cache allocation functions are expected to be
621 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
622 * conditional to this static branch, we'll have to allow modules that does
623 * kmem_cache_alloc and the such to see this symbol as well
624 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800625struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800626EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800627
628static void disarm_kmem_keys(struct mem_cgroup *memcg)
629{
Glauber Costa55007d82012-12-18 14:22:38 -0800630 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800631 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800632 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
633 }
Glauber Costabea207c2012-12-18 14:22:11 -0800634 /*
635 * This check can't live in kmem destruction function,
636 * since the charges will outlive the cgroup
637 */
638 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800639}
640#else
641static void disarm_kmem_keys(struct mem_cgroup *memcg)
642{
643}
644#endif /* CONFIG_MEMCG_KMEM */
645
646static void disarm_static_keys(struct mem_cgroup *memcg)
647{
648 disarm_sock_keys(memcg);
649 disarm_kmem_keys(memcg);
650}
651
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700652static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800653
Balbir Singhf64c3f52009-09-23 15:56:37 -0700654static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700655mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700656{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800657 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700658 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700659}
660
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700661struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100662{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700663 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100664}
665
Balbir Singhf64c3f52009-09-23 15:56:37 -0700666static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700667page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700668{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700669 int nid = page_to_nid(page);
670 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700671
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700672 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700673}
674
675static struct mem_cgroup_tree_per_zone *
676soft_limit_tree_node_zone(int nid, int zid)
677{
678 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
679}
680
681static struct mem_cgroup_tree_per_zone *
682soft_limit_tree_from_page(struct page *page)
683{
684 int nid = page_to_nid(page);
685 int zid = page_zonenum(page);
686
687 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
688}
689
690static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700691__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700692 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700693 struct mem_cgroup_tree_per_zone *mctz,
694 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700695{
696 struct rb_node **p = &mctz->rb_root.rb_node;
697 struct rb_node *parent = NULL;
698 struct mem_cgroup_per_zone *mz_node;
699
700 if (mz->on_tree)
701 return;
702
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700703 mz->usage_in_excess = new_usage_in_excess;
704 if (!mz->usage_in_excess)
705 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700706 while (*p) {
707 parent = *p;
708 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
709 tree_node);
710 if (mz->usage_in_excess < mz_node->usage_in_excess)
711 p = &(*p)->rb_left;
712 /*
713 * We can't avoid mem cgroups that are over their soft
714 * limit by the same amount
715 */
716 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
717 p = &(*p)->rb_right;
718 }
719 rb_link_node(&mz->tree_node, parent, p);
720 rb_insert_color(&mz->tree_node, &mctz->rb_root);
721 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700722}
723
724static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700725__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700726 struct mem_cgroup_per_zone *mz,
727 struct mem_cgroup_tree_per_zone *mctz)
728{
729 if (!mz->on_tree)
730 return;
731 rb_erase(&mz->tree_node, &mctz->rb_root);
732 mz->on_tree = false;
733}
734
735static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700736mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700737 struct mem_cgroup_per_zone *mz,
738 struct mem_cgroup_tree_per_zone *mctz)
739{
740 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700741 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700742 spin_unlock(&mctz->lock);
743}
744
Balbir Singhf64c3f52009-09-23 15:56:37 -0700745
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700746static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700747{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700748 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700749 struct mem_cgroup_per_zone *mz;
750 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700751 int nid = page_to_nid(page);
752 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700753 mctz = soft_limit_tree_from_page(page);
754
755 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700756 * Necessary to update all ancestors when hierarchy is used.
757 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700758 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700759 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
760 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
761 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700762 /*
763 * We have to update the tree if mz is on RB-tree or
764 * mem is over its softlimit.
765 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700766 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700767 spin_lock(&mctz->lock);
768 /* if on-tree, remove it */
769 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700770 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700771 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700772 * Insert again. mz->usage_in_excess will be updated.
773 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700774 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700775 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700776 spin_unlock(&mctz->lock);
777 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700778 }
779}
780
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700781static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700782{
783 int node, zone;
784 struct mem_cgroup_per_zone *mz;
785 struct mem_cgroup_tree_per_zone *mctz;
786
Bob Liu3ed28fa2012-01-12 17:19:04 -0800787 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700788 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700789 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700790 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700791 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700792 }
793 }
794}
795
Balbir Singh4e416952009-09-23 15:56:39 -0700796static struct mem_cgroup_per_zone *
797__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
798{
799 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700800 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700801
802retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700803 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700804 rightmost = rb_last(&mctz->rb_root);
805 if (!rightmost)
806 goto done; /* Nothing to reclaim from */
807
808 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
809 /*
810 * Remove the node now but someone else can add it back,
811 * we will to add it back at the end of reclaim to its correct
812 * position in the tree.
813 */
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700814 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
815 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
816 !css_tryget(&mz->memcg->css))
Balbir Singh4e416952009-09-23 15:56:39 -0700817 goto retry;
818done:
819 return mz;
820}
821
822static struct mem_cgroup_per_zone *
823mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
824{
825 struct mem_cgroup_per_zone *mz;
826
827 spin_lock(&mctz->lock);
828 mz = __mem_cgroup_largest_soft_limit_node(mctz);
829 spin_unlock(&mctz->lock);
830 return mz;
831}
832
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700833/*
834 * Implementation Note: reading percpu statistics for memcg.
835 *
836 * Both of vmstat[] and percpu_counter has threshold and do periodic
837 * synchronization to implement "quick" read. There are trade-off between
838 * reading cost and precision of value. Then, we may have a chance to implement
839 * a periodic synchronizion of counter in memcg's counter.
840 *
841 * But this _read() function is used for user interface now. The user accounts
842 * memory usage by memory cgroup and he _always_ requires exact value because
843 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
844 * have to visit all online cpus and make sum. So, for now, unnecessary
845 * synchronization is not implemented. (just implemented for cpu hotplug)
846 *
847 * If there are kernel internal actions which can make use of some not-exact
848 * value, and reading all cpu value can be performance bottleneck in some
849 * common workload, threashold and synchonization as vmstat[] should be
850 * implemented.
851 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700852static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700853 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800854{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700855 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800856 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800857
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700858 get_online_cpus();
859 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700860 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700861#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700862 spin_lock(&memcg->pcp_counter_lock);
863 val += memcg->nocpu_base.count[idx];
864 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700865#endif
866 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800867 return val;
868}
869
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700870static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700871 bool charge)
872{
873 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700874 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700875}
876
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700877static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700878 enum mem_cgroup_events_index idx)
879{
880 unsigned long val = 0;
881 int cpu;
882
883 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700884 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700885#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700886 spin_lock(&memcg->pcp_counter_lock);
887 val += memcg->nocpu_base.events[idx];
888 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700889#endif
890 return val;
891}
892
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700893static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700894 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700895 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800896{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800897 preempt_disable();
898
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700899 /*
900 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
901 * counted as CACHE even if it's on ANON LRU.
902 */
903 if (anon)
904 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700905 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800906 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700907 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700908 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700909
David Rientjesb070e652013-05-07 16:18:09 -0700910 if (PageTransHuge(page))
911 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
912 nr_pages);
913
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800914 /* pagein of a big page is an event. So, ignore page size */
915 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700916 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800917 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700918 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800919 nr_pages = -nr_pages; /* for event */
920 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800921
Johannes Weiner13114712012-05-29 15:07:07 -0700922 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800923
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800924 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800925}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800926
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700927unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700928mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700929{
930 struct mem_cgroup_per_zone *mz;
931
932 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
933 return mz->lru_size[lru];
934}
935
936static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700937mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700938 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700939{
940 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700941 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700942 unsigned long ret = 0;
943
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700944 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700945
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700946 for_each_lru(lru) {
947 if (BIT(lru) & lru_mask)
948 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700949 }
950 return ret;
951}
952
953static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700954mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700955 int nid, unsigned int lru_mask)
956{
Ying Han889976d2011-05-26 16:25:33 -0700957 u64 total = 0;
958 int zid;
959
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700960 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700961 total += mem_cgroup_zone_nr_lru_pages(memcg,
962 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700963
Ying Han889976d2011-05-26 16:25:33 -0700964 return total;
965}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700966
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700967static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700968 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800969{
Ying Han889976d2011-05-26 16:25:33 -0700970 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800971 u64 total = 0;
972
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800973 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700974 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800975 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800976}
977
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800978static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
979 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800980{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700981 unsigned long val, next;
982
Johannes Weiner13114712012-05-29 15:07:07 -0700983 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700984 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700985 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800986 if ((long)next - (long)val < 0) {
987 switch (target) {
988 case MEM_CGROUP_TARGET_THRESH:
989 next = val + THRESHOLDS_EVENTS_TARGET;
990 break;
991 case MEM_CGROUP_TARGET_SOFTLIMIT:
992 next = val + SOFTLIMIT_EVENTS_TARGET;
993 break;
994 case MEM_CGROUP_TARGET_NUMAINFO:
995 next = val + NUMAINFO_EVENTS_TARGET;
996 break;
997 default:
998 break;
999 }
1000 __this_cpu_write(memcg->stat->targets[target], next);
1001 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001002 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001003 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001004}
1005
1006/*
1007 * Check events in order.
1008 *
1009 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001010static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001011{
Steven Rostedt47994012011-11-02 13:38:33 -07001012 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001013 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001014 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1015 MEM_CGROUP_TARGET_THRESH))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001016 bool do_softlimit;
1017 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001018
1019 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1020 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001021#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001022 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1023 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001024#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001025 preempt_enable();
1026
1027 mem_cgroup_threshold(memcg);
1028 if (unlikely(do_softlimit))
1029 mem_cgroup_update_tree(memcg, page);
1030#if MAX_NUMNODES > 1
1031 if (unlikely(do_numainfo))
1032 atomic_inc(&memcg->numainfo_events);
1033#endif
1034 } else
1035 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001036}
1037
Glauber Costad1a4c0b2011-12-11 21:47:04 +00001038struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001039{
Wanpeng Lib2145142012-07-31 16:46:01 -07001040 return mem_cgroup_from_css(
1041 cgroup_subsys_state(cont, mem_cgroup_subsys_id));
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001042}
1043
Balbir Singhcf475ad2008-04-29 01:00:16 -07001044struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001045{
Balbir Singh31a78f22008-09-28 23:09:31 +01001046 /*
1047 * mm_update_next_owner() may clear mm->owner to NULL
1048 * if it races with swapoff, page migration, etc.
1049 * So this can be called with p == NULL.
1050 */
1051 if (unlikely(!p))
1052 return NULL;
1053
Wanpeng Lib2145142012-07-31 16:46:01 -07001054 return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001055}
1056
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001057struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001058{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001059 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001060
1061 if (!mm)
1062 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001063 /*
1064 * Because we have no locks, mm->owner's may be being moved to other
1065 * cgroup. We use css_tryget() here even if this looks
1066 * pessimistic (rather than adding locks here).
1067 */
1068 rcu_read_lock();
1069 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001070 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1071 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001072 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001073 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001074 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001075 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001076}
1077
Michal Hocko16248d82013-04-29 15:07:19 -07001078/*
1079 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1080 * ref. count) or NULL if the whole root's subtree has been visited.
1081 *
1082 * helper function to be used by mem_cgroup_iter
1083 */
1084static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
1085 struct mem_cgroup *last_visited)
1086{
1087 struct cgroup *prev_cgroup, *next_cgroup;
1088
1089 /*
1090 * Root is not visited by cgroup iterators so it needs an
1091 * explicit visit.
1092 */
1093 if (!last_visited)
1094 return root;
1095
1096 prev_cgroup = (last_visited == root) ? NULL
1097 : last_visited->css.cgroup;
1098skip_node:
1099 next_cgroup = cgroup_next_descendant_pre(
1100 prev_cgroup, root->css.cgroup);
1101
1102 /*
1103 * Even if we found a group we have to make sure it is
1104 * alive. css && !memcg means that the groups should be
1105 * skipped and we should continue the tree walk.
1106 * last_visited css is safe to use because it is
1107 * protected by css_get and the tree walk is rcu safe.
1108 */
1109 if (next_cgroup) {
1110 struct mem_cgroup *mem = mem_cgroup_from_cont(
1111 next_cgroup);
1112 if (css_tryget(&mem->css))
1113 return mem;
1114 else {
1115 prev_cgroup = next_cgroup;
1116 goto skip_node;
1117 }
1118 }
1119
1120 return NULL;
1121}
1122
Johannes Weiner519ebea2013-07-03 15:04:51 -07001123static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1124{
1125 /*
1126 * When a group in the hierarchy below root is destroyed, the
1127 * hierarchy iterator can no longer be trusted since it might
1128 * have pointed to the destroyed group. Invalidate it.
1129 */
1130 atomic_inc(&root->dead_count);
1131}
1132
1133static struct mem_cgroup *
1134mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1135 struct mem_cgroup *root,
1136 int *sequence)
1137{
1138 struct mem_cgroup *position = NULL;
1139 /*
1140 * A cgroup destruction happens in two stages: offlining and
1141 * release. They are separated by a RCU grace period.
1142 *
1143 * If the iterator is valid, we may still race with an
1144 * offlining. The RCU lock ensures the object won't be
1145 * released, tryget will fail if we lost the race.
1146 */
1147 *sequence = atomic_read(&root->dead_count);
1148 if (iter->last_dead_count == *sequence) {
1149 smp_rmb();
1150 position = iter->last_visited;
1151 if (position && !css_tryget(&position->css))
1152 position = NULL;
1153 }
1154 return position;
1155}
1156
1157static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1158 struct mem_cgroup *last_visited,
1159 struct mem_cgroup *new_position,
1160 int sequence)
1161{
1162 if (last_visited)
1163 css_put(&last_visited->css);
1164 /*
1165 * We store the sequence count from the time @last_visited was
1166 * loaded successfully instead of rereading it here so that we
1167 * don't lose destruction events in between. We could have
1168 * raced with the destruction of @new_position after all.
1169 */
1170 iter->last_visited = new_position;
1171 smp_wmb();
1172 iter->last_dead_count = sequence;
1173}
1174
Johannes Weiner56600482012-01-12 17:17:59 -08001175/**
1176 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1177 * @root: hierarchy root
1178 * @prev: previously returned memcg, NULL on first invocation
1179 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1180 *
1181 * Returns references to children of the hierarchy below @root, or
1182 * @root itself, or %NULL after a full round-trip.
1183 *
1184 * Caller must pass the return value in @prev on subsequent
1185 * invocations for reference counting, or use mem_cgroup_iter_break()
1186 * to cancel a hierarchy walk before the round-trip is complete.
1187 *
1188 * Reclaimers can specify a zone and a priority level in @reclaim to
1189 * divide up the memcgs in the hierarchy among all concurrent
1190 * reclaimers operating on the same zone and priority.
1191 */
1192struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
1193 struct mem_cgroup *prev,
1194 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001195{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001196 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001197 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001198
Johannes Weiner56600482012-01-12 17:17:59 -08001199 if (mem_cgroup_disabled())
1200 return NULL;
1201
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001202 if (!root)
1203 root = root_mem_cgroup;
1204
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001205 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001206 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001207
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001208 if (!root->use_hierarchy && root != root_mem_cgroup) {
1209 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001210 goto out_css_put;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001211 return root;
1212 }
1213
Michal Hocko542f85f2013-04-29 15:07:15 -07001214 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001215 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001216 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001217 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001218
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001219 if (reclaim) {
1220 int nid = zone_to_nid(reclaim->zone);
1221 int zid = zone_idx(reclaim->zone);
1222 struct mem_cgroup_per_zone *mz;
1223
1224 mz = mem_cgroup_zoneinfo(root, nid, zid);
1225 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001226 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001227 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001228 goto out_unlock;
1229 }
Michal Hocko5f578162013-04-29 15:07:17 -07001230
Johannes Weiner519ebea2013-07-03 15:04:51 -07001231 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001232 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001233
Michal Hocko16248d82013-04-29 15:07:19 -07001234 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001235
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001236 if (reclaim) {
Johannes Weiner519ebea2013-07-03 15:04:51 -07001237 mem_cgroup_iter_update(iter, last_visited, memcg, seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001238
Michal Hocko19f39402013-04-29 15:07:18 -07001239 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001240 iter->generation++;
1241 else if (!prev && memcg)
1242 reclaim->generation = iter->generation;
1243 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001244
Michal Hocko19f39402013-04-29 15:07:18 -07001245 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001246 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001247 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001248out_unlock:
1249 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001250out_css_put:
1251 if (prev && prev != root)
1252 css_put(&prev->css);
1253
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001254 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001255}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001256
Johannes Weiner56600482012-01-12 17:17:59 -08001257/**
1258 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1259 * @root: hierarchy root
1260 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1261 */
1262void mem_cgroup_iter_break(struct mem_cgroup *root,
1263 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001264{
1265 if (!root)
1266 root = root_mem_cgroup;
1267 if (prev && prev != root)
1268 css_put(&prev->css);
1269}
1270
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001271/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001272 * Iteration constructs for visiting all cgroups (under a tree). If
1273 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1274 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001275 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001276#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001277 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001278 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001279 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001280
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001281#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001282 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001283 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001284 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001285
David Rientjes68ae5642012-12-12 13:51:57 -08001286void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001287{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001289
Ying Han456f9982011-05-26 16:25:38 -07001290 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001291 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1292 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001293 goto out;
1294
1295 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001296 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001297 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1298 break;
1299 case PGMAJFAULT:
1300 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001301 break;
1302 default:
1303 BUG();
1304 }
1305out:
1306 rcu_read_unlock();
1307}
David Rientjes68ae5642012-12-12 13:51:57 -08001308EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001309
Johannes Weiner925b7672012-01-12 17:18:15 -08001310/**
1311 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1312 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001313 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001314 *
1315 * Returns the lru list vector holding pages for the given @zone and
1316 * @mem. This can be the global zone lruvec, if the memory controller
1317 * is disabled.
1318 */
1319struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1320 struct mem_cgroup *memcg)
1321{
1322 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001323 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001324
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001325 if (mem_cgroup_disabled()) {
1326 lruvec = &zone->lruvec;
1327 goto out;
1328 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001329
1330 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001331 lruvec = &mz->lruvec;
1332out:
1333 /*
1334 * Since a node can be onlined after the mem_cgroup was created,
1335 * we have to be prepared to initialize lruvec->zone here;
1336 * and if offlined then reonlined, we need to reinitialize it.
1337 */
1338 if (unlikely(lruvec->zone != zone))
1339 lruvec->zone = zone;
1340 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001341}
1342
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001343/*
1344 * Following LRU functions are allowed to be used without PCG_LOCK.
1345 * Operations are called by routine of global LRU independently from memcg.
1346 * What we have to take care of here is validness of pc->mem_cgroup.
1347 *
1348 * Changes to pc->mem_cgroup happens when
1349 * 1. charge
1350 * 2. moving account
1351 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1352 * It is added to LRU before charge.
1353 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1354 * When moving account, the page is not on LRU. It's isolated.
1355 */
1356
Johannes Weiner925b7672012-01-12 17:18:15 -08001357/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001358 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001359 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001360 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001361 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001362struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001363{
1364 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001365 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001366 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001367 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001368
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001369 if (mem_cgroup_disabled()) {
1370 lruvec = &zone->lruvec;
1371 goto out;
1372 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001373
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001374 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001375 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001376
1377 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001378 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001379 * an uncharged page off lru does nothing to secure
1380 * its former mem_cgroup from sudden removal.
1381 *
1382 * Our caller holds lru_lock, and PageCgroupUsed is updated
1383 * under page_cgroup lock: between them, they make all uses
1384 * of pc->mem_cgroup safe.
1385 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001386 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001387 pc->mem_cgroup = memcg = root_mem_cgroup;
1388
Johannes Weiner925b7672012-01-12 17:18:15 -08001389 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001390 lruvec = &mz->lruvec;
1391out:
1392 /*
1393 * Since a node can be onlined after the mem_cgroup was created,
1394 * we have to be prepared to initialize lruvec->zone here;
1395 * and if offlined then reonlined, we need to reinitialize it.
1396 */
1397 if (unlikely(lruvec->zone != zone))
1398 lruvec->zone = zone;
1399 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001400}
1401
1402/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001403 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1404 * @lruvec: mem_cgroup per zone lru vector
1405 * @lru: index of lru list the page is sitting on
1406 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001407 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001408 * This function must be called when a page is added to or removed from an
1409 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001410 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001411void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1412 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001413{
1414 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001415 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001416
1417 if (mem_cgroup_disabled())
1418 return;
1419
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001420 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1421 lru_size = mz->lru_size + lru;
1422 *lru_size += nr_pages;
1423 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001424}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001425
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001426/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001427 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001428 * hierarchy subtree
1429 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001430bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1431 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001432{
Johannes Weiner91c637342012-05-29 15:06:24 -07001433 if (root_memcg == memcg)
1434 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001435 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001436 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001437 return css_is_ancestor(&memcg->css, &root_memcg->css);
1438}
1439
1440static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1441 struct mem_cgroup *memcg)
1442{
1443 bool ret;
1444
Johannes Weiner91c637342012-05-29 15:06:24 -07001445 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001446 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001447 rcu_read_unlock();
1448 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001449}
1450
David Rientjesffbdccf2013-07-03 15:01:23 -07001451bool task_in_mem_cgroup(struct task_struct *task,
1452 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001453{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001454 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001455 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001456 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001457
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001458 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001459 if (p) {
1460 curr = try_get_mem_cgroup_from_mm(p->mm);
1461 task_unlock(p);
1462 } else {
1463 /*
1464 * All threads may have already detached their mm's, but the oom
1465 * killer still needs to detect if they have already been oom
1466 * killed to prevent needlessly killing additional tasks.
1467 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001468 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001469 curr = mem_cgroup_from_task(task);
1470 if (curr)
1471 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001472 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001473 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001474 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001475 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001476 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001477 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001478 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001479 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1480 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001481 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001482 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001483 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001484 return ret;
1485}
1486
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001487int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001488{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001489 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001490 unsigned long inactive;
1491 unsigned long active;
1492 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001493
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001494 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1495 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001496
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001497 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1498 if (gb)
1499 inactive_ratio = int_sqrt(10 * gb);
1500 else
1501 inactive_ratio = 1;
1502
Johannes Weiner9b272972011-11-02 13:38:23 -07001503 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001504}
1505
Balbir Singh6d61ef42009-01-07 18:08:06 -08001506#define mem_cgroup_from_res_counter(counter, member) \
1507 container_of(counter, struct mem_cgroup, member)
1508
Johannes Weiner19942822011-02-01 15:52:43 -08001509/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001510 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001511 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001512 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001513 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001514 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001515 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001516static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001517{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001518 unsigned long long margin;
1519
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001520 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001521 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001522 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001523 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001524}
1525
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001526int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001527{
1528 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001529
1530 /* root ? */
1531 if (cgrp->parent == NULL)
1532 return vm_swappiness;
1533
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001534 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001535}
1536
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001537/*
1538 * memcg->moving_account is used for checking possibility that some thread is
1539 * calling move_account(). When a thread on CPU-A starts moving pages under
1540 * a memcg, other threads should check memcg->moving_account under
1541 * rcu_read_lock(), like this:
1542 *
1543 * CPU-A CPU-B
1544 * rcu_read_lock()
1545 * memcg->moving_account+1 if (memcg->mocing_account)
1546 * take heavy locks.
1547 * synchronize_rcu() update something.
1548 * rcu_read_unlock()
1549 * start move here.
1550 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001551
1552/* for quick checking without looking up memcg */
1553atomic_t memcg_moving __read_mostly;
1554
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001555static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001556{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001557 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001558 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001559 synchronize_rcu();
1560}
1561
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001562static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001563{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001564 /*
1565 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1566 * We check NULL in callee rather than caller.
1567 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001568 if (memcg) {
1569 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001570 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001571 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001572}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001573
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001574/*
1575 * 2 routines for checking "mem" is under move_account() or not.
1576 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001577 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1578 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001579 * pc->mem_cgroup may be overwritten.
1580 *
1581 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1582 * under hierarchy of moving cgroups. This is for
1583 * waiting at hith-memory prressure caused by "move".
1584 */
1585
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001586static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001587{
1588 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001589 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001590}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001591
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001592static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001593{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001594 struct mem_cgroup *from;
1595 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001596 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001597 /*
1598 * Unlike task_move routines, we access mc.to, mc.from not under
1599 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1600 */
1601 spin_lock(&mc.lock);
1602 from = mc.from;
1603 to = mc.to;
1604 if (!from)
1605 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001606
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001607 ret = mem_cgroup_same_or_subtree(memcg, from)
1608 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001609unlock:
1610 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001611 return ret;
1612}
1613
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001614static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001615{
1616 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001617 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001618 DEFINE_WAIT(wait);
1619 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1620 /* moving charge context might have finished. */
1621 if (mc.moving_task)
1622 schedule();
1623 finish_wait(&mc.waitq, &wait);
1624 return true;
1625 }
1626 }
1627 return false;
1628}
1629
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001630/*
1631 * Take this lock when
1632 * - a code tries to modify page's memcg while it's USED.
1633 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001634 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001635 */
1636static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1637 unsigned long *flags)
1638{
1639 spin_lock_irqsave(&memcg->move_lock, *flags);
1640}
1641
1642static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1643 unsigned long *flags)
1644{
1645 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1646}
1647
Sha Zhengju58cf1882013-02-22 16:32:05 -08001648#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001649/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001650 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001651 * @memcg: The memory cgroup that went over limit
1652 * @p: Task that is going to be killed
1653 *
1654 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1655 * enabled
1656 */
1657void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1658{
1659 struct cgroup *task_cgrp;
1660 struct cgroup *mem_cgrp;
1661 /*
1662 * Need a buffer in BSS, can't rely on allocations. The code relies
1663 * on the assumption that OOM is serialized for memory controller.
1664 * If this assumption is broken, revisit this code.
1665 */
1666 static char memcg_name[PATH_MAX];
1667 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001668 struct mem_cgroup *iter;
1669 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001670
Sha Zhengju58cf1882013-02-22 16:32:05 -08001671 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001672 return;
1673
Balbir Singhe2224322009-04-02 16:57:39 -07001674 rcu_read_lock();
1675
1676 mem_cgrp = memcg->css.cgroup;
1677 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1678
1679 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1680 if (ret < 0) {
1681 /*
1682 * Unfortunately, we are unable to convert to a useful name
1683 * But we'll still print out the usage information
1684 */
1685 rcu_read_unlock();
1686 goto done;
1687 }
1688 rcu_read_unlock();
1689
Andrew Mortond0451972013-02-22 16:32:06 -08001690 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001691
1692 rcu_read_lock();
1693 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1694 if (ret < 0) {
1695 rcu_read_unlock();
1696 goto done;
1697 }
1698 rcu_read_unlock();
1699
1700 /*
1701 * Continues from above, so we don't need an KERN_ level
1702 */
Andrew Mortond0451972013-02-22 16:32:06 -08001703 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001704done:
1705
Andrew Mortond0451972013-02-22 16:32:06 -08001706 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001707 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1708 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1709 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001710 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001711 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1712 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1713 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001714 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001715 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1716 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1717 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001718
1719 for_each_mem_cgroup_tree(iter, memcg) {
1720 pr_info("Memory cgroup stats");
1721
1722 rcu_read_lock();
1723 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1724 if (!ret)
1725 pr_cont(" for %s", memcg_name);
1726 rcu_read_unlock();
1727 pr_cont(":");
1728
1729 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1730 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1731 continue;
1732 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1733 K(mem_cgroup_read_stat(iter, i)));
1734 }
1735
1736 for (i = 0; i < NR_LRU_LISTS; i++)
1737 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1738 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1739
1740 pr_cont("\n");
1741 }
Balbir Singhe2224322009-04-02 16:57:39 -07001742}
1743
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001744/*
1745 * This function returns the number of memcg under hierarchy tree. Returns
1746 * 1(self count) if no children.
1747 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001748static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001749{
1750 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001751 struct mem_cgroup *iter;
1752
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001753 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001754 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001755 return num;
1756}
1757
Balbir Singh6d61ef42009-01-07 18:08:06 -08001758/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001759 * Return the memory (and swap, if configured) limit for a memcg.
1760 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001761static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001762{
1763 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001764
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001765 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001766
David Rientjesa63d83f2010-08-09 17:19:46 -07001767 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001768 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001769 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001770 if (mem_cgroup_swappiness(memcg)) {
1771 u64 memsw;
1772
1773 limit += total_swap_pages << PAGE_SHIFT;
1774 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1775
1776 /*
1777 * If memsw is finite and limits the amount of swap space
1778 * available to this memcg, return that limit.
1779 */
1780 limit = min(limit, memsw);
1781 }
1782
1783 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001784}
1785
David Rientjes19965462012-12-11 16:00:26 -08001786static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1787 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001788{
1789 struct mem_cgroup *iter;
1790 unsigned long chosen_points = 0;
1791 unsigned long totalpages;
1792 unsigned int points = 0;
1793 struct task_struct *chosen = NULL;
1794
David Rientjes876aafb2012-07-31 16:43:48 -07001795 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001796 * If current has a pending SIGKILL or is exiting, then automatically
1797 * select it. The goal is to allow it to allocate so that it may
1798 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001799 */
David Rientjes465adcf2013-04-29 15:08:45 -07001800 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001801 set_thread_flag(TIF_MEMDIE);
1802 return;
1803 }
1804
1805 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001806 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1807 for_each_mem_cgroup_tree(iter, memcg) {
1808 struct cgroup *cgroup = iter->css.cgroup;
1809 struct cgroup_iter it;
1810 struct task_struct *task;
1811
1812 cgroup_iter_start(cgroup, &it);
1813 while ((task = cgroup_iter_next(cgroup, &it))) {
1814 switch (oom_scan_process_thread(task, totalpages, NULL,
1815 false)) {
1816 case OOM_SCAN_SELECT:
1817 if (chosen)
1818 put_task_struct(chosen);
1819 chosen = task;
1820 chosen_points = ULONG_MAX;
1821 get_task_struct(chosen);
1822 /* fall through */
1823 case OOM_SCAN_CONTINUE:
1824 continue;
1825 case OOM_SCAN_ABORT:
1826 cgroup_iter_end(cgroup, &it);
1827 mem_cgroup_iter_break(memcg, iter);
1828 if (chosen)
1829 put_task_struct(chosen);
1830 return;
1831 case OOM_SCAN_OK:
1832 break;
1833 };
1834 points = oom_badness(task, memcg, NULL, totalpages);
1835 if (points > chosen_points) {
1836 if (chosen)
1837 put_task_struct(chosen);
1838 chosen = task;
1839 chosen_points = points;
1840 get_task_struct(chosen);
1841 }
1842 }
1843 cgroup_iter_end(cgroup, &it);
1844 }
1845
1846 if (!chosen)
1847 return;
1848 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001849 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1850 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001851}
1852
Johannes Weiner56600482012-01-12 17:17:59 -08001853static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1854 gfp_t gfp_mask,
1855 unsigned long flags)
1856{
1857 unsigned long total = 0;
1858 bool noswap = false;
1859 int loop;
1860
1861 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1862 noswap = true;
1863 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1864 noswap = true;
1865
1866 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1867 if (loop)
1868 drain_all_stock_async(memcg);
1869 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1870 /*
1871 * Allow limit shrinkers, which are triggered directly
1872 * by userspace, to catch signals and stop reclaim
1873 * after minimal progress, regardless of the margin.
1874 */
1875 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1876 break;
1877 if (mem_cgroup_margin(memcg))
1878 break;
1879 /*
1880 * If nothing was reclaimed after two attempts, there
1881 * may be no reclaimable pages in this hierarchy.
1882 */
1883 if (loop && !total)
1884 break;
1885 }
1886 return total;
1887}
1888
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001889/**
1890 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001891 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001892 * @nid: the node ID to be checked.
1893 * @noswap : specify true here if the user wants flle only information.
1894 *
1895 * This function returns whether the specified memcg contains any
1896 * reclaimable pages on a node. Returns true if there are any reclaimable
1897 * pages in the node.
1898 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001899static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001900 int nid, bool noswap)
1901{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001902 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001903 return true;
1904 if (noswap || !total_swap_pages)
1905 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001906 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001907 return true;
1908 return false;
1909
1910}
Ying Han889976d2011-05-26 16:25:33 -07001911#if MAX_NUMNODES > 1
1912
1913/*
1914 * Always updating the nodemask is not very good - even if we have an empty
1915 * list or the wrong list here, we can start from some node and traverse all
1916 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1917 *
1918 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001919static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001920{
1921 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001922 /*
1923 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1924 * pagein/pageout changes since the last update.
1925 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001927 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001928 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001929 return;
1930
Ying Han889976d2011-05-26 16:25:33 -07001931 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001932 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001933
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001934 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001935
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001936 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1937 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001938 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001939
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001940 atomic_set(&memcg->numainfo_events, 0);
1941 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001942}
1943
1944/*
1945 * Selecting a node where we start reclaim from. Because what we need is just
1946 * reducing usage counter, start from anywhere is O,K. Considering
1947 * memory reclaim from current node, there are pros. and cons.
1948 *
1949 * Freeing memory from current node means freeing memory from a node which
1950 * we'll use or we've used. So, it may make LRU bad. And if several threads
1951 * hit limits, it will see a contention on a node. But freeing from remote
1952 * node means more costs for memory reclaim because of memory latency.
1953 *
1954 * Now, we use round-robin. Better algorithm is welcomed.
1955 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001956int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001957{
1958 int node;
1959
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001960 mem_cgroup_may_update_nodemask(memcg);
1961 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001962
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001963 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001964 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001965 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001966 /*
1967 * We call this when we hit limit, not when pages are added to LRU.
1968 * No LRU may hold pages because all pages are UNEVICTABLE or
1969 * memcg is too small and all pages are not on LRU. In that case,
1970 * we use curret node.
1971 */
1972 if (unlikely(node == MAX_NUMNODES))
1973 node = numa_node_id();
1974
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001975 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001976 return node;
1977}
1978
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001979/*
1980 * Check all nodes whether it contains reclaimable pages or not.
1981 * For quick scan, we make use of scan_nodes. This will allow us to skip
1982 * unused nodes. But scan_nodes is lazily updated and may not cotain
1983 * enough new information. We need to do double check.
1984 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07001985static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001986{
1987 int nid;
1988
1989 /*
1990 * quick check...making use of scan_node.
1991 * We can skip unused nodes.
1992 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001993 if (!nodes_empty(memcg->scan_nodes)) {
1994 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001995 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001996 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001997
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001998 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001999 return true;
2000 }
2001 }
2002 /*
2003 * Check rest of nodes.
2004 */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08002005 for_each_node_state(nid, N_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002006 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002007 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002008 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002009 return true;
2010 }
2011 return false;
2012}
2013
Ying Han889976d2011-05-26 16:25:33 -07002014#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002015int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002016{
2017 return 0;
2018}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002019
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002020static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002021{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002022 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002023}
Ying Han889976d2011-05-26 16:25:33 -07002024#endif
2025
Johannes Weiner56600482012-01-12 17:17:59 -08002026static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2027 struct zone *zone,
2028 gfp_t gfp_mask,
2029 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002030{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002031 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08002032 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002033 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002034 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07002035 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08002036 struct mem_cgroup_reclaim_cookie reclaim = {
2037 .zone = zone,
2038 .priority = 0,
2039 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002040
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002041 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002042
Balbir Singh4e416952009-09-23 15:56:39 -07002043 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08002044 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002045 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002046 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07002047 if (loop >= 2) {
2048 /*
2049 * If we have not been able to reclaim
2050 * anything, it might because there are
2051 * no reclaimable pages under this hierarchy
2052 */
Johannes Weiner56600482012-01-12 17:17:59 -08002053 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07002054 break;
Balbir Singh4e416952009-09-23 15:56:39 -07002055 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002056 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07002057 * excess >> 2 is not to excessive so as to
2058 * reclaim too much, nor too less that we keep
2059 * coming back to reclaim from this cgroup
2060 */
2061 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002062 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07002063 break;
Balbir Singh4e416952009-09-23 15:56:39 -07002064 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002065 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07002066 }
Johannes Weiner56600482012-01-12 17:17:59 -08002067 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08002068 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08002069 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2070 zone, &nr_scanned);
2071 *total_scanned += nr_scanned;
2072 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002073 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002074 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002075 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002076 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002077}
2078
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002079/*
2080 * Check OOM-Killer is already running under our hierarchy.
2081 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07002082 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002083 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002084static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002085{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002086 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002087
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002088 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002089 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002090 /*
2091 * this subtree of our hierarchy is already locked
2092 * so we cannot give a lock.
2093 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002094 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002095 mem_cgroup_iter_break(memcg, iter);
2096 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002097 } else
2098 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002099 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002100
Michal Hocko79dfdac2011-07-26 16:08:23 -07002101 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07002102 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002103
2104 /*
2105 * OK, we failed to lock the whole subtree so we have to clean up
2106 * what we set up to the failing subtree
2107 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002108 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002109 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002110 mem_cgroup_iter_break(memcg, iter);
2111 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002112 }
2113 iter->oom_lock = false;
2114 }
Johannes Weiner23751be2011-08-25 15:59:16 -07002115 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002116}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002117
Michal Hocko79dfdac2011-07-26 16:08:23 -07002118/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07002119 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07002120 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002121static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002122{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002123 struct mem_cgroup *iter;
2124
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002125 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002126 iter->oom_lock = false;
2127 return 0;
2128}
2129
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002130static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002131{
2132 struct mem_cgroup *iter;
2133
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002134 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002135 atomic_inc(&iter->under_oom);
2136}
2137
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002138static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002139{
2140 struct mem_cgroup *iter;
2141
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002142 /*
2143 * When a new child is created while the hierarchy is under oom,
2144 * mem_cgroup_oom_lock() may not be called. We have to use
2145 * atomic_add_unless() here.
2146 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002147 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002148 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002149}
2150
Michal Hocko1af8efe2011-07-26 16:08:24 -07002151static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002152static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2153
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002154struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002155 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002156 wait_queue_t wait;
2157};
2158
2159static int memcg_oom_wake_function(wait_queue_t *wait,
2160 unsigned mode, int sync, void *arg)
2161{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002162 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2163 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002164 struct oom_wait_info *oom_wait_info;
2165
2166 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002167 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002168
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002169 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002170 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002171 * Then we can use css_is_ancestor without taking care of RCU.
2172 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002173 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2174 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002175 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002176 return autoremove_wake_function(wait, mode, sync, arg);
2177}
2178
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002179static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002180{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002181 /* for filtering, pass "memcg" as argument. */
2182 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002183}
2184
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002185static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002186{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002187 if (memcg && atomic_read(&memcg->under_oom))
2188 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002189}
2190
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002191/*
2192 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
2193 */
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07002194static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
2195 int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002196{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002197 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002198 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002199
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002200 owait.memcg = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002201 owait.wait.flags = 0;
2202 owait.wait.func = memcg_oom_wake_function;
2203 owait.wait.private = current;
2204 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002205 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002206 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002207
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002208 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07002209 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002210 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002211 /*
2212 * Even if signal_pending(), we can't quit charge() loop without
2213 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
2214 * under OOM is always welcomed, use TASK_KILLABLE here.
2215 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002216 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002217 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002218 need_to_kill = false;
2219 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002220 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002221 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002222
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002223 if (need_to_kill) {
2224 finish_wait(&memcg_oom_waitq, &owait.wait);
David Rientjese845e192012-03-21 16:34:10 -07002225 mem_cgroup_out_of_memory(memcg, mask, order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002226 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002227 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002228 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002229 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07002230 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002231 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002232 mem_cgroup_oom_unlock(memcg);
2233 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07002234 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002235
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002236 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002237
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002238 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
2239 return false;
2240 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07002241 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002242 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002243}
2244
Balbir Singhd69b0422009-06-17 16:26:34 -07002245/*
2246 * Currently used to update mapped file statistics, but the routine can be
2247 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002248 *
2249 * Notes: Race condition
2250 *
2251 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2252 * it tends to be costly. But considering some conditions, we doesn't need
2253 * to do so _always_.
2254 *
2255 * Considering "charge", lock_page_cgroup() is not required because all
2256 * file-stat operations happen after a page is attached to radix-tree. There
2257 * are no race with "charge".
2258 *
2259 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2260 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2261 * if there are race with "uncharge". Statistics itself is properly handled
2262 * by flags.
2263 *
2264 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002265 * small, we check mm->moving_account and detect there are possibility of race
2266 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002267 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002268
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002269void __mem_cgroup_begin_update_page_stat(struct page *page,
2270 bool *locked, unsigned long *flags)
2271{
2272 struct mem_cgroup *memcg;
2273 struct page_cgroup *pc;
2274
2275 pc = lookup_page_cgroup(page);
2276again:
2277 memcg = pc->mem_cgroup;
2278 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2279 return;
2280 /*
2281 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002282 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002283 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002284 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002285 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002286 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002287 return;
2288
2289 move_lock_mem_cgroup(memcg, flags);
2290 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2291 move_unlock_mem_cgroup(memcg, flags);
2292 goto again;
2293 }
2294 *locked = true;
2295}
2296
2297void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2298{
2299 struct page_cgroup *pc = lookup_page_cgroup(page);
2300
2301 /*
2302 * It's guaranteed that pc->mem_cgroup never changes while
2303 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002304 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002305 */
2306 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2307}
2308
Greg Thelen2a7106f2011-01-13 15:47:37 -08002309void mem_cgroup_update_page_stat(struct page *page,
2310 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002311{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002312 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002313 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002314 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002315
Johannes Weinercfa44942012-01-12 17:18:38 -08002316 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002317 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002318
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002319 memcg = pc->mem_cgroup;
2320 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002321 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002322
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002323 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08002324 case MEMCG_NR_FILE_MAPPED:
Greg Thelen2a7106f2011-01-13 15:47:37 -08002325 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002326 break;
2327 default:
2328 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002329 }
Balbir Singhd69b0422009-06-17 16:26:34 -07002330
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002331 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002332}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002333
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002334/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002335 * size of first charge trial. "32" comes from vmscan.c's magic value.
2336 * TODO: maybe necessary to use big numbers in big irons.
2337 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002338#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002339struct memcg_stock_pcp {
2340 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002341 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002342 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002343 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002344#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002345};
2346static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002347static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002348
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002349/**
2350 * consume_stock: Try to consume stocked charge on this cpu.
2351 * @memcg: memcg to consume from.
2352 * @nr_pages: how many pages to charge.
2353 *
2354 * The charges will only happen if @memcg matches the current cpu's memcg
2355 * stock, and at least @nr_pages are available in that stock. Failure to
2356 * service an allocation will refill the stock.
2357 *
2358 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002359 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002360static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002361{
2362 struct memcg_stock_pcp *stock;
2363 bool ret = true;
2364
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002365 if (nr_pages > CHARGE_BATCH)
2366 return false;
2367
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002368 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002369 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2370 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002371 else /* need to call res_counter_charge */
2372 ret = false;
2373 put_cpu_var(memcg_stock);
2374 return ret;
2375}
2376
2377/*
2378 * Returns stocks cached in percpu to res_counter and reset cached information.
2379 */
2380static void drain_stock(struct memcg_stock_pcp *stock)
2381{
2382 struct mem_cgroup *old = stock->cached;
2383
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002384 if (stock->nr_pages) {
2385 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2386
2387 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002388 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002389 res_counter_uncharge(&old->memsw, bytes);
2390 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002391 }
2392 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002393}
2394
2395/*
2396 * This must be called under preempt disabled or must be called by
2397 * a thread which is pinned to local cpu.
2398 */
2399static void drain_local_stock(struct work_struct *dummy)
2400{
2401 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2402 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002403 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002404}
2405
Michal Hockoe4777492013-02-22 16:35:40 -08002406static void __init memcg_stock_init(void)
2407{
2408 int cpu;
2409
2410 for_each_possible_cpu(cpu) {
2411 struct memcg_stock_pcp *stock =
2412 &per_cpu(memcg_stock, cpu);
2413 INIT_WORK(&stock->work, drain_local_stock);
2414 }
2415}
2416
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002417/*
2418 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002419 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002420 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002421static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002422{
2423 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2424
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002425 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002426 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002427 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002428 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002429 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002430 put_cpu_var(memcg_stock);
2431}
2432
2433/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002434 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002435 * of the hierarchy under it. sync flag says whether we should block
2436 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002437 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002438static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002439{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002440 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002441
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002442 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002443 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002444 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002445 for_each_online_cpu(cpu) {
2446 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002447 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002448
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002449 memcg = stock->cached;
2450 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002451 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002452 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002453 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002454 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2455 if (cpu == curcpu)
2456 drain_local_stock(&stock->work);
2457 else
2458 schedule_work_on(cpu, &stock->work);
2459 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002460 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002461 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002462
2463 if (!sync)
2464 goto out;
2465
2466 for_each_online_cpu(cpu) {
2467 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002468 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002469 flush_work(&stock->work);
2470 }
2471out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002472 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002473}
2474
2475/*
2476 * Tries to drain stocked charges in other cpus. This function is asynchronous
2477 * and just put a work per cpu for draining localy on each cpu. Caller can
2478 * expects some charges will be back to res_counter later but cannot wait for
2479 * it.
2480 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002481static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002482{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002483 /*
2484 * If someone calls draining, avoid adding more kworker runs.
2485 */
2486 if (!mutex_trylock(&percpu_charge_mutex))
2487 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002488 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002489 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002490}
2491
2492/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002493static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002494{
2495 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002496 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002497 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002498 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002499}
2500
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002501/*
2502 * This function drains percpu counter value from DEAD cpu and
2503 * move it to local cpu. Note that this function can be preempted.
2504 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002505static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002506{
2507 int i;
2508
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002509 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002510 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002511 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002512
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002513 per_cpu(memcg->stat->count[i], cpu) = 0;
2514 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002515 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002516 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002517 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002519 per_cpu(memcg->stat->events[i], cpu) = 0;
2520 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002521 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002522 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002523}
2524
2525static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002526 unsigned long action,
2527 void *hcpu)
2528{
2529 int cpu = (unsigned long)hcpu;
2530 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002531 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002532
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002533 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002534 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002535
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002536 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002537 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002538
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002539 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002540 mem_cgroup_drain_pcp_counter(iter, cpu);
2541
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002542 stock = &per_cpu(memcg_stock, cpu);
2543 drain_stock(stock);
2544 return NOTIFY_OK;
2545}
2546
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002547
2548/* See __mem_cgroup_try_charge() for details */
2549enum {
2550 CHARGE_OK, /* success */
2551 CHARGE_RETRY, /* need to retry but retry is not bad */
2552 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2553 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2554 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2555};
2556
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002557static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002558 unsigned int nr_pages, unsigned int min_pages,
2559 bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002560{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002561 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002562 struct mem_cgroup *mem_over_limit;
2563 struct res_counter *fail_res;
2564 unsigned long flags = 0;
2565 int ret;
2566
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002567 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002568
2569 if (likely(!ret)) {
2570 if (!do_swap_account)
2571 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002572 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002573 if (likely(!ret))
2574 return CHARGE_OK;
2575
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002576 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002577 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2578 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2579 } else
2580 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002581 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002582 * Never reclaim on behalf of optional batching, retry with a
2583 * single page instead.
2584 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002585 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002586 return CHARGE_RETRY;
2587
2588 if (!(gfp_mask & __GFP_WAIT))
2589 return CHARGE_WOULDBLOCK;
2590
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002591 if (gfp_mask & __GFP_NORETRY)
2592 return CHARGE_NOMEM;
2593
Johannes Weiner56600482012-01-12 17:17:59 -08002594 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002595 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002596 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002597 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002598 * Even though the limit is exceeded at this point, reclaim
2599 * may have been able to free some pages. Retry the charge
2600 * before killing the task.
2601 *
2602 * Only for regular pages, though: huge pages are rather
2603 * unlikely to succeed so close to the limit, and we fall back
2604 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002605 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002606 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002607 return CHARGE_RETRY;
2608
2609 /*
2610 * At task move, charge accounts can be doubly counted. So, it's
2611 * better to wait until the end of task_move if something is going on.
2612 */
2613 if (mem_cgroup_wait_acct_move(mem_over_limit))
2614 return CHARGE_RETRY;
2615
2616 /* If we don't need to call oom-killer at el, return immediately */
2617 if (!oom_check)
2618 return CHARGE_NOMEM;
2619 /* check OOM */
David Rientjese845e192012-03-21 16:34:10 -07002620 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002621 return CHARGE_OOM_DIE;
2622
2623 return CHARGE_RETRY;
2624}
2625
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002626/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002627 * __mem_cgroup_try_charge() does
2628 * 1. detect memcg to be charged against from passed *mm and *ptr,
2629 * 2. update res_counter
2630 * 3. call memory reclaim if necessary.
2631 *
2632 * In some special case, if the task is fatal, fatal_signal_pending() or
2633 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2634 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2635 * as possible without any hazards. 2: all pages should have a valid
2636 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2637 * pointer, that is treated as a charge to root_mem_cgroup.
2638 *
2639 * So __mem_cgroup_try_charge() will return
2640 * 0 ... on success, filling *ptr with a valid memcg pointer.
2641 * -ENOMEM ... charge failure because of resource limits.
2642 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2643 *
2644 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2645 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002646 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002647static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002648 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002649 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002650 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002651 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002652{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002653 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002654 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002655 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002656 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002657
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002658 /*
2659 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2660 * in system level. So, allow to go ahead dying process in addition to
2661 * MEMDIE process.
2662 */
2663 if (unlikely(test_thread_flag(TIF_MEMDIE)
2664 || fatal_signal_pending(current)))
2665 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002666
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002667 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002668 * We always charge the cgroup the mm_struct belongs to.
2669 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002670 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002671 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002672 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002673 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002674 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002675again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002676 if (*ptr) { /* css should be a valid one */
2677 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002678 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002679 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002680 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002681 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002682 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002683 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002684 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002685
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002686 rcu_read_lock();
2687 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002688 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002689 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002690 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002691 * race with swapoff. Then, we have small risk of mis-accouning.
2692 * But such kind of mis-account by race always happens because
2693 * we don't have cgroup_mutex(). It's overkill and we allo that
2694 * small race, here.
2695 * (*) swapoff at el will charge against mm-struct not against
2696 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002697 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002698 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002699 if (!memcg)
2700 memcg = root_mem_cgroup;
2701 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002702 rcu_read_unlock();
2703 goto done;
2704 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002705 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002706 /*
2707 * It seems dagerous to access memcg without css_get().
2708 * But considering how consume_stok works, it's not
2709 * necessary. If consume_stock success, some charges
2710 * from this memcg are cached on this cpu. So, we
2711 * don't need to call css_get()/css_tryget() before
2712 * calling consume_stock().
2713 */
2714 rcu_read_unlock();
2715 goto done;
2716 }
2717 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002718 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002719 rcu_read_unlock();
2720 goto again;
2721 }
2722 rcu_read_unlock();
2723 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002724
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002725 do {
2726 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002727
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002728 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002729 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002730 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002731 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002732 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002733
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002734 oom_check = false;
2735 if (oom && !nr_oom_retries) {
2736 oom_check = true;
2737 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2738 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002739
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002740 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages,
2741 oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002742 switch (ret) {
2743 case CHARGE_OK:
2744 break;
2745 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002746 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002747 css_put(&memcg->css);
2748 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002749 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002750 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002751 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002752 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002753 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002754 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002755 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002756 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002757 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002758 /* If oom, we never return -ENOMEM */
2759 nr_oom_retries--;
2760 break;
2761 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002762 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002763 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002764 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002765 } while (ret != CHARGE_OK);
2766
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002767 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002768 refill_stock(memcg, batch - nr_pages);
2769 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002770done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002771 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002772 return 0;
2773nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002774 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002775 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002776bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002777 *ptr = root_mem_cgroup;
2778 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002779}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002780
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002781/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002782 * Somemtimes we have to undo a charge we got by try_charge().
2783 * This function is for that and do uncharge, put css's refcnt.
2784 * gotten by try_charge().
2785 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002786static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002787 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002788{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002789 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002790 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002791
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002792 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002793 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002794 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002795 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002796}
2797
2798/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002799 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2800 * This is useful when moving usage to parent cgroup.
2801 */
2802static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2803 unsigned int nr_pages)
2804{
2805 unsigned long bytes = nr_pages * PAGE_SIZE;
2806
2807 if (mem_cgroup_is_root(memcg))
2808 return;
2809
2810 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2811 if (do_swap_account)
2812 res_counter_uncharge_until(&memcg->memsw,
2813 memcg->memsw.parent, bytes);
2814}
2815
2816/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002817 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002818 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2819 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2820 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002821 */
2822static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2823{
2824 struct cgroup_subsys_state *css;
2825
2826 /* ID 0 is unused ID */
2827 if (!id)
2828 return NULL;
2829 css = css_lookup(&mem_cgroup_subsys, id);
2830 if (!css)
2831 return NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -07002832 return mem_cgroup_from_css(css);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002833}
2834
Wu Fengguange42d9d52009-12-16 12:19:59 +01002835struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002836{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002837 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002838 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002839 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002840 swp_entry_t ent;
2841
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002842 VM_BUG_ON(!PageLocked(page));
2843
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002844 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002845 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002846 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002847 memcg = pc->mem_cgroup;
2848 if (memcg && !css_tryget(&memcg->css))
2849 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002850 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002851 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002852 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002853 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002854 memcg = mem_cgroup_lookup(id);
2855 if (memcg && !css_tryget(&memcg->css))
2856 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002857 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002858 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002859 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002860 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002861}
2862
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002863static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002864 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002865 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002866 enum charge_type ctype,
2867 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002868{
Johannes Weinerce587e62012-04-24 20:22:33 +02002869 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002870 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002871 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002872 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002873 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002874
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002875 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002876 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002877 /*
2878 * we don't need page_cgroup_lock about tail pages, becase they are not
2879 * accessed by any other context at this point.
2880 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002881
2882 /*
2883 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2884 * may already be on some other mem_cgroup's LRU. Take care of it.
2885 */
2886 if (lrucare) {
2887 zone = page_zone(page);
2888 spin_lock_irq(&zone->lru_lock);
2889 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002890 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002891 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002892 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002893 was_on_lru = true;
2894 }
2895 }
2896
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002897 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002898 /*
2899 * We access a page_cgroup asynchronously without lock_page_cgroup().
2900 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2901 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2902 * before USED bit, we need memory barrier here.
2903 * See mem_cgroup_add_lru_list(), etc.
2904 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002905 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002906 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002907
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002908 if (lrucare) {
2909 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002910 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002911 VM_BUG_ON(PageLRU(page));
2912 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002913 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002914 }
2915 spin_unlock_irq(&zone->lru_lock);
2916 }
2917
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002918 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002919 anon = true;
2920 else
2921 anon = false;
2922
David Rientjesb070e652013-05-07 16:18:09 -07002923 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002924 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002925
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002926 /*
2927 * "charge_statistics" updated event counter. Then, check it.
2928 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2929 * if they exceeds softlimit.
2930 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002931 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002932}
2933
Glauber Costa7cf27982012-12-18 14:22:55 -08002934static DEFINE_MUTEX(set_limit_mutex);
2935
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002936#ifdef CONFIG_MEMCG_KMEM
2937static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2938{
2939 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
2940 (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK);
2941}
2942
Glauber Costa1f458cb2012-12-18 14:22:50 -08002943/*
2944 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2945 * in the memcg_cache_params struct.
2946 */
2947static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2948{
2949 struct kmem_cache *cachep;
2950
2951 VM_BUG_ON(p->is_root_cache);
2952 cachep = p->root_cache;
2953 return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)];
2954}
2955
Glauber Costa749c5412012-12-18 14:23:01 -08002956#ifdef CONFIG_SLABINFO
2957static int mem_cgroup_slabinfo_read(struct cgroup *cont, struct cftype *cft,
2958 struct seq_file *m)
2959{
2960 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
2961 struct memcg_cache_params *params;
2962
2963 if (!memcg_can_account_kmem(memcg))
2964 return -EIO;
2965
2966 print_slabinfo_header(m);
2967
2968 mutex_lock(&memcg->slab_caches_mutex);
2969 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2970 cache_show(memcg_params_to_cache(params), m);
2971 mutex_unlock(&memcg->slab_caches_mutex);
2972
2973 return 0;
2974}
2975#endif
2976
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002977static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2978{
2979 struct res_counter *fail_res;
2980 struct mem_cgroup *_memcg;
2981 int ret = 0;
2982 bool may_oom;
2983
2984 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2985 if (ret)
2986 return ret;
2987
2988 /*
2989 * Conditions under which we can wait for the oom_killer. Those are
2990 * the same conditions tested by the core page allocator
2991 */
2992 may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY);
2993
2994 _memcg = memcg;
2995 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
2996 &_memcg, may_oom);
2997
2998 if (ret == -EINTR) {
2999 /*
3000 * __mem_cgroup_try_charge() chosed to bypass to root due to
3001 * OOM kill or fatal signal. Since our only options are to
3002 * either fail the allocation or charge it to this cgroup, do
3003 * it as a temporary condition. But we can't fail. From a
3004 * kmem/slab perspective, the cache has already been selected,
3005 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3006 * our minds.
3007 *
3008 * This condition will only trigger if the task entered
3009 * memcg_charge_kmem in a sane state, but was OOM-killed during
3010 * __mem_cgroup_try_charge() above. Tasks that were already
3011 * dying when the allocation triggers should have been already
3012 * directed to the root cgroup in memcontrol.h
3013 */
3014 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3015 if (do_swap_account)
3016 res_counter_charge_nofail(&memcg->memsw, size,
3017 &fail_res);
3018 ret = 0;
3019 } else if (ret)
3020 res_counter_uncharge(&memcg->kmem, size);
3021
3022 return ret;
3023}
3024
3025static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3026{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003027 res_counter_uncharge(&memcg->res, size);
3028 if (do_swap_account)
3029 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003030
3031 /* Not down to 0 */
3032 if (res_counter_uncharge(&memcg->kmem, size))
3033 return;
3034
Li Zefan10d5ebf2013-07-08 16:00:33 -07003035 /*
3036 * Releases a reference taken in kmem_cgroup_css_offline in case
3037 * this last uncharge is racing with the offlining code or it is
3038 * outliving the memcg existence.
3039 *
3040 * The memory barrier imposed by test&clear is paired with the
3041 * explicit one in memcg_kmem_mark_dead().
3042 */
Glauber Costa7de37682012-12-18 14:22:07 -08003043 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003044 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003045}
3046
Glauber Costa2633d7a2012-12-18 14:22:34 -08003047void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
3048{
3049 if (!memcg)
3050 return;
3051
3052 mutex_lock(&memcg->slab_caches_mutex);
3053 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
3054 mutex_unlock(&memcg->slab_caches_mutex);
3055}
3056
3057/*
3058 * helper for acessing a memcg's index. It will be used as an index in the
3059 * child cache array in kmem_cache, and also to derive its name. This function
3060 * will return -1 when this is not a kmem-limited memcg.
3061 */
3062int memcg_cache_id(struct mem_cgroup *memcg)
3063{
3064 return memcg ? memcg->kmemcg_id : -1;
3065}
3066
Glauber Costa55007d82012-12-18 14:22:38 -08003067/*
3068 * This ends up being protected by the set_limit mutex, during normal
3069 * operation, because that is its main call site.
3070 *
3071 * But when we create a new cache, we can call this as well if its parent
3072 * is kmem-limited. That will have to hold set_limit_mutex as well.
3073 */
3074int memcg_update_cache_sizes(struct mem_cgroup *memcg)
3075{
3076 int num, ret;
3077
3078 num = ida_simple_get(&kmem_limited_groups,
3079 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
3080 if (num < 0)
3081 return num;
3082 /*
3083 * After this point, kmem_accounted (that we test atomically in
3084 * the beginning of this conditional), is no longer 0. This
3085 * guarantees only one process will set the following boolean
3086 * to true. We don't need test_and_set because we're protected
3087 * by the set_limit_mutex anyway.
3088 */
3089 memcg_kmem_set_activated(memcg);
3090
3091 ret = memcg_update_all_caches(num+1);
3092 if (ret) {
3093 ida_simple_remove(&kmem_limited_groups, num);
3094 memcg_kmem_clear_activated(memcg);
3095 return ret;
3096 }
3097
3098 memcg->kmemcg_id = num;
3099 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
3100 mutex_init(&memcg->slab_caches_mutex);
3101 return 0;
3102}
3103
3104static size_t memcg_caches_array_size(int num_groups)
3105{
3106 ssize_t size;
3107 if (num_groups <= 0)
3108 return 0;
3109
3110 size = 2 * num_groups;
3111 if (size < MEMCG_CACHES_MIN_SIZE)
3112 size = MEMCG_CACHES_MIN_SIZE;
3113 else if (size > MEMCG_CACHES_MAX_SIZE)
3114 size = MEMCG_CACHES_MAX_SIZE;
3115
3116 return size;
3117}
3118
3119/*
3120 * We should update the current array size iff all caches updates succeed. This
3121 * can only be done from the slab side. The slab mutex needs to be held when
3122 * calling this.
3123 */
3124void memcg_update_array_size(int num)
3125{
3126 if (num > memcg_limited_groups_array_size)
3127 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3128}
3129
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003130static void kmem_cache_destroy_work_func(struct work_struct *w);
3131
Glauber Costa55007d82012-12-18 14:22:38 -08003132int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3133{
3134 struct memcg_cache_params *cur_params = s->memcg_params;
3135
3136 VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache);
3137
3138 if (num_groups > memcg_limited_groups_array_size) {
3139 int i;
3140 ssize_t size = memcg_caches_array_size(num_groups);
3141
3142 size *= sizeof(void *);
3143 size += sizeof(struct memcg_cache_params);
3144
3145 s->memcg_params = kzalloc(size, GFP_KERNEL);
3146 if (!s->memcg_params) {
3147 s->memcg_params = cur_params;
3148 return -ENOMEM;
3149 }
3150
3151 s->memcg_params->is_root_cache = true;
3152
3153 /*
3154 * There is the chance it will be bigger than
3155 * memcg_limited_groups_array_size, if we failed an allocation
3156 * in a cache, in which case all caches updated before it, will
3157 * have a bigger array.
3158 *
3159 * But if that is the case, the data after
3160 * memcg_limited_groups_array_size is certainly unused
3161 */
3162 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3163 if (!cur_params->memcg_caches[i])
3164 continue;
3165 s->memcg_params->memcg_caches[i] =
3166 cur_params->memcg_caches[i];
3167 }
3168
3169 /*
3170 * Ideally, we would wait until all caches succeed, and only
3171 * then free the old one. But this is not worth the extra
3172 * pointer per-cache we'd have to have for this.
3173 *
3174 * It is not a big deal if some caches are left with a size
3175 * bigger than the others. And all updates will reset this
3176 * anyway.
3177 */
3178 kfree(cur_params);
3179 }
3180 return 0;
3181}
3182
Glauber Costa943a4512012-12-18 14:23:03 -08003183int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
3184 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003185{
3186 size_t size = sizeof(struct memcg_cache_params);
3187
3188 if (!memcg_kmem_enabled())
3189 return 0;
3190
Glauber Costa55007d82012-12-18 14:22:38 -08003191 if (!memcg)
3192 size += memcg_limited_groups_array_size * sizeof(void *);
3193
Glauber Costa2633d7a2012-12-18 14:22:34 -08003194 s->memcg_params = kzalloc(size, GFP_KERNEL);
3195 if (!s->memcg_params)
3196 return -ENOMEM;
3197
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003198 INIT_WORK(&s->memcg_params->destroy,
3199 kmem_cache_destroy_work_func);
Glauber Costa943a4512012-12-18 14:23:03 -08003200 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003201 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003202 s->memcg_params->root_cache = root_cache;
Glauber Costa4ba902b2013-02-12 13:46:22 -08003203 } else
3204 s->memcg_params->is_root_cache = true;
3205
Glauber Costa2633d7a2012-12-18 14:22:34 -08003206 return 0;
3207}
3208
3209void memcg_release_cache(struct kmem_cache *s)
3210{
Glauber Costad7f25f82012-12-18 14:22:40 -08003211 struct kmem_cache *root;
3212 struct mem_cgroup *memcg;
3213 int id;
3214
3215 /*
3216 * This happens, for instance, when a root cache goes away before we
3217 * add any memcg.
3218 */
3219 if (!s->memcg_params)
3220 return;
3221
3222 if (s->memcg_params->is_root_cache)
3223 goto out;
3224
3225 memcg = s->memcg_params->memcg;
3226 id = memcg_cache_id(memcg);
3227
3228 root = s->memcg_params->root_cache;
3229 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08003230
3231 mutex_lock(&memcg->slab_caches_mutex);
3232 list_del(&s->memcg_params->list);
3233 mutex_unlock(&memcg->slab_caches_mutex);
3234
Li Zefan20f05312013-07-08 16:00:31 -07003235 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003236out:
Glauber Costa2633d7a2012-12-18 14:22:34 -08003237 kfree(s->memcg_params);
3238}
3239
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003240/*
3241 * During the creation a new cache, we need to disable our accounting mechanism
3242 * altogether. This is true even if we are not creating, but rather just
3243 * enqueing new caches to be created.
3244 *
3245 * This is because that process will trigger allocations; some visible, like
3246 * explicit kmallocs to auxiliary data structures, name strings and internal
3247 * cache structures; some well concealed, like INIT_WORK() that can allocate
3248 * objects during debug.
3249 *
3250 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3251 * to it. This may not be a bounded recursion: since the first cache creation
3252 * failed to complete (waiting on the allocation), we'll just try to create the
3253 * cache again, failing at the same point.
3254 *
3255 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3256 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3257 * inside the following two functions.
3258 */
3259static inline void memcg_stop_kmem_account(void)
3260{
3261 VM_BUG_ON(!current->mm);
3262 current->memcg_kmem_skip_account++;
3263}
3264
3265static inline void memcg_resume_kmem_account(void)
3266{
3267 VM_BUG_ON(!current->mm);
3268 current->memcg_kmem_skip_account--;
3269}
3270
Glauber Costa1f458cb2012-12-18 14:22:50 -08003271static void kmem_cache_destroy_work_func(struct work_struct *w)
3272{
3273 struct kmem_cache *cachep;
3274 struct memcg_cache_params *p;
3275
3276 p = container_of(w, struct memcg_cache_params, destroy);
3277
3278 cachep = memcg_params_to_cache(p);
3279
Glauber Costa22933152012-12-18 14:22:59 -08003280 /*
3281 * If we get down to 0 after shrink, we could delete right away.
3282 * However, memcg_release_pages() already puts us back in the workqueue
3283 * in that case. If we proceed deleting, we'll get a dangling
3284 * reference, and removing the object from the workqueue in that case
3285 * is unnecessary complication. We are not a fast path.
3286 *
3287 * Note that this case is fundamentally different from racing with
3288 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3289 * kmem_cache_shrink, not only we would be reinserting a dead cache
3290 * into the queue, but doing so from inside the worker racing to
3291 * destroy it.
3292 *
3293 * So if we aren't down to zero, we'll just schedule a worker and try
3294 * again
3295 */
3296 if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
3297 kmem_cache_shrink(cachep);
3298 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
3299 return;
3300 } else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003301 kmem_cache_destroy(cachep);
3302}
3303
3304void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3305{
3306 if (!cachep->memcg_params->dead)
3307 return;
3308
3309 /*
Glauber Costa22933152012-12-18 14:22:59 -08003310 * There are many ways in which we can get here.
3311 *
3312 * We can get to a memory-pressure situation while the delayed work is
3313 * still pending to run. The vmscan shrinkers can then release all
3314 * cache memory and get us to destruction. If this is the case, we'll
3315 * be executed twice, which is a bug (the second time will execute over
3316 * bogus data). In this case, cancelling the work should be fine.
3317 *
3318 * But we can also get here from the worker itself, if
3319 * kmem_cache_shrink is enough to shake all the remaining objects and
3320 * get the page count to 0. In this case, we'll deadlock if we try to
3321 * cancel the work (the worker runs with an internal lock held, which
3322 * is the same lock we would hold for cancel_work_sync().)
3323 *
3324 * Since we can't possibly know who got us here, just refrain from
3325 * running if there is already work pending
3326 */
3327 if (work_pending(&cachep->memcg_params->destroy))
3328 return;
3329 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003330 * We have to defer the actual destroying to a workqueue, because
3331 * we might currently be in a context that cannot sleep.
3332 */
3333 schedule_work(&cachep->memcg_params->destroy);
3334}
3335
Glauber Costad7f25f82012-12-18 14:22:40 -08003336/*
3337 * This lock protects updaters, not readers. We want readers to be as fast as
3338 * they can, and they will either see NULL or a valid cache value. Our model
3339 * allow them to see NULL, in which case the root memcg will be selected.
3340 *
3341 * We need this lock because multiple allocations to the same cache from a non
3342 * will span more than one worker. Only one of them can create the cache.
3343 */
3344static DEFINE_MUTEX(memcg_cache_mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003345
3346/*
3347 * Called with memcg_cache_mutex held
3348 */
3349static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
3350 struct kmem_cache *s)
3351{
3352 struct kmem_cache *new;
3353 static char *tmp_name = NULL;
3354
3355 lockdep_assert_held(&memcg_cache_mutex);
3356
3357 /*
3358 * kmem_cache_create_memcg duplicates the given name and
3359 * cgroup_name for this name requires RCU context.
3360 * This static temporary buffer is used to prevent from
3361 * pointless shortliving allocation.
3362 */
3363 if (!tmp_name) {
3364 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3365 if (!tmp_name)
3366 return NULL;
3367 }
3368
3369 rcu_read_lock();
3370 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3371 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3372 rcu_read_unlock();
3373
3374 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3375 (s->flags & ~SLAB_PANIC), s->ctor, s);
3376
3377 if (new)
3378 new->allocflags |= __GFP_KMEMCG;
3379
3380 return new;
3381}
3382
Glauber Costad7f25f82012-12-18 14:22:40 -08003383static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3384 struct kmem_cache *cachep)
3385{
3386 struct kmem_cache *new_cachep;
3387 int idx;
3388
3389 BUG_ON(!memcg_can_account_kmem(memcg));
3390
3391 idx = memcg_cache_id(memcg);
3392
3393 mutex_lock(&memcg_cache_mutex);
3394 new_cachep = cachep->memcg_params->memcg_caches[idx];
Li Zefan20f05312013-07-08 16:00:31 -07003395 if (new_cachep) {
3396 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003397 goto out;
Li Zefan20f05312013-07-08 16:00:31 -07003398 }
Glauber Costad7f25f82012-12-18 14:22:40 -08003399
3400 new_cachep = kmem_cache_dup(memcg, cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003401 if (new_cachep == NULL) {
3402 new_cachep = cachep;
Li Zefan20f05312013-07-08 16:00:31 -07003403 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003404 goto out;
3405 }
3406
Glauber Costa1f458cb2012-12-18 14:22:50 -08003407 atomic_set(&new_cachep->memcg_params->nr_pages , 0);
Glauber Costad7f25f82012-12-18 14:22:40 -08003408
3409 cachep->memcg_params->memcg_caches[idx] = new_cachep;
3410 /*
3411 * the readers won't lock, make sure everybody sees the updated value,
3412 * so they won't put stuff in the queue again for no reason
3413 */
3414 wmb();
3415out:
3416 mutex_unlock(&memcg_cache_mutex);
3417 return new_cachep;
3418}
3419
Glauber Costa7cf27982012-12-18 14:22:55 -08003420void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3421{
3422 struct kmem_cache *c;
3423 int i;
3424
3425 if (!s->memcg_params)
3426 return;
3427 if (!s->memcg_params->is_root_cache)
3428 return;
3429
3430 /*
3431 * If the cache is being destroyed, we trust that there is no one else
3432 * requesting objects from it. Even if there are, the sanity checks in
3433 * kmem_cache_destroy should caught this ill-case.
3434 *
3435 * Still, we don't want anyone else freeing memcg_caches under our
3436 * noses, which can happen if a new memcg comes to life. As usual,
3437 * we'll take the set_limit_mutex to protect ourselves against this.
3438 */
3439 mutex_lock(&set_limit_mutex);
3440 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3441 c = s->memcg_params->memcg_caches[i];
3442 if (!c)
3443 continue;
3444
3445 /*
3446 * We will now manually delete the caches, so to avoid races
3447 * we need to cancel all pending destruction workers and
3448 * proceed with destruction ourselves.
3449 *
3450 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3451 * and that could spawn the workers again: it is likely that
3452 * the cache still have active pages until this very moment.
3453 * This would lead us back to mem_cgroup_destroy_cache.
3454 *
3455 * But that will not execute at all if the "dead" flag is not
3456 * set, so flip it down to guarantee we are in control.
3457 */
3458 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003459 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003460 kmem_cache_destroy(c);
3461 }
3462 mutex_unlock(&set_limit_mutex);
3463}
3464
Glauber Costad7f25f82012-12-18 14:22:40 -08003465struct create_work {
3466 struct mem_cgroup *memcg;
3467 struct kmem_cache *cachep;
3468 struct work_struct work;
3469};
3470
Glauber Costa1f458cb2012-12-18 14:22:50 -08003471static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3472{
3473 struct kmem_cache *cachep;
3474 struct memcg_cache_params *params;
3475
3476 if (!memcg_kmem_is_active(memcg))
3477 return;
3478
3479 mutex_lock(&memcg->slab_caches_mutex);
3480 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3481 cachep = memcg_params_to_cache(params);
3482 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003483 schedule_work(&cachep->memcg_params->destroy);
3484 }
3485 mutex_unlock(&memcg->slab_caches_mutex);
3486}
3487
Glauber Costad7f25f82012-12-18 14:22:40 -08003488static void memcg_create_cache_work_func(struct work_struct *w)
3489{
3490 struct create_work *cw;
3491
3492 cw = container_of(w, struct create_work, work);
3493 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003494 kfree(cw);
3495}
3496
3497/*
3498 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003499 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003500static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3501 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003502{
3503 struct create_work *cw;
3504
3505 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003506 if (cw == NULL) {
3507 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003508 return;
3509 }
3510
3511 cw->memcg = memcg;
3512 cw->cachep = cachep;
3513
3514 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3515 schedule_work(&cw->work);
3516}
3517
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003518static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3519 struct kmem_cache *cachep)
3520{
3521 /*
3522 * We need to stop accounting when we kmalloc, because if the
3523 * corresponding kmalloc cache is not yet created, the first allocation
3524 * in __memcg_create_cache_enqueue will recurse.
3525 *
3526 * However, it is better to enclose the whole function. Depending on
3527 * the debugging options enabled, INIT_WORK(), for instance, can
3528 * trigger an allocation. This too, will make us recurse. Because at
3529 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3530 * the safest choice is to do it like this, wrapping the whole function.
3531 */
3532 memcg_stop_kmem_account();
3533 __memcg_create_cache_enqueue(memcg, cachep);
3534 memcg_resume_kmem_account();
3535}
Glauber Costad7f25f82012-12-18 14:22:40 -08003536/*
3537 * Return the kmem_cache we're supposed to use for a slab allocation.
3538 * We try to use the current memcg's version of the cache.
3539 *
3540 * If the cache does not exist yet, if we are the first user of it,
3541 * we either create it immediately, if possible, or create it asynchronously
3542 * in a workqueue.
3543 * In the latter case, we will let the current allocation go through with
3544 * the original cache.
3545 *
3546 * Can't be called in interrupt context or from kernel threads.
3547 * This function needs to be called with rcu_read_lock() held.
3548 */
3549struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3550 gfp_t gfp)
3551{
3552 struct mem_cgroup *memcg;
3553 int idx;
3554
3555 VM_BUG_ON(!cachep->memcg_params);
3556 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3557
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003558 if (!current->mm || current->memcg_kmem_skip_account)
3559 return cachep;
3560
Glauber Costad7f25f82012-12-18 14:22:40 -08003561 rcu_read_lock();
3562 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003563
3564 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003565 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003566
3567 idx = memcg_cache_id(memcg);
3568
3569 /*
3570 * barrier to mare sure we're always seeing the up to date value. The
3571 * code updating memcg_caches will issue a write barrier to match this.
3572 */
3573 read_barrier_depends();
Li Zefanca0dde92013-04-29 15:08:57 -07003574 if (likely(cachep->memcg_params->memcg_caches[idx])) {
3575 cachep = cachep->memcg_params->memcg_caches[idx];
3576 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003577 }
3578
Li Zefanca0dde92013-04-29 15:08:57 -07003579 /* The corresponding put will be done in the workqueue. */
3580 if (!css_tryget(&memcg->css))
3581 goto out;
3582 rcu_read_unlock();
3583
3584 /*
3585 * If we are in a safe context (can wait, and not in interrupt
3586 * context), we could be be predictable and return right away.
3587 * This would guarantee that the allocation being performed
3588 * already belongs in the new cache.
3589 *
3590 * However, there are some clashes that can arrive from locking.
3591 * For instance, because we acquire the slab_mutex while doing
3592 * kmem_cache_dup, this means no further allocation could happen
3593 * with the slab_mutex held.
3594 *
3595 * Also, because cache creation issue get_online_cpus(), this
3596 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3597 * that ends up reversed during cpu hotplug. (cpuset allocates
3598 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3599 * better to defer everything.
3600 */
3601 memcg_create_cache_enqueue(memcg, cachep);
3602 return cachep;
3603out:
3604 rcu_read_unlock();
3605 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003606}
3607EXPORT_SYMBOL(__memcg_kmem_get_cache);
3608
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003609/*
3610 * We need to verify if the allocation against current->mm->owner's memcg is
3611 * possible for the given order. But the page is not allocated yet, so we'll
3612 * need a further commit step to do the final arrangements.
3613 *
3614 * It is possible for the task to switch cgroups in this mean time, so at
3615 * commit time, we can't rely on task conversion any longer. We'll then use
3616 * the handle argument to return to the caller which cgroup we should commit
3617 * against. We could also return the memcg directly and avoid the pointer
3618 * passing, but a boolean return value gives better semantics considering
3619 * the compiled-out case as well.
3620 *
3621 * Returning true means the allocation is possible.
3622 */
3623bool
3624__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3625{
3626 struct mem_cgroup *memcg;
3627 int ret;
3628
3629 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003630
3631 /*
3632 * Disabling accounting is only relevant for some specific memcg
3633 * internal allocations. Therefore we would initially not have such
3634 * check here, since direct calls to the page allocator that are marked
3635 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3636 * concerned with cache allocations, and by having this test at
3637 * memcg_kmem_get_cache, we are already able to relay the allocation to
3638 * the root cache and bypass the memcg cache altogether.
3639 *
3640 * There is one exception, though: the SLUB allocator does not create
3641 * large order caches, but rather service large kmallocs directly from
3642 * the page allocator. Therefore, the following sequence when backed by
3643 * the SLUB allocator:
3644 *
3645 * memcg_stop_kmem_account();
3646 * kmalloc(<large_number>)
3647 * memcg_resume_kmem_account();
3648 *
3649 * would effectively ignore the fact that we should skip accounting,
3650 * since it will drive us directly to this function without passing
3651 * through the cache selector memcg_kmem_get_cache. Such large
3652 * allocations are extremely rare but can happen, for instance, for the
3653 * cache arrays. We bring this test here.
3654 */
3655 if (!current->mm || current->memcg_kmem_skip_account)
3656 return true;
3657
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003658 memcg = try_get_mem_cgroup_from_mm(current->mm);
3659
3660 /*
3661 * very rare case described in mem_cgroup_from_task. Unfortunately there
3662 * isn't much we can do without complicating this too much, and it would
3663 * be gfp-dependent anyway. Just let it go
3664 */
3665 if (unlikely(!memcg))
3666 return true;
3667
3668 if (!memcg_can_account_kmem(memcg)) {
3669 css_put(&memcg->css);
3670 return true;
3671 }
3672
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003673 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3674 if (!ret)
3675 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003676
3677 css_put(&memcg->css);
3678 return (ret == 0);
3679}
3680
3681void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3682 int order)
3683{
3684 struct page_cgroup *pc;
3685
3686 VM_BUG_ON(mem_cgroup_is_root(memcg));
3687
3688 /* The page allocation failed. Revert */
3689 if (!page) {
3690 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003691 return;
3692 }
3693
3694 pc = lookup_page_cgroup(page);
3695 lock_page_cgroup(pc);
3696 pc->mem_cgroup = memcg;
3697 SetPageCgroupUsed(pc);
3698 unlock_page_cgroup(pc);
3699}
3700
3701void __memcg_kmem_uncharge_pages(struct page *page, int order)
3702{
3703 struct mem_cgroup *memcg = NULL;
3704 struct page_cgroup *pc;
3705
3706
3707 pc = lookup_page_cgroup(page);
3708 /*
3709 * Fast unlocked return. Theoretically might have changed, have to
3710 * check again after locking.
3711 */
3712 if (!PageCgroupUsed(pc))
3713 return;
3714
3715 lock_page_cgroup(pc);
3716 if (PageCgroupUsed(pc)) {
3717 memcg = pc->mem_cgroup;
3718 ClearPageCgroupUsed(pc);
3719 }
3720 unlock_page_cgroup(pc);
3721
3722 /*
3723 * We trust that only if there is a memcg associated with the page, it
3724 * is a valid allocation
3725 */
3726 if (!memcg)
3727 return;
3728
3729 VM_BUG_ON(mem_cgroup_is_root(memcg));
3730 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003731}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003732#else
3733static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3734{
3735}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003736#endif /* CONFIG_MEMCG_KMEM */
3737
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003738#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3739
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003740#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003741/*
3742 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003743 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3744 * charge/uncharge will be never happen and move_account() is done under
3745 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003746 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003747void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003748{
3749 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003750 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003751 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003752 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003753
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003754 if (mem_cgroup_disabled())
3755 return;
David Rientjesb070e652013-05-07 16:18:09 -07003756
3757 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003758 for (i = 1; i < HPAGE_PMD_NR; i++) {
3759 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003760 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003761 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003762 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3763 }
David Rientjesb070e652013-05-07 16:18:09 -07003764 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3765 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003766}
Hugh Dickins12d27102012-01-12 17:19:52 -08003767#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003768
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003769/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003770 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003771 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003772 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003773 * @pc: page_cgroup of the page.
3774 * @from: mem_cgroup which the page is moved from.
3775 * @to: mem_cgroup which the page is moved to. @from != @to.
3776 *
3777 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003778 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003779 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003780 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003781 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3782 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003783 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003784static int mem_cgroup_move_account(struct page *page,
3785 unsigned int nr_pages,
3786 struct page_cgroup *pc,
3787 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003788 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003789{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003790 unsigned long flags;
3791 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003792 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003793
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003794 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07003795 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07003796 /*
3797 * The page is isolated from LRU. So, collapse function
3798 * will not handle this page. But page splitting can happen.
3799 * Do this check under compound_page_lock(). The caller should
3800 * hold it.
3801 */
3802 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003803 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003804 goto out;
3805
3806 lock_page_cgroup(pc);
3807
3808 ret = -EINVAL;
3809 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3810 goto unlock;
3811
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003812 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003813
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07003814 if (!anon && page_mapped(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003815 /* Update mapped_file data for mem_cgroup */
3816 preempt_disable();
3817 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3818 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
3819 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07003820 }
David Rientjesb070e652013-05-07 16:18:09 -07003821 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003822
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003823 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003824 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003825 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003826 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003827 ret = 0;
3828unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003829 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003830 /*
3831 * check events
3832 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003833 memcg_check_events(to, page);
3834 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003835out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003836 return ret;
3837}
3838
Michal Hocko2ef37d32012-10-26 13:37:30 +02003839/**
3840 * mem_cgroup_move_parent - moves page to the parent group
3841 * @page: the page to move
3842 * @pc: page_cgroup of the page
3843 * @child: page's cgroup
3844 *
3845 * move charges to its parent or the root cgroup if the group has no
3846 * parent (aka use_hierarchy==0).
3847 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3848 * mem_cgroup_move_account fails) the failure is always temporary and
3849 * it signals a race with a page removal/uncharge or migration. In the
3850 * first case the page is on the way out and it will vanish from the LRU
3851 * on the next attempt and the call should be retried later.
3852 * Isolation from the LRU fails only if page has been isolated from
3853 * the LRU since we looked at it and that usually means either global
3854 * reclaim or migration going on. The page will either get back to the
3855 * LRU or vanish.
3856 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3857 * (!PageCgroupUsed) or moved to a different group. The page will
3858 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003859 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003860static int mem_cgroup_move_parent(struct page *page,
3861 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003862 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003863{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003864 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003865 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003866 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003867 int ret;
3868
Michal Hockod8423012012-10-26 13:37:29 +02003869 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003870
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003871 ret = -EBUSY;
3872 if (!get_page_unless_zero(page))
3873 goto out;
3874 if (isolate_lru_page(page))
3875 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003876
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003877 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003878
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003879 parent = parent_mem_cgroup(child);
3880 /*
3881 * If no parent, move charges to root cgroup.
3882 */
3883 if (!parent)
3884 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003885
Michal Hocko2ef37d32012-10-26 13:37:30 +02003886 if (nr_pages > 1) {
3887 VM_BUG_ON(!PageTransHuge(page));
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003888 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003889 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003890
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003891 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003892 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003893 if (!ret)
3894 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003895
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003896 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003897 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003898 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003899put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003900 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003901out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003902 return ret;
3903}
3904
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003905/*
3906 * Charge the memory controller for page usage.
3907 * Return
3908 * 0 if the charge was successful
3909 * < 0 if the cgroup is over its limit
3910 */
3911static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003912 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003913{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003914 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003915 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003916 bool oom = true;
3917 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003918
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003919 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003920 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003921 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08003922 /*
3923 * Never OOM-kill a process for a huge page. The
3924 * fault handler will fall back to regular pages.
3925 */
3926 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003927 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003928
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003929 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003930 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003931 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003932 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003933 return 0;
3934}
3935
3936int mem_cgroup_newpage_charge(struct page *page,
3937 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003938{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003939 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003940 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003941 VM_BUG_ON(page_mapped(page));
3942 VM_BUG_ON(page->mapping && !PageAnon(page));
3943 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003944 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003945 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003946}
3947
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003948/*
3949 * While swap-in, try_charge -> commit or cancel, the page is locked.
3950 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003951 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003952 * "commit()" or removed by "cancel()"
3953 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003954static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3955 struct page *page,
3956 gfp_t mask,
3957 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003958{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003959 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003960 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003961 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003962
Johannes Weiner90deb782012-07-31 16:45:47 -07003963 pc = lookup_page_cgroup(page);
3964 /*
3965 * Every swap fault against a single page tries to charge the
3966 * page, bail as early as possible. shmem_unuse() encounters
3967 * already charged pages, too. The USED bit is protected by
3968 * the page lock, which serializes swap cache removal, which
3969 * in turn serializes uncharging.
3970 */
3971 if (PageCgroupUsed(pc))
3972 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003973 if (!do_swap_account)
3974 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003975 memcg = try_get_mem_cgroup_from_page(page);
3976 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003977 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08003978 *memcgp = memcg;
3979 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003980 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003981 if (ret == -EINTR)
3982 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003983 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003984charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003985 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
3986 if (ret == -EINTR)
3987 ret = 0;
3988 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003989}
3990
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003991int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
3992 gfp_t gfp_mask, struct mem_cgroup **memcgp)
3993{
3994 *memcgp = NULL;
3995 if (mem_cgroup_disabled())
3996 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07003997 /*
3998 * A racing thread's fault, or swapoff, may have already
3999 * updated the pte, and even removed page from swap cache: in
4000 * those cases unuse_pte()'s pte_same() test will fail; but
4001 * there's also a KSM case which does need to charge the page.
4002 */
4003 if (!PageSwapCache(page)) {
4004 int ret;
4005
4006 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
4007 if (ret == -EINTR)
4008 ret = 0;
4009 return ret;
4010 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004011 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
4012}
4013
Johannes Weiner827a03d2012-07-31 16:45:36 -07004014void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
4015{
4016 if (mem_cgroup_disabled())
4017 return;
4018 if (!memcg)
4019 return;
4020 __mem_cgroup_cancel_charge(memcg, 1);
4021}
4022
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004023static void
Johannes Weiner72835c82012-01-12 17:18:32 -08004024__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004025 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004026{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004027 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004028 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08004029 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004030 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07004031
Johannes Weinerce587e62012-04-24 20:22:33 +02004032 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004033 /*
4034 * Now swap is on-memory. This means this page may be
4035 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004036 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
4037 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
4038 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004039 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004040 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004041 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004042 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004043 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004044}
4045
Johannes Weiner72835c82012-01-12 17:18:32 -08004046void mem_cgroup_commit_charge_swapin(struct page *page,
4047 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004048{
Johannes Weiner72835c82012-01-12 17:18:32 -08004049 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004050 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004051}
4052
Johannes Weiner827a03d2012-07-31 16:45:36 -07004053int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4054 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004055{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004056 struct mem_cgroup *memcg = NULL;
4057 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4058 int ret;
4059
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004060 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004061 return 0;
4062 if (PageCompound(page))
4063 return 0;
4064
Johannes Weiner827a03d2012-07-31 16:45:36 -07004065 if (!PageSwapCache(page))
4066 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4067 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004068 ret = __mem_cgroup_try_charge_swapin(mm, page,
4069 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004070 if (!ret)
4071 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4072 }
4073 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004074}
4075
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004076static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004077 unsigned int nr_pages,
4078 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004079{
4080 struct memcg_batch_info *batch = NULL;
4081 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004082
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004083 /* If swapout, usage of swap doesn't decrease */
4084 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4085 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004086
4087 batch = &current->memcg_batch;
4088 /*
4089 * In usual, we do css_get() when we remember memcg pointer.
4090 * But in this case, we keep res->usage until end of a series of
4091 * uncharges. Then, it's ok to ignore memcg's refcnt.
4092 */
4093 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004094 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004095 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004096 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004097 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004098 * the same cgroup and we have chance to coalesce uncharges.
4099 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4100 * because we want to do uncharge as soon as possible.
4101 */
4102
4103 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4104 goto direct_uncharge;
4105
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004106 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004107 goto direct_uncharge;
4108
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004109 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004110 * In typical case, batch->memcg == mem. This means we can
4111 * merge a series of uncharges to an uncharge of res_counter.
4112 * If not, we uncharge res_counter ony by one.
4113 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004114 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004115 goto direct_uncharge;
4116 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004117 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004118 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004119 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004120 return;
4121direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004122 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004123 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004124 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4125 if (unlikely(batch->memcg != memcg))
4126 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004127}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004128
Balbir Singh8697d332008-02-07 00:13:59 -08004129/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004130 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004131 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004132static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004133__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4134 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004135{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004136 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004137 unsigned int nr_pages = 1;
4138 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004139 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004140
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004141 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004142 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004143
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004144 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004145 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004146 VM_BUG_ON(!PageTransHuge(page));
4147 }
Balbir Singh8697d332008-02-07 00:13:59 -08004148 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004149 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004150 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004151 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004152 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004153 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004154
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004155 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004156
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004157 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004158
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004159 if (!PageCgroupUsed(pc))
4160 goto unlock_out;
4161
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004162 anon = PageAnon(page);
4163
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004164 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004165 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004166 /*
4167 * Generally PageAnon tells if it's the anon statistics to be
4168 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4169 * used before page reached the stage of being marked PageAnon.
4170 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004171 anon = true;
4172 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004173 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004174 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004175 if (page_mapped(page))
4176 goto unlock_out;
4177 /*
4178 * Pages under migration may not be uncharged. But
4179 * end_migration() /must/ be the one uncharging the
4180 * unused post-migration page and so it has to call
4181 * here with the migration bit still set. See the
4182 * res_counter handling below.
4183 */
4184 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004185 goto unlock_out;
4186 break;
4187 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4188 if (!PageAnon(page)) { /* Shared memory */
4189 if (page->mapping && !page_is_file_cache(page))
4190 goto unlock_out;
4191 } else if (page_mapped(page)) /* Anon */
4192 goto unlock_out;
4193 break;
4194 default:
4195 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004196 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004197
David Rientjesb070e652013-05-07 16:18:09 -07004198 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004199
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004200 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004201 /*
4202 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4203 * freed from LRU. This is safe because uncharged page is expected not
4204 * to be reused (freed soon). Exception is SwapCache, it's handled by
4205 * special functions.
4206 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004207
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004208 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004209 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004210 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004211 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004212 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004213 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004214 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004215 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004216 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004217 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004218 /*
4219 * Migration does not charge the res_counter for the
4220 * replacement page, so leave it alone when phasing out the
4221 * page that is unused after the migration.
4222 */
4223 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004224 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004225
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004226 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004227
4228unlock_out:
4229 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004230 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004231}
4232
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004233void mem_cgroup_uncharge_page(struct page *page)
4234{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004235 /* early check. */
4236 if (page_mapped(page))
4237 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08004238 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004239 /*
4240 * If the page is in swap cache, uncharge should be deferred
4241 * to the swap path, which also properly accounts swap usage
4242 * and handles memcg lifetime.
4243 *
4244 * Note that this check is not stable and reclaim may add the
4245 * page to swap cache at any time after this. However, if the
4246 * page is not in swap cache by the time page->mapcount hits
4247 * 0, there won't be any page table references to the swap
4248 * slot, and reclaim will free it and not actually write the
4249 * page to disk.
4250 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004251 if (PageSwapCache(page))
4252 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004253 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004254}
4255
4256void mem_cgroup_uncharge_cache_page(struct page *page)
4257{
4258 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07004259 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07004260 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004261}
4262
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004263/*
4264 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4265 * In that cases, pages are freed continuously and we can expect pages
4266 * are in the same memcg. All these calls itself limits the number of
4267 * pages freed at once, then uncharge_start/end() is called properly.
4268 * This may be called prural(2) times in a context,
4269 */
4270
4271void mem_cgroup_uncharge_start(void)
4272{
4273 current->memcg_batch.do_batch++;
4274 /* We can do nest. */
4275 if (current->memcg_batch.do_batch == 1) {
4276 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004277 current->memcg_batch.nr_pages = 0;
4278 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004279 }
4280}
4281
4282void mem_cgroup_uncharge_end(void)
4283{
4284 struct memcg_batch_info *batch = &current->memcg_batch;
4285
4286 if (!batch->do_batch)
4287 return;
4288
4289 batch->do_batch--;
4290 if (batch->do_batch) /* If stacked, do nothing. */
4291 return;
4292
4293 if (!batch->memcg)
4294 return;
4295 /*
4296 * This "batch->memcg" is valid without any css_get/put etc...
4297 * bacause we hide charges behind us.
4298 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004299 if (batch->nr_pages)
4300 res_counter_uncharge(&batch->memcg->res,
4301 batch->nr_pages * PAGE_SIZE);
4302 if (batch->memsw_nr_pages)
4303 res_counter_uncharge(&batch->memcg->memsw,
4304 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004305 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004306 /* forget this pointer (for sanity check) */
4307 batch->memcg = NULL;
4308}
4309
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004310#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004311/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004312 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004313 * memcg information is recorded to swap_cgroup of "ent"
4314 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004315void
4316mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004317{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004318 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004319 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004320
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004321 if (!swapout) /* this was a swap cache but the swap is unused ! */
4322 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4323
Johannes Weiner0030f532012-07-31 16:45:25 -07004324 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004325
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004326 /*
4327 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004328 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004329 */
4330 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004331 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004332}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004333#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004334
Andrew Mortonc255a452012-07-31 16:43:02 -07004335#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004336/*
4337 * called from swap_entry_free(). remove record in swap_cgroup and
4338 * uncharge "memsw" account.
4339 */
4340void mem_cgroup_uncharge_swap(swp_entry_t ent)
4341{
4342 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004343 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004344
4345 if (!do_swap_account)
4346 return;
4347
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004348 id = swap_cgroup_record(ent, 0);
4349 rcu_read_lock();
4350 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004351 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004352 /*
4353 * We uncharge this because swap is freed.
4354 * This memcg can be obsolete one. We avoid calling css_tryget
4355 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004356 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004357 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004358 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004359 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004360 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004361 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004362}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004363
4364/**
4365 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4366 * @entry: swap entry to be moved
4367 * @from: mem_cgroup which the entry is moved from
4368 * @to: mem_cgroup which the entry is moved to
4369 *
4370 * It succeeds only when the swap_cgroup's record for this entry is the same
4371 * as the mem_cgroup's id of @from.
4372 *
4373 * Returns 0 on success, -EINVAL on failure.
4374 *
4375 * The caller must have charged to @to, IOW, called res_counter_charge() about
4376 * both res and memsw, and called css_get().
4377 */
4378static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004379 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004380{
4381 unsigned short old_id, new_id;
4382
4383 old_id = css_id(&from->css);
4384 new_id = css_id(&to->css);
4385
4386 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004387 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004388 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004389 /*
4390 * This function is only called from task migration context now.
4391 * It postpones res_counter and refcount handling till the end
4392 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004393 * improvement. But we cannot postpone css_get(to) because if
4394 * the process that has been moved to @to does swap-in, the
4395 * refcount of @to might be decreased to 0.
4396 *
4397 * We are in attach() phase, so the cgroup is guaranteed to be
4398 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004399 */
Li Zefan40503772013-07-08 16:00:34 -07004400 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004401 return 0;
4402 }
4403 return -EINVAL;
4404}
4405#else
4406static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004407 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004408{
4409 return -EINVAL;
4410}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004411#endif
4412
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004413/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004414 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4415 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004416 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004417void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4418 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004419{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004420 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004421 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004422 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004423 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004424
Johannes Weiner72835c82012-01-12 17:18:32 -08004425 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004426
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004427 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004428 return;
Balbir Singh40779602008-04-04 14:29:59 -07004429
Mel Gormanb32967f2012-11-19 12:35:47 +00004430 if (PageTransHuge(page))
4431 nr_pages <<= compound_order(page);
4432
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004433 pc = lookup_page_cgroup(page);
4434 lock_page_cgroup(pc);
4435 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004436 memcg = pc->mem_cgroup;
4437 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004438 /*
4439 * At migrating an anonymous page, its mapcount goes down
4440 * to 0 and uncharge() will be called. But, even if it's fully
4441 * unmapped, migration may fail and this page has to be
4442 * charged again. We set MIGRATION flag here and delay uncharge
4443 * until end_migration() is called
4444 *
4445 * Corner Case Thinking
4446 * A)
4447 * When the old page was mapped as Anon and it's unmap-and-freed
4448 * while migration was ongoing.
4449 * If unmap finds the old page, uncharge() of it will be delayed
4450 * until end_migration(). If unmap finds a new page, it's
4451 * uncharged when it make mapcount to be 1->0. If unmap code
4452 * finds swap_migration_entry, the new page will not be mapped
4453 * and end_migration() will find it(mapcount==0).
4454 *
4455 * B)
4456 * When the old page was mapped but migraion fails, the kernel
4457 * remaps it. A charge for it is kept by MIGRATION flag even
4458 * if mapcount goes down to 0. We can do remap successfully
4459 * without charging it again.
4460 *
4461 * C)
4462 * The "old" page is under lock_page() until the end of
4463 * migration, so, the old page itself will not be swapped-out.
4464 * If the new page is swapped out before end_migraton, our
4465 * hook to usual swap-out path will catch the event.
4466 */
4467 if (PageAnon(page))
4468 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004469 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004470 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004471 /*
4472 * If the page is not charged at this point,
4473 * we return here.
4474 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004475 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004476 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004477
Johannes Weiner72835c82012-01-12 17:18:32 -08004478 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004479 /*
4480 * We charge new page before it's used/mapped. So, even if unlock_page()
4481 * is called before end_migration, we can catch all events on this new
4482 * page. In the case new page is migrated but not remapped, new page's
4483 * mapcount will be finally 0 and we call uncharge in end_migration().
4484 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004485 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004486 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004487 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004488 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004489 /*
4490 * The page is committed to the memcg, but it's not actually
4491 * charged to the res_counter since we plan on replacing the
4492 * old one and only one page is going to be left afterwards.
4493 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004494 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004495}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004496
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004497/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004498void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004499 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004500{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004501 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004502 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004503 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004504
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004505 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004506 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004507
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004508 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004509 used = oldpage;
4510 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004511 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004512 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004513 unused = oldpage;
4514 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004515 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004516 __mem_cgroup_uncharge_common(unused,
4517 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4518 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4519 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004520 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004521 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004522 * We disallowed uncharge of pages under migration because mapcount
4523 * of the page goes down to zero, temporarly.
4524 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004525 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004526 pc = lookup_page_cgroup(oldpage);
4527 lock_page_cgroup(pc);
4528 ClearPageCgroupMigration(pc);
4529 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004530
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004531 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004532 * If a page is a file cache, radix-tree replacement is very atomic
4533 * and we can skip this check. When it was an Anon page, its mapcount
4534 * goes down to 0. But because we added MIGRATION flage, it's not
4535 * uncharged yet. There are several case but page->mapcount check
4536 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4537 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004538 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004539 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004540 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004541}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004542
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004543/*
4544 * At replace page cache, newpage is not under any memcg but it's on
4545 * LRU. So, this function doesn't touch res_counter but handles LRU
4546 * in correct way. Both pages are locked so we cannot race with uncharge.
4547 */
4548void mem_cgroup_replace_page_cache(struct page *oldpage,
4549 struct page *newpage)
4550{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004551 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004552 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004553 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004554
4555 if (mem_cgroup_disabled())
4556 return;
4557
4558 pc = lookup_page_cgroup(oldpage);
4559 /* fix accounting on old pages */
4560 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004561 if (PageCgroupUsed(pc)) {
4562 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004563 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004564 ClearPageCgroupUsed(pc);
4565 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004566 unlock_page_cgroup(pc);
4567
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004568 /*
4569 * When called from shmem_replace_page(), in some cases the
4570 * oldpage has already been charged, and in some cases not.
4571 */
4572 if (!memcg)
4573 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004574 /*
4575 * Even if newpage->mapping was NULL before starting replacement,
4576 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4577 * LRU while we overwrite pc->mem_cgroup.
4578 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004579 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004580}
4581
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004582#ifdef CONFIG_DEBUG_VM
4583static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4584{
4585 struct page_cgroup *pc;
4586
4587 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004588 /*
4589 * Can be NULL while feeding pages into the page allocator for
4590 * the first time, i.e. during boot or memory hotplug;
4591 * or when mem_cgroup_disabled().
4592 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004593 if (likely(pc) && PageCgroupUsed(pc))
4594 return pc;
4595 return NULL;
4596}
4597
4598bool mem_cgroup_bad_page_check(struct page *page)
4599{
4600 if (mem_cgroup_disabled())
4601 return false;
4602
4603 return lookup_page_cgroup_used(page) != NULL;
4604}
4605
4606void mem_cgroup_print_bad_page(struct page *page)
4607{
4608 struct page_cgroup *pc;
4609
4610 pc = lookup_page_cgroup_used(page);
4611 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004612 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4613 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004614 }
4615}
4616#endif
4617
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004618static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004619 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004620{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004621 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004622 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004623 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004624 int children = mem_cgroup_count_children(memcg);
4625 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004626 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004627
4628 /*
4629 * For keeping hierarchical_reclaim simple, how long we should retry
4630 * is depends on callers. We set our retry-count to be function
4631 * of # of children which we should visit in this loop.
4632 */
4633 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4634
4635 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004636
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004637 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004638 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004639 if (signal_pending(current)) {
4640 ret = -EINTR;
4641 break;
4642 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004643 /*
4644 * Rather than hide all in some function, I do this in
4645 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004646 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004647 */
4648 mutex_lock(&set_limit_mutex);
4649 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4650 if (memswlimit < val) {
4651 ret = -EINVAL;
4652 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004653 break;
4654 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004655
4656 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4657 if (memlimit < val)
4658 enlarge = 1;
4659
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004660 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004661 if (!ret) {
4662 if (memswlimit == val)
4663 memcg->memsw_is_minimum = true;
4664 else
4665 memcg->memsw_is_minimum = false;
4666 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004667 mutex_unlock(&set_limit_mutex);
4668
4669 if (!ret)
4670 break;
4671
Johannes Weiner56600482012-01-12 17:17:59 -08004672 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4673 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004674 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4675 /* Usage is reduced ? */
4676 if (curusage >= oldusage)
4677 retry_count--;
4678 else
4679 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004680 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004681 if (!ret && enlarge)
4682 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004683
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004684 return ret;
4685}
4686
Li Zefan338c8432009-06-17 16:27:15 -07004687static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4688 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004689{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004690 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004691 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004692 int children = mem_cgroup_count_children(memcg);
4693 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004694 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004695
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004696 /* see mem_cgroup_resize_res_limit */
4697 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
4698 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004699 while (retry_count) {
4700 if (signal_pending(current)) {
4701 ret = -EINTR;
4702 break;
4703 }
4704 /*
4705 * Rather than hide all in some function, I do this in
4706 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004707 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004708 */
4709 mutex_lock(&set_limit_mutex);
4710 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4711 if (memlimit > val) {
4712 ret = -EINVAL;
4713 mutex_unlock(&set_limit_mutex);
4714 break;
4715 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004716 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4717 if (memswlimit < val)
4718 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004719 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004720 if (!ret) {
4721 if (memlimit == val)
4722 memcg->memsw_is_minimum = true;
4723 else
4724 memcg->memsw_is_minimum = false;
4725 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004726 mutex_unlock(&set_limit_mutex);
4727
4728 if (!ret)
4729 break;
4730
Johannes Weiner56600482012-01-12 17:17:59 -08004731 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4732 MEM_CGROUP_RECLAIM_NOSWAP |
4733 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004734 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004735 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004736 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004737 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004738 else
4739 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004740 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004741 if (!ret && enlarge)
4742 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004743 return ret;
4744}
4745
Balbir Singh4e416952009-09-23 15:56:39 -07004746unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07004747 gfp_t gfp_mask,
4748 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07004749{
4750 unsigned long nr_reclaimed = 0;
4751 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4752 unsigned long reclaimed;
4753 int loop = 0;
4754 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07004755 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07004756 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07004757
4758 if (order > 0)
4759 return 0;
4760
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07004761 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07004762 /*
4763 * This loop can run a while, specially if mem_cgroup's continuously
4764 * keep exceeding their soft limit and putting the system under
4765 * pressure
4766 */
4767 do {
4768 if (next_mz)
4769 mz = next_mz;
4770 else
4771 mz = mem_cgroup_largest_soft_limit_node(mctz);
4772 if (!mz)
4773 break;
4774
Ying Han0ae5e892011-05-26 16:25:25 -07004775 nr_scanned = 0;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004776 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
Johannes Weiner56600482012-01-12 17:17:59 -08004777 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07004778 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07004779 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07004780 spin_lock(&mctz->lock);
4781
4782 /*
4783 * If we failed to reclaim anything from this memory cgroup
4784 * it is time to move on to the next cgroup
4785 */
4786 next_mz = NULL;
4787 if (!reclaimed) {
4788 do {
4789 /*
4790 * Loop until we find yet another one.
4791 *
4792 * By the time we get the soft_limit lock
4793 * again, someone might have aded the
4794 * group back on the RB tree. Iterate to
4795 * make sure we get a different mem.
4796 * mem_cgroup_largest_soft_limit_node returns
4797 * NULL if no other cgroup is present on
4798 * the tree
4799 */
4800 next_mz =
4801 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07004802 if (next_mz == mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004803 css_put(&next_mz->memcg->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07004804 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07004805 break;
4806 } while (1);
4807 }
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004808 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4809 excess = res_counter_soft_limit_excess(&mz->memcg->res);
Balbir Singh4e416952009-09-23 15:56:39 -07004810 /*
4811 * One school of thought says that we should not add
4812 * back the node to the tree if reclaim returns 0.
4813 * But our reclaim could return 0, simply because due
4814 * to priority we are exposing a smaller subset of
4815 * memory to reclaim from. Consider this as a longer
4816 * term TODO.
4817 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07004818 /* If excess == 0, no tree ops */
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004819 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07004820 spin_unlock(&mctz->lock);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004821 css_put(&mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07004822 loop++;
4823 /*
4824 * Could not reclaim anything and there are no more
4825 * mem cgroups to try or we seem to be looping without
4826 * reclaiming anything.
4827 */
4828 if (!nr_reclaimed &&
4829 (next_mz == NULL ||
4830 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4831 break;
4832 } while (!nr_reclaimed);
4833 if (next_mz)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004834 css_put(&next_mz->memcg->css);
Balbir Singh4e416952009-09-23 15:56:39 -07004835 return nr_reclaimed;
4836}
4837
Michal Hocko2ef37d32012-10-26 13:37:30 +02004838/**
4839 * mem_cgroup_force_empty_list - clears LRU of a group
4840 * @memcg: group to clear
4841 * @node: NUMA node
4842 * @zid: zone id
4843 * @lru: lru to to clear
4844 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004845 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004846 * reclaim the pages page themselves - pages are moved to the parent (or root)
4847 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004848 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004849static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004850 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004851{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004852 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004853 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004854 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004855 struct page *busy;
4856 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004857
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004858 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004859 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4860 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004861
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004862 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004863 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004864 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004865 struct page *page;
4866
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004867 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004868 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004869 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004870 break;
4871 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004872 page = list_entry(list->prev, struct page, lru);
4873 if (busy == page) {
4874 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004875 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004876 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004877 continue;
4878 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004879 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004880
Johannes Weiner925b7672012-01-12 17:18:15 -08004881 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004882
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004883 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004884 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004885 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004886 cond_resched();
4887 } else
4888 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004889 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004890}
4891
4892/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004893 * make mem_cgroup's charge to be 0 if there is no task by moving
4894 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004895 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004896 *
4897 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004898 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004899static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004900{
Michal Hockoc26251f2012-10-26 13:37:28 +02004901 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004902 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004903
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004904 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004905 /* This is for making all *used* pages to be on LRU. */
4906 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004907 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004908 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004909 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004910 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004911 enum lru_list lru;
4912 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004913 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004914 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004915 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004916 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004917 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004918 mem_cgroup_end_move(memcg);
4919 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004920 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004921
Michal Hocko2ef37d32012-10-26 13:37:30 +02004922 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004923 * Kernel memory may not necessarily be trackable to a specific
4924 * process. So they are not migrated, and therefore we can't
4925 * expect their value to drop to 0 here.
4926 * Having res filled up with kmem only is enough.
4927 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004928 * This is a safety check because mem_cgroup_force_empty_list
4929 * could have raced with mem_cgroup_replace_page_cache callers
4930 * so the lru seemed empty but the page could have been added
4931 * right after the check. RES_USAGE should be safe as we always
4932 * charge before adding to the LRU.
4933 */
Glauber Costabea207c2012-12-18 14:22:11 -08004934 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4935 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4936 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004937}
4938
4939/*
Glauber Costab5f99b52013-02-22 16:34:53 -08004940 * This mainly exists for tests during the setting of set of use_hierarchy.
4941 * Since this is the very setting we are changing, the current hierarchy value
4942 * is meaningless
4943 */
4944static inline bool __memcg_has_children(struct mem_cgroup *memcg)
4945{
4946 struct cgroup *pos;
4947
4948 /* bounce at first found */
4949 cgroup_for_each_child(pos, memcg->css.cgroup)
4950 return true;
4951 return false;
4952}
4953
4954/*
Glauber Costa09998212013-02-22 16:34:55 -08004955 * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed
4956 * to be already dead (as in mem_cgroup_force_empty, for instance). This is
Glauber Costab5f99b52013-02-22 16:34:53 -08004957 * from mem_cgroup_count_children(), in the sense that we don't really care how
4958 * many children we have; we only need to know if we have any. It also counts
4959 * any memcg without hierarchy as infertile.
4960 */
4961static inline bool memcg_has_children(struct mem_cgroup *memcg)
4962{
4963 return memcg->use_hierarchy && __memcg_has_children(memcg);
4964}
4965
4966/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004967 * Reclaims as many pages from the given memcg as possible and moves
4968 * the rest to the parent.
4969 *
4970 * Caller is responsible for holding css reference for memcg.
4971 */
4972static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4973{
4974 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4975 struct cgroup *cgrp = memcg->css.cgroup;
4976
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004977 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004978 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4979 return -EBUSY;
4980
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004981 /* we call try-to-free pages for make this cgroup empty */
4982 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004983 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004984 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004985 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004986
Michal Hockoc26251f2012-10-26 13:37:28 +02004987 if (signal_pending(current))
4988 return -EINTR;
4989
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004990 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07004991 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004992 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004993 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004994 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02004995 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004996 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004997
4998 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004999 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02005000 mem_cgroup_reparent_charges(memcg);
5001
5002 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08005003}
5004
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -07005005static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005006{
Michal Hockoc26251f2012-10-26 13:37:28 +02005007 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5008 int ret;
5009
Michal Hockod8423012012-10-26 13:37:29 +02005010 if (mem_cgroup_is_root(memcg))
5011 return -EINVAL;
Michal Hockoc26251f2012-10-26 13:37:28 +02005012 css_get(&memcg->css);
5013 ret = mem_cgroup_force_empty(memcg);
5014 css_put(&memcg->css);
5015
5016 return ret;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005017}
5018
5019
Balbir Singh18f59ea2009-01-07 18:08:07 -08005020static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
5021{
5022 return mem_cgroup_from_cont(cont)->use_hierarchy;
5023}
5024
5025static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
5026 u64 val)
5027{
5028 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005029 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005030 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005031 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005032
5033 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005034 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005035
Glauber Costa09998212013-02-22 16:34:55 -08005036 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07005037
5038 if (memcg->use_hierarchy == val)
5039 goto out;
5040
Balbir Singh18f59ea2009-01-07 18:08:07 -08005041 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005042 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08005043 * in the child subtrees. If it is unset, then the change can
5044 * occur, provided the current cgroup has no children.
5045 *
5046 * For the root cgroup, parent_mem is NULL, we allow value to be
5047 * set if there are no children.
5048 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005049 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08005050 (val == 1 || val == 0)) {
Glauber Costab5f99b52013-02-22 16:34:53 -08005051 if (!__memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005052 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005053 else
5054 retval = -EBUSY;
5055 } else
5056 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005057
5058out:
Glauber Costa09998212013-02-22 16:34:55 -08005059 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005060
5061 return retval;
5062}
5063
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005064
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005065static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005066 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005067{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005068 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005069 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005070
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005071 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005073 val += mem_cgroup_read_stat(iter, idx);
5074
5075 if (val < 0) /* race ? */
5076 val = 0;
5077 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005078}
5079
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005080static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005081{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005082 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005083
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005084 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005085 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005086 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005087 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005088 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005089 }
5090
David Rientjesb070e652013-05-07 16:18:09 -07005091 /*
5092 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5093 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5094 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005095 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5096 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005097
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005098 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005099 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005100
5101 return val << PAGE_SHIFT;
5102}
5103
Tejun Heoaf36f902012-04-01 12:09:55 -07005104static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
5105 struct file *file, char __user *buf,
5106 size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005107{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005108 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Tejun Heoaf36f902012-04-01 12:09:55 -07005109 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005110 u64 val;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005111 int name, len;
5112 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005113
5114 type = MEMFILE_TYPE(cft->private);
5115 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005116
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005117 switch (type) {
5118 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005119 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005120 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005121 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005122 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005123 break;
5124 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005125 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005126 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005127 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005128 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005129 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005130 case _KMEM:
5131 val = res_counter_read_u64(&memcg->kmem, name);
5132 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005133 default:
5134 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005135 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005136
5137 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
5138 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005139}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005140
5141static int memcg_update_kmem_limit(struct cgroup *cont, u64 val)
5142{
5143 int ret = -EINVAL;
5144#ifdef CONFIG_MEMCG_KMEM
5145 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5146 /*
5147 * For simplicity, we won't allow this to be disabled. It also can't
5148 * be changed if the cgroup has children already, or if tasks had
5149 * already joined.
5150 *
5151 * If tasks join before we set the limit, a person looking at
5152 * kmem.usage_in_bytes will have no way to determine when it took
5153 * place, which makes the value quite meaningless.
5154 *
5155 * After it first became limited, changes in the value of the limit are
5156 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005157 */
Glauber Costa09998212013-02-22 16:34:55 -08005158 mutex_lock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005159 mutex_lock(&set_limit_mutex);
5160 if (!memcg->kmem_account_flags && val != RESOURCE_MAX) {
Glauber Costab5f99b52013-02-22 16:34:53 -08005161 if (cgroup_task_count(cont) || memcg_has_children(memcg)) {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005162 ret = -EBUSY;
5163 goto out;
5164 }
5165 ret = res_counter_set_limit(&memcg->kmem, val);
5166 VM_BUG_ON(ret);
5167
Glauber Costa55007d82012-12-18 14:22:38 -08005168 ret = memcg_update_cache_sizes(memcg);
5169 if (ret) {
5170 res_counter_set_limit(&memcg->kmem, RESOURCE_MAX);
5171 goto out;
5172 }
Glauber Costa692e89a2013-02-22 16:34:56 -08005173 static_key_slow_inc(&memcg_kmem_enabled_key);
5174 /*
5175 * setting the active bit after the inc will guarantee no one
5176 * starts accounting before all call sites are patched
5177 */
5178 memcg_kmem_set_active(memcg);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005179 } else
5180 ret = res_counter_set_limit(&memcg->kmem, val);
5181out:
5182 mutex_unlock(&set_limit_mutex);
Glauber Costa09998212013-02-22 16:34:55 -08005183 mutex_unlock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005184#endif
5185 return ret;
5186}
5187
Hugh Dickins6d0439902013-02-22 16:35:50 -08005188#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -08005189static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005190{
Glauber Costa55007d82012-12-18 14:22:38 -08005191 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005192 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5193 if (!parent)
Glauber Costa55007d82012-12-18 14:22:38 -08005194 goto out;
5195
Glauber Costa510fc4e2012-12-18 14:21:47 -08005196 memcg->kmem_account_flags = parent->kmem_account_flags;
Glauber Costaa8964b92012-12-18 14:22:09 -08005197 /*
5198 * When that happen, we need to disable the static branch only on those
5199 * memcgs that enabled it. To achieve this, we would be forced to
5200 * complicate the code by keeping track of which memcgs were the ones
5201 * that actually enabled limits, and which ones got it from its
5202 * parents.
5203 *
5204 * It is a lot simpler just to do static_key_slow_inc() on every child
5205 * that is accounted.
5206 */
Glauber Costa55007d82012-12-18 14:22:38 -08005207 if (!memcg_kmem_is_active(memcg))
5208 goto out;
5209
5210 /*
Li Zefan10d5ebf2013-07-08 16:00:33 -07005211 * __mem_cgroup_free() will issue static_key_slow_dec() because this
5212 * memcg is active already. If the later initialization fails then the
5213 * cgroup core triggers the cleanup so we do not have to do it here.
Glauber Costa55007d82012-12-18 14:22:38 -08005214 */
Glauber Costa55007d82012-12-18 14:22:38 -08005215 static_key_slow_inc(&memcg_kmem_enabled_key);
5216
5217 mutex_lock(&set_limit_mutex);
Glauber Costa425c5982013-07-08 16:00:01 -07005218 memcg_stop_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005219 ret = memcg_update_cache_sizes(memcg);
Glauber Costa425c5982013-07-08 16:00:01 -07005220 memcg_resume_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005221 mutex_unlock(&set_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005222out:
5223 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005224}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005225#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005226
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005227/*
5228 * The user of this function is...
5229 * RES_LIMIT.
5230 */
Paul Menage856c13a2008-07-25 01:47:04 -07005231static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
5232 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005233{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005234 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005235 enum res_type type;
5236 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005237 unsigned long long val;
5238 int ret;
5239
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005240 type = MEMFILE_TYPE(cft->private);
5241 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005242
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005243 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005244 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005245 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5246 ret = -EINVAL;
5247 break;
5248 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005249 /* This function does all necessary parse...reuse it */
5250 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005251 if (ret)
5252 break;
5253 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005254 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005255 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005256 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005257 else if (type == _KMEM)
5258 ret = memcg_update_kmem_limit(cont, val);
5259 else
5260 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005261 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005262 case RES_SOFT_LIMIT:
5263 ret = res_counter_memparse_write_strategy(buffer, &val);
5264 if (ret)
5265 break;
5266 /*
5267 * For memsw, soft limits are hard to implement in terms
5268 * of semantics, for now, we support soft limits for
5269 * control without swap
5270 */
5271 if (type == _MEM)
5272 ret = res_counter_set_soft_limit(&memcg->res, val);
5273 else
5274 ret = -EINVAL;
5275 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005276 default:
5277 ret = -EINVAL; /* should be BUG() ? */
5278 break;
5279 }
5280 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005281}
5282
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005283static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5284 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5285{
5286 struct cgroup *cgroup;
5287 unsigned long long min_limit, min_memsw_limit, tmp;
5288
5289 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5290 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5291 cgroup = memcg->css.cgroup;
5292 if (!memcg->use_hierarchy)
5293 goto out;
5294
5295 while (cgroup->parent) {
5296 cgroup = cgroup->parent;
5297 memcg = mem_cgroup_from_cont(cgroup);
5298 if (!memcg->use_hierarchy)
5299 break;
5300 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5301 min_limit = min(min_limit, tmp);
5302 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5303 min_memsw_limit = min(min_memsw_limit, tmp);
5304 }
5305out:
5306 *mem_limit = min_limit;
5307 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005308}
5309
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005310static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005311{
Tejun Heoaf36f902012-04-01 12:09:55 -07005312 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005313 int name;
5314 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005315
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005316 type = MEMFILE_TYPE(event);
5317 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005318
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005319 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005320 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005321 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005322 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005323 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005324 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005325 else if (type == _KMEM)
5326 res_counter_reset_max(&memcg->kmem);
5327 else
5328 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005329 break;
5330 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005331 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005332 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005333 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005334 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005335 else if (type == _KMEM)
5336 res_counter_reset_failcnt(&memcg->kmem);
5337 else
5338 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005339 break;
5340 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005341
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005342 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005343}
5344
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005345static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
5346 struct cftype *cft)
5347{
5348 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
5349}
5350
Daisuke Nishimura02491442010-03-10 15:22:17 -08005351#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005352static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
5353 struct cftype *cft, u64 val)
5354{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005355 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005356
5357 if (val >= (1 << NR_MOVE_TYPE))
5358 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005359
Glauber Costaee5e8472013-02-22 16:34:50 -08005360 /*
5361 * No kind of locking is needed in here, because ->can_attach() will
5362 * check this value once in the beginning of the process, and then carry
5363 * on with stale data. This means that changes to this value will only
5364 * affect task migrations starting after the change.
5365 */
5366 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005367 return 0;
5368}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005369#else
5370static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
5371 struct cftype *cft, u64 val)
5372{
5373 return -ENOSYS;
5374}
5375#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005376
Ying Han406eb0c2011-05-26 16:25:37 -07005377#ifdef CONFIG_NUMA
Wanpeng Liab215882012-07-31 16:43:09 -07005378static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weinerfada52c2012-05-29 15:07:06 -07005379 struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07005380{
5381 int nid;
5382 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
5383 unsigned long node_nr;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005384 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Ying Han406eb0c2011-05-26 16:25:37 -07005385
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005386 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005387 seq_printf(m, "total=%lu", total_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005388 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005389 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005390 seq_printf(m, " N%d=%lu", nid, node_nr);
5391 }
5392 seq_putc(m, '\n');
5393
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005394 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005395 seq_printf(m, "file=%lu", file_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005396 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005397 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005398 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005399 seq_printf(m, " N%d=%lu", nid, node_nr);
5400 }
5401 seq_putc(m, '\n');
5402
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005403 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005404 seq_printf(m, "anon=%lu", anon_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005405 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005406 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005407 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005408 seq_printf(m, " N%d=%lu", nid, node_nr);
5409 }
5410 seq_putc(m, '\n');
5411
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005412 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005413 seq_printf(m, "unevictable=%lu", unevictable_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005414 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005415 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005416 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005417 seq_printf(m, " N%d=%lu", nid, node_nr);
5418 }
5419 seq_putc(m, '\n');
5420 return 0;
5421}
5422#endif /* CONFIG_NUMA */
5423
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005424static inline void mem_cgroup_lru_names_not_uptodate(void)
5425{
5426 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5427}
5428
Wanpeng Liab215882012-07-31 16:43:09 -07005429static int memcg_stat_show(struct cgroup *cont, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005430 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005431{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005432 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005433 struct mem_cgroup *mi;
5434 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005435
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005436 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005437 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005438 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005439 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5440 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005441 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005442
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005443 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5444 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5445 mem_cgroup_read_events(memcg, i));
5446
5447 for (i = 0; i < NR_LRU_LISTS; i++)
5448 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5449 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5450
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005451 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005452 {
5453 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005454 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005455 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005456 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005457 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5458 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005459 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005460
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005461 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5462 long long val = 0;
5463
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005464 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005465 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005466 for_each_mem_cgroup_tree(mi, memcg)
5467 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5468 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5469 }
5470
5471 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5472 unsigned long long val = 0;
5473
5474 for_each_mem_cgroup_tree(mi, memcg)
5475 val += mem_cgroup_read_events(mi, i);
5476 seq_printf(m, "total_%s %llu\n",
5477 mem_cgroup_events_names[i], val);
5478 }
5479
5480 for (i = 0; i < NR_LRU_LISTS; i++) {
5481 unsigned long long val = 0;
5482
5483 for_each_mem_cgroup_tree(mi, memcg)
5484 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5485 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005486 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005487
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005488#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005489 {
5490 int nid, zid;
5491 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005492 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005493 unsigned long recent_rotated[2] = {0, 0};
5494 unsigned long recent_scanned[2] = {0, 0};
5495
5496 for_each_online_node(nid)
5497 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005498 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005499 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005500
Hugh Dickins89abfab2012-05-29 15:06:53 -07005501 recent_rotated[0] += rstat->recent_rotated[0];
5502 recent_rotated[1] += rstat->recent_rotated[1];
5503 recent_scanned[0] += rstat->recent_scanned[0];
5504 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005505 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005506 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5507 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5508 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5509 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005510 }
5511#endif
5512
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005513 return 0;
5514}
5515
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005516static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
5517{
5518 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5519
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005520 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005521}
5522
5523static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
5524 u64 val)
5525{
5526 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5527 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08005528
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005529 if (val > 100)
5530 return -EINVAL;
5531
5532 if (cgrp->parent == NULL)
5533 return -EINVAL;
5534
5535 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08005536
Glauber Costa09998212013-02-22 16:34:55 -08005537 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005538
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005539 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005540 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005541 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005542 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005543 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005544
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005545 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005546
Glauber Costa09998212013-02-22 16:34:55 -08005547 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005548
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005549 return 0;
5550}
5551
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005552static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5553{
5554 struct mem_cgroup_threshold_ary *t;
5555 u64 usage;
5556 int i;
5557
5558 rcu_read_lock();
5559 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005560 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005561 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005562 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005563
5564 if (!t)
5565 goto unlock;
5566
5567 usage = mem_cgroup_usage(memcg, swap);
5568
5569 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005570 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005571 * If it's not true, a threshold was crossed after last
5572 * call of __mem_cgroup_threshold().
5573 */
Phil Carmody5407a562010-05-26 14:42:42 -07005574 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005575
5576 /*
5577 * Iterate backward over array of thresholds starting from
5578 * current_threshold and check if a threshold is crossed.
5579 * If none of thresholds below usage is crossed, we read
5580 * only one element of the array here.
5581 */
5582 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5583 eventfd_signal(t->entries[i].eventfd, 1);
5584
5585 /* i = current_threshold + 1 */
5586 i++;
5587
5588 /*
5589 * Iterate forward over array of thresholds starting from
5590 * current_threshold+1 and check if a threshold is crossed.
5591 * If none of thresholds above usage is crossed, we read
5592 * only one element of the array here.
5593 */
5594 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5595 eventfd_signal(t->entries[i].eventfd, 1);
5596
5597 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005598 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005599unlock:
5600 rcu_read_unlock();
5601}
5602
5603static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5604{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005605 while (memcg) {
5606 __mem_cgroup_threshold(memcg, false);
5607 if (do_swap_account)
5608 __mem_cgroup_threshold(memcg, true);
5609
5610 memcg = parent_mem_cgroup(memcg);
5611 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005612}
5613
5614static int compare_thresholds(const void *a, const void *b)
5615{
5616 const struct mem_cgroup_threshold *_a = a;
5617 const struct mem_cgroup_threshold *_b = b;
5618
5619 return _a->threshold - _b->threshold;
5620}
5621
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005622static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005623{
5624 struct mem_cgroup_eventfd_list *ev;
5625
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005626 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005627 eventfd_signal(ev->eventfd, 1);
5628 return 0;
5629}
5630
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005631static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005632{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005633 struct mem_cgroup *iter;
5634
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005635 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005636 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005637}
5638
5639static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
5640 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005641{
5642 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005643 struct mem_cgroup_thresholds *thresholds;
5644 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005645 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005646 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005647 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005648
5649 ret = res_counter_memparse_write_strategy(args, &threshold);
5650 if (ret)
5651 return ret;
5652
5653 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005654
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005655 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005656 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005657 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005658 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005659 else
5660 BUG();
5661
5662 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5663
5664 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005665 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005666 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5667
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005668 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005669
5670 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005671 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005672 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005673 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005674 ret = -ENOMEM;
5675 goto unlock;
5676 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005677 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005678
5679 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005680 if (thresholds->primary) {
5681 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005682 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005683 }
5684
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005685 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005686 new->entries[size - 1].eventfd = eventfd;
5687 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005688
5689 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005690 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005691 compare_thresholds, NULL);
5692
5693 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005694 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005695 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005696 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005697 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005698 * new->current_threshold will not be used until
5699 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005700 * it here.
5701 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005702 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005703 } else
5704 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005705 }
5706
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005707 /* Free old spare buffer and save old primary buffer as spare */
5708 kfree(thresholds->spare);
5709 thresholds->spare = thresholds->primary;
5710
5711 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005712
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005713 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005714 synchronize_rcu();
5715
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005716unlock:
5717 mutex_unlock(&memcg->thresholds_lock);
5718
5719 return ret;
5720}
5721
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005722static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005723 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005724{
5725 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005726 struct mem_cgroup_thresholds *thresholds;
5727 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005728 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005729 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005730 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005731
5732 mutex_lock(&memcg->thresholds_lock);
5733 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005734 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005735 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005736 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005737 else
5738 BUG();
5739
Anton Vorontsov371528c2012-02-24 05:14:46 +04005740 if (!thresholds->primary)
5741 goto unlock;
5742
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005743 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5744
5745 /* Check if a threshold crossed before removing */
5746 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5747
5748 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005749 size = 0;
5750 for (i = 0; i < thresholds->primary->size; i++) {
5751 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005752 size++;
5753 }
5754
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005755 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005756
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005757 /* Set thresholds array to NULL if we don't have thresholds */
5758 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005759 kfree(new);
5760 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005761 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005762 }
5763
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005764 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005765
5766 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005767 new->current_threshold = -1;
5768 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5769 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005770 continue;
5771
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005772 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005773 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005774 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005775 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005776 * until rcu_assign_pointer(), so it's safe to increment
5777 * it here.
5778 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005779 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005780 }
5781 j++;
5782 }
5783
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005784swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005785 /* Swap primary and spare array */
5786 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005787 /* If all events are unregistered, free the spare array */
5788 if (!new) {
5789 kfree(thresholds->spare);
5790 thresholds->spare = NULL;
5791 }
5792
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005793 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005794
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005795 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005796 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005797unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005798 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005799}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005800
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005801static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
5802 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5803{
5804 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
5805 struct mem_cgroup_eventfd_list *event;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005806 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005807
5808 BUG_ON(type != _OOM_TYPE);
5809 event = kmalloc(sizeof(*event), GFP_KERNEL);
5810 if (!event)
5811 return -ENOMEM;
5812
Michal Hocko1af8efe2011-07-26 16:08:24 -07005813 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005814
5815 event->eventfd = eventfd;
5816 list_add(&event->list, &memcg->oom_notify);
5817
5818 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005819 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005820 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005821 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005822
5823 return 0;
5824}
5825
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005826static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005827 struct cftype *cft, struct eventfd_ctx *eventfd)
5828{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005829 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005830 struct mem_cgroup_eventfd_list *ev, *tmp;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005831 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005832
5833 BUG_ON(type != _OOM_TYPE);
5834
Michal Hocko1af8efe2011-07-26 16:08:24 -07005835 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005836
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005837 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005838 if (ev->eventfd == eventfd) {
5839 list_del(&ev->list);
5840 kfree(ev);
5841 }
5842 }
5843
Michal Hocko1af8efe2011-07-26 16:08:24 -07005844 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005845}
5846
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005847static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
5848 struct cftype *cft, struct cgroup_map_cb *cb)
5849{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005850 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005851
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005852 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005853
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005854 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005855 cb->fill(cb, "under_oom", 1);
5856 else
5857 cb->fill(cb, "under_oom", 0);
5858 return 0;
5859}
5860
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005861static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
5862 struct cftype *cft, u64 val)
5863{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005864 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005865 struct mem_cgroup *parent;
5866
5867 /* cannot set to root cgroup and only 0 and 1 are allowed */
5868 if (!cgrp->parent || !((val == 0) || (val == 1)))
5869 return -EINVAL;
5870
5871 parent = mem_cgroup_from_cont(cgrp->parent);
5872
Glauber Costa09998212013-02-22 16:34:55 -08005873 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005874 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005875 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005876 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005877 return -EINVAL;
5878 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005879 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005880 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005881 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005882 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005883 return 0;
5884}
5885
Andrew Mortonc255a452012-07-31 16:43:02 -07005886#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005887static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005888{
Glauber Costa55007d82012-12-18 14:22:38 -08005889 int ret;
5890
Glauber Costa2633d7a2012-12-18 14:22:34 -08005891 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005892 ret = memcg_propagate_kmem(memcg);
5893 if (ret)
5894 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005895
Glauber Costa1d62e432012-04-09 19:36:33 -03005896 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005897}
Glauber Costae5671df2011-12-11 21:47:01 +00005898
Li Zefan10d5ebf2013-07-08 16:00:33 -07005899static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005900{
Glauber Costa1d62e432012-04-09 19:36:33 -03005901 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005902}
5903
5904static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5905{
5906 if (!memcg_kmem_is_active(memcg))
5907 return;
5908
5909 /*
5910 * kmem charges can outlive the cgroup. In the case of slab
5911 * pages, for instance, a page contain objects from various
5912 * processes. As we prevent from taking a reference for every
5913 * such allocation we have to be careful when doing uncharge
5914 * (see memcg_uncharge_kmem) and here during offlining.
5915 *
5916 * The idea is that that only the _last_ uncharge which sees
5917 * the dead memcg will drop the last reference. An additional
5918 * reference is taken here before the group is marked dead
5919 * which is then paired with css_put during uncharge resp. here.
5920 *
5921 * Although this might sound strange as this path is called from
5922 * css_offline() when the referencemight have dropped down to 0
5923 * and shouldn't be incremented anymore (css_tryget would fail)
5924 * we do not have other options because of the kmem allocations
5925 * lifetime.
5926 */
5927 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005928
5929 memcg_kmem_mark_dead(memcg);
5930
5931 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5932 return;
5933
Glauber Costa7de37682012-12-18 14:22:07 -08005934 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005935 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005936}
Glauber Costae5671df2011-12-11 21:47:01 +00005937#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005938static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005939{
5940 return 0;
5941}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005942
Li Zefan10d5ebf2013-07-08 16:00:33 -07005943static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5944{
5945}
5946
5947static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005948{
5949}
Glauber Costae5671df2011-12-11 21:47:01 +00005950#endif
5951
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005952static struct cftype mem_cgroup_files[] = {
5953 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005954 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005955 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07005956 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005957 .register_event = mem_cgroup_usage_register_event,
5958 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005959 },
5960 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005961 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005962 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005963 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005964 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005965 },
5966 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005967 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005968 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07005969 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005970 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005971 },
5972 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005973 .name = "soft_limit_in_bytes",
5974 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
5975 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005976 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07005977 },
5978 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005979 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005980 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005981 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005982 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005983 },
Balbir Singh8697d332008-02-07 00:13:59 -08005984 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005985 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005986 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005987 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005988 {
5989 .name = "force_empty",
5990 .trigger = mem_cgroup_force_empty_write,
5991 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005992 {
5993 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07005994 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08005995 .write_u64 = mem_cgroup_hierarchy_write,
5996 .read_u64 = mem_cgroup_hierarchy_read,
5997 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005998 {
5999 .name = "swappiness",
6000 .read_u64 = mem_cgroup_swappiness_read,
6001 .write_u64 = mem_cgroup_swappiness_write,
6002 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006003 {
6004 .name = "move_charge_at_immigrate",
6005 .read_u64 = mem_cgroup_move_charge_read,
6006 .write_u64 = mem_cgroup_move_charge_write,
6007 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006008 {
6009 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006010 .read_map = mem_cgroup_oom_control_read,
6011 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006012 .register_event = mem_cgroup_oom_register_event,
6013 .unregister_event = mem_cgroup_oom_unregister_event,
6014 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6015 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006016 {
6017 .name = "pressure_level",
6018 .register_event = vmpressure_register_event,
6019 .unregister_event = vmpressure_unregister_event,
6020 },
Ying Han406eb0c2011-05-26 16:25:37 -07006021#ifdef CONFIG_NUMA
6022 {
6023 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07006024 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006025 },
6026#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006027#ifdef CONFIG_MEMCG_KMEM
6028 {
6029 .name = "kmem.limit_in_bytes",
6030 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6031 .write_string = mem_cgroup_write,
6032 .read = mem_cgroup_read,
6033 },
6034 {
6035 .name = "kmem.usage_in_bytes",
6036 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
6037 .read = mem_cgroup_read,
6038 },
6039 {
6040 .name = "kmem.failcnt",
6041 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6042 .trigger = mem_cgroup_reset,
6043 .read = mem_cgroup_read,
6044 },
6045 {
6046 .name = "kmem.max_usage_in_bytes",
6047 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6048 .trigger = mem_cgroup_reset,
6049 .read = mem_cgroup_read,
6050 },
Glauber Costa749c5412012-12-18 14:23:01 -08006051#ifdef CONFIG_SLABINFO
6052 {
6053 .name = "kmem.slabinfo",
6054 .read_seq_string = mem_cgroup_slabinfo_read,
6055 },
6056#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006057#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006058 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006059};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006060
Michal Hocko2d110852013-02-22 16:34:43 -08006061#ifdef CONFIG_MEMCG_SWAP
6062static struct cftype memsw_cgroup_files[] = {
6063 {
6064 .name = "memsw.usage_in_bytes",
6065 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6066 .read = mem_cgroup_read,
6067 .register_event = mem_cgroup_usage_register_event,
6068 .unregister_event = mem_cgroup_usage_unregister_event,
6069 },
6070 {
6071 .name = "memsw.max_usage_in_bytes",
6072 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6073 .trigger = mem_cgroup_reset,
6074 .read = mem_cgroup_read,
6075 },
6076 {
6077 .name = "memsw.limit_in_bytes",
6078 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6079 .write_string = mem_cgroup_write,
6080 .read = mem_cgroup_read,
6081 },
6082 {
6083 .name = "memsw.failcnt",
6084 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6085 .trigger = mem_cgroup_reset,
6086 .read = mem_cgroup_read,
6087 },
6088 { }, /* terminate */
6089};
6090#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006091static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006092{
6093 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006094 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006095 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006096 /*
6097 * This routine is called against possible nodes.
6098 * But it's BUG to call kmalloc() against offline node.
6099 *
6100 * TODO: this routine can waste much memory for nodes which will
6101 * never be onlined. It's better to use memory hotplug callback
6102 * function.
6103 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006104 if (!node_state(node, N_NORMAL_MEMORY))
6105 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006106 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006107 if (!pn)
6108 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006109
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006110 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6111 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006112 lruvec_init(&mz->lruvec);
Balbir Singhf64c3f52009-09-23 15:56:37 -07006113 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07006114 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006115 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006116 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006117 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006118 return 0;
6119}
6120
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006121static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006122{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006123 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006124}
6125
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006126static struct mem_cgroup *mem_cgroup_alloc(void)
6127{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006128 struct mem_cgroup *memcg;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006129 size_t size = memcg_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006130
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006131 /* Can be very big if nr_node_ids is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08006132 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006133 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006134 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006135 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006136
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006137 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006138 return NULL;
6139
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006140 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6141 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006142 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006143 spin_lock_init(&memcg->pcp_counter_lock);
6144 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006145
6146out_free:
6147 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006148 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006149 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006150 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006151 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006152}
6153
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006154/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006155 * At destroying mem_cgroup, references from swap_cgroup can remain.
6156 * (scanning all at force_empty is too costly...)
6157 *
6158 * Instead of clearing all references at force_empty, we remember
6159 * the number of reference from swap_cgroup and free mem_cgroup when
6160 * it goes down to 0.
6161 *
6162 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006163 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006164
6165static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006166{
Glauber Costac8b2a362012-12-18 14:22:13 -08006167 int node;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006168 size_t size = memcg_size();
Hugh Dickins59927fb2012-03-15 15:17:07 -07006169
Glauber Costac8b2a362012-12-18 14:22:13 -08006170 mem_cgroup_remove_from_trees(memcg);
6171 free_css_id(&mem_cgroup_subsys, &memcg->css);
6172
6173 for_each_node(node)
6174 free_mem_cgroup_per_zone_info(memcg, node);
6175
6176 free_percpu(memcg->stat);
6177
Glauber Costa3f134612012-05-29 15:07:11 -07006178 /*
6179 * We need to make sure that (at least for now), the jump label
6180 * destruction code runs outside of the cgroup lock. This is because
6181 * get_online_cpus(), which is called from the static_branch update,
6182 * can't be called inside the cgroup_lock. cpusets are the ones
6183 * enforcing this dependency, so if they ever change, we might as well.
6184 *
6185 * schedule_work() will guarantee this happens. Be careful if you need
6186 * to move this code around, and make sure it is outside
6187 * the cgroup_lock.
6188 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006189 disarm_static_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006190 if (size < PAGE_SIZE)
6191 kfree(memcg);
6192 else
6193 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006194}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006195
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006196/*
6197 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6198 */
Glauber Costae1aab162011-12-11 21:47:03 +00006199struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006200{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006201 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006202 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006203 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006204}
Glauber Costae1aab162011-12-11 21:47:03 +00006205EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006206
Michal Hocko8787a1d2013-02-22 16:35:39 -08006207static void __init mem_cgroup_soft_limit_tree_init(void)
Balbir Singhf64c3f52009-09-23 15:56:37 -07006208{
6209 struct mem_cgroup_tree_per_node *rtpn;
6210 struct mem_cgroup_tree_per_zone *rtpz;
6211 int tmp, node, zone;
6212
Bob Liu3ed28fa2012-01-12 17:19:04 -08006213 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07006214 tmp = node;
6215 if (!node_state(node, N_NORMAL_MEMORY))
6216 tmp = -1;
6217 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
Michal Hocko8787a1d2013-02-22 16:35:39 -08006218 BUG_ON(!rtpn);
Balbir Singhf64c3f52009-09-23 15:56:37 -07006219
6220 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6221
6222 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6223 rtpz = &rtpn->rb_tree_per_zone[zone];
6224 rtpz->rb_root = RB_ROOT;
6225 spin_lock_init(&rtpz->lock);
6226 }
6227 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07006228}
6229
Li Zefan0eb253e2009-01-15 13:51:25 -08006230static struct cgroup_subsys_state * __ref
Tejun Heo92fb9742012-11-19 08:13:38 -08006231mem_cgroup_css_alloc(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006232{
Glauber Costad142e3e2013-02-22 16:34:52 -08006233 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006234 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006235 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006236
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006237 memcg = mem_cgroup_alloc();
6238 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006239 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006240
Bob Liu3ed28fa2012-01-12 17:19:04 -08006241 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006242 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006243 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006244
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006245 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006246 if (cont->parent == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006247 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006248 res_counter_init(&memcg->res, NULL);
6249 res_counter_init(&memcg->memsw, NULL);
6250 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006251 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006252
Glauber Costad142e3e2013-02-22 16:34:52 -08006253 memcg->last_scanned_node = MAX_NUMNODES;
6254 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006255 memcg->move_charge_at_immigrate = 0;
6256 mutex_init(&memcg->thresholds_lock);
6257 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006258 vmpressure_init(&memcg->vmpressure);
Glauber Costad142e3e2013-02-22 16:34:52 -08006259
6260 return &memcg->css;
6261
6262free_out:
6263 __mem_cgroup_free(memcg);
6264 return ERR_PTR(error);
6265}
6266
6267static int
6268mem_cgroup_css_online(struct cgroup *cont)
6269{
6270 struct mem_cgroup *memcg, *parent;
6271 int error = 0;
6272
6273 if (!cont->parent)
6274 return 0;
6275
Glauber Costa09998212013-02-22 16:34:55 -08006276 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006277 memcg = mem_cgroup_from_cont(cont);
6278 parent = mem_cgroup_from_cont(cont->parent);
6279
6280 memcg->use_hierarchy = parent->use_hierarchy;
6281 memcg->oom_kill_disable = parent->oom_kill_disable;
6282 memcg->swappiness = mem_cgroup_swappiness(parent);
6283
6284 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006285 res_counter_init(&memcg->res, &parent->res);
6286 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006287 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006288
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006289 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006290 * No need to take a reference to the parent because cgroup
6291 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006292 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006293 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006294 res_counter_init(&memcg->res, NULL);
6295 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006296 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006297 /*
6298 * Deeper hierachy with use_hierarchy == false doesn't make
6299 * much sense so let cgroup subsystem know about this
6300 * unfortunate state in our controller.
6301 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006302 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006303 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006304 }
Glauber Costacbe128e32012-04-09 19:36:34 -03006305
6306 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
Glauber Costa09998212013-02-22 16:34:55 -08006307 mutex_unlock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006308 return error;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006309}
6310
Michal Hocko5f578162013-04-29 15:07:17 -07006311/*
6312 * Announce all parents that a group from their hierarchy is gone.
6313 */
6314static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6315{
6316 struct mem_cgroup *parent = memcg;
6317
6318 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006319 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006320
6321 /*
6322 * if the root memcg is not hierarchical we have to check it
6323 * explicitely.
6324 */
6325 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006326 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006327}
6328
Tejun Heo92fb9742012-11-19 08:13:38 -08006329static void mem_cgroup_css_offline(struct cgroup *cont)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006330{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006331 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006332
Li Zefan10d5ebf2013-07-08 16:00:33 -07006333 kmem_cgroup_css_offline(memcg);
6334
Michal Hocko5f578162013-04-29 15:07:17 -07006335 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006336 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006337 mem_cgroup_destroy_all_caches(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006338}
6339
Tejun Heo92fb9742012-11-19 08:13:38 -08006340static void mem_cgroup_css_free(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006341{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006342 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006343
Li Zefan10d5ebf2013-07-08 16:00:33 -07006344 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006345 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006346}
6347
Daisuke Nishimura02491442010-03-10 15:22:17 -08006348#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006349/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006350#define PRECHARGE_COUNT_AT_ONCE 256
6351static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006352{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006353 int ret = 0;
6354 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006355 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006356
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006357 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006358 mc.precharge += count;
6359 /* we don't need css_get for root */
6360 return ret;
6361 }
6362 /* try to charge at once */
6363 if (count > 1) {
6364 struct res_counter *dummy;
6365 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006366 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006367 * by cgroup_lock_live_cgroup() that it is not removed and we
6368 * are still under the same cgroup_mutex. So we can postpone
6369 * css_get().
6370 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006371 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006372 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006373 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006374 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006375 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006376 goto one_by_one;
6377 }
6378 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006379 return ret;
6380 }
6381one_by_one:
6382 /* fall back to one by one charge */
6383 while (count--) {
6384 if (signal_pending(current)) {
6385 ret = -EINTR;
6386 break;
6387 }
6388 if (!batch_count--) {
6389 batch_count = PRECHARGE_COUNT_AT_ONCE;
6390 cond_resched();
6391 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006392 ret = __mem_cgroup_try_charge(NULL,
6393 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006394 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006395 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006396 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006397 mc.precharge++;
6398 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006399 return ret;
6400}
6401
6402/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006403 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006404 * @vma: the vma the pte to be checked belongs
6405 * @addr: the address corresponding to the pte to be checked
6406 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006407 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006408 *
6409 * Returns
6410 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6411 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6412 * move charge. if @target is not NULL, the page is stored in target->page
6413 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006414 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6415 * target for charge migration. if @target is not NULL, the entry is stored
6416 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006417 *
6418 * Called with pte lock held.
6419 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006420union mc_target {
6421 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006422 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006423};
6424
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006425enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006426 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006427 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006428 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006429};
6430
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006431static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6432 unsigned long addr, pte_t ptent)
6433{
6434 struct page *page = vm_normal_page(vma, addr, ptent);
6435
6436 if (!page || !page_mapped(page))
6437 return NULL;
6438 if (PageAnon(page)) {
6439 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006440 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006441 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006442 } else if (!move_file())
6443 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006444 return NULL;
6445 if (!get_page_unless_zero(page))
6446 return NULL;
6447
6448 return page;
6449}
6450
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006451#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006452static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6453 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6454{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006455 struct page *page = NULL;
6456 swp_entry_t ent = pte_to_swp_entry(ptent);
6457
6458 if (!move_anon() || non_swap_entry(ent))
6459 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006460 /*
6461 * Because lookup_swap_cache() updates some statistics counter,
6462 * we call find_get_page() with swapper_space directly.
6463 */
Shaohua Li33806f02013-02-22 16:34:37 -08006464 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006465 if (do_swap_account)
6466 entry->val = ent.val;
6467
6468 return page;
6469}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006470#else
6471static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6472 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6473{
6474 return NULL;
6475}
6476#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006477
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006478static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6479 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6480{
6481 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006482 struct address_space *mapping;
6483 pgoff_t pgoff;
6484
6485 if (!vma->vm_file) /* anonymous vma */
6486 return NULL;
6487 if (!move_file())
6488 return NULL;
6489
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006490 mapping = vma->vm_file->f_mapping;
6491 if (pte_none(ptent))
6492 pgoff = linear_page_index(vma, addr);
6493 else /* pte_file(ptent) is true */
6494 pgoff = pte_to_pgoff(ptent);
6495
6496 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006497 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006498
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006499#ifdef CONFIG_SWAP
6500 /* shmem/tmpfs may report page out on swap: account for that too. */
6501 if (radix_tree_exceptional_entry(page)) {
6502 swp_entry_t swap = radix_to_swp_entry(page);
6503 if (do_swap_account)
6504 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006505 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006506 }
6507#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006508 return page;
6509}
6510
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006511static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006512 unsigned long addr, pte_t ptent, union mc_target *target)
6513{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006514 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006515 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006516 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006517 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006518
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006519 if (pte_present(ptent))
6520 page = mc_handle_present_pte(vma, addr, ptent);
6521 else if (is_swap_pte(ptent))
6522 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006523 else if (pte_none(ptent) || pte_file(ptent))
6524 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006525
6526 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006527 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006528 if (page) {
6529 pc = lookup_page_cgroup(page);
6530 /*
6531 * Do only loose check w/o page_cgroup lock.
6532 * mem_cgroup_move_account() checks the pc is valid or not under
6533 * the lock.
6534 */
6535 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6536 ret = MC_TARGET_PAGE;
6537 if (target)
6538 target->page = page;
6539 }
6540 if (!ret || !target)
6541 put_page(page);
6542 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006543 /* There is a swap entry and a page doesn't exist or isn't charged */
6544 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08006545 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006546 ret = MC_TARGET_SWAP;
6547 if (target)
6548 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006549 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006550 return ret;
6551}
6552
Naoya Horiguchi12724852012-03-21 16:34:28 -07006553#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6554/*
6555 * We don't consider swapping or file mapped pages because THP does not
6556 * support them for now.
6557 * Caller should make sure that pmd_trans_huge(pmd) is true.
6558 */
6559static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6560 unsigned long addr, pmd_t pmd, union mc_target *target)
6561{
6562 struct page *page = NULL;
6563 struct page_cgroup *pc;
6564 enum mc_target_type ret = MC_TARGET_NONE;
6565
6566 page = pmd_page(pmd);
6567 VM_BUG_ON(!page || !PageHead(page));
6568 if (!move_anon())
6569 return ret;
6570 pc = lookup_page_cgroup(page);
6571 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6572 ret = MC_TARGET_PAGE;
6573 if (target) {
6574 get_page(page);
6575 target->page = page;
6576 }
6577 }
6578 return ret;
6579}
6580#else
6581static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6582 unsigned long addr, pmd_t pmd, union mc_target *target)
6583{
6584 return MC_TARGET_NONE;
6585}
6586#endif
6587
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006588static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6589 unsigned long addr, unsigned long end,
6590 struct mm_walk *walk)
6591{
6592 struct vm_area_struct *vma = walk->private;
6593 pte_t *pte;
6594 spinlock_t *ptl;
6595
Naoya Horiguchi12724852012-03-21 16:34:28 -07006596 if (pmd_trans_huge_lock(pmd, vma) == 1) {
6597 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6598 mc.precharge += HPAGE_PMD_NR;
6599 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006600 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006601 }
Dave Hansen03319322011-03-22 16:32:56 -07006602
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006603 if (pmd_trans_unstable(pmd))
6604 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006605 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6606 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006607 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006608 mc.precharge++; /* increment precharge temporarily */
6609 pte_unmap_unlock(pte - 1, ptl);
6610 cond_resched();
6611
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006612 return 0;
6613}
6614
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006615static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6616{
6617 unsigned long precharge;
6618 struct vm_area_struct *vma;
6619
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006620 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006621 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6622 struct mm_walk mem_cgroup_count_precharge_walk = {
6623 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6624 .mm = mm,
6625 .private = vma,
6626 };
6627 if (is_vm_hugetlb_page(vma))
6628 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006629 walk_page_range(vma->vm_start, vma->vm_end,
6630 &mem_cgroup_count_precharge_walk);
6631 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006632 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006633
6634 precharge = mc.precharge;
6635 mc.precharge = 0;
6636
6637 return precharge;
6638}
6639
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006640static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6641{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006642 unsigned long precharge = mem_cgroup_count_precharge(mm);
6643
6644 VM_BUG_ON(mc.moving_task);
6645 mc.moving_task = current;
6646 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006647}
6648
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006649/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6650static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006651{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006652 struct mem_cgroup *from = mc.from;
6653 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006654 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006655
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006656 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006657 if (mc.precharge) {
6658 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6659 mc.precharge = 0;
6660 }
6661 /*
6662 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6663 * we must uncharge here.
6664 */
6665 if (mc.moved_charge) {
6666 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6667 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006668 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006669 /* we must fixup refcnts and charges */
6670 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006671 /* uncharge swap account from the old cgroup */
6672 if (!mem_cgroup_is_root(mc.from))
6673 res_counter_uncharge(&mc.from->memsw,
6674 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006675
6676 for (i = 0; i < mc.moved_swap; i++)
6677 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006678
6679 if (!mem_cgroup_is_root(mc.to)) {
6680 /*
6681 * we charged both to->res and to->memsw, so we should
6682 * uncharge to->res.
6683 */
6684 res_counter_uncharge(&mc.to->res,
6685 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006686 }
Li Zefan40503772013-07-08 16:00:34 -07006687 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006688 mc.moved_swap = 0;
6689 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006690 memcg_oom_recover(from);
6691 memcg_oom_recover(to);
6692 wake_up_all(&mc.waitq);
6693}
6694
6695static void mem_cgroup_clear_mc(void)
6696{
6697 struct mem_cgroup *from = mc.from;
6698
6699 /*
6700 * we must clear moving_task before waking up waiters at the end of
6701 * task migration.
6702 */
6703 mc.moving_task = NULL;
6704 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006705 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006706 mc.from = NULL;
6707 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006708 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006709 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006710}
6711
Li Zefan761b3ef2012-01-31 13:47:36 +08006712static int mem_cgroup_can_attach(struct cgroup *cgroup,
6713 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006714{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006715 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006716 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006717 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Glauber Costaee5e8472013-02-22 16:34:50 -08006718 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006719
Glauber Costaee5e8472013-02-22 16:34:50 -08006720 /*
6721 * We are now commited to this value whatever it is. Changes in this
6722 * tunable will only affect upcoming migrations, not the current one.
6723 * So we need to save it, and keep it going.
6724 */
6725 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6726 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006727 struct mm_struct *mm;
6728 struct mem_cgroup *from = mem_cgroup_from_task(p);
6729
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006730 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006731
6732 mm = get_task_mm(p);
6733 if (!mm)
6734 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006735 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006736 if (mm->owner == p) {
6737 VM_BUG_ON(mc.from);
6738 VM_BUG_ON(mc.to);
6739 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006740 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006741 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006742 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006743 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006744 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006745 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006746 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006747 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006748 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006749
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006750 ret = mem_cgroup_precharge_mc(mm);
6751 if (ret)
6752 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006753 }
6754 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006755 }
6756 return ret;
6757}
6758
Li Zefan761b3ef2012-01-31 13:47:36 +08006759static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
6760 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006761{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006762 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006763}
6764
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006765static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6766 unsigned long addr, unsigned long end,
6767 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006768{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006769 int ret = 0;
6770 struct vm_area_struct *vma = walk->private;
6771 pte_t *pte;
6772 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006773 enum mc_target_type target_type;
6774 union mc_target target;
6775 struct page *page;
6776 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006777
Naoya Horiguchi12724852012-03-21 16:34:28 -07006778 /*
6779 * We don't take compound_lock() here but no race with splitting thp
6780 * happens because:
6781 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6782 * under splitting, which means there's no concurrent thp split,
6783 * - if another thread runs into split_huge_page() just after we
6784 * entered this if-block, the thread must wait for page table lock
6785 * to be unlocked in __split_huge_page_splitting(), where the main
6786 * part of thp split is not executed yet.
6787 */
6788 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006789 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006790 spin_unlock(&vma->vm_mm->page_table_lock);
6791 return 0;
6792 }
6793 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6794 if (target_type == MC_TARGET_PAGE) {
6795 page = target.page;
6796 if (!isolate_lru_page(page)) {
6797 pc = lookup_page_cgroup(page);
6798 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006799 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006800 mc.precharge -= HPAGE_PMD_NR;
6801 mc.moved_charge += HPAGE_PMD_NR;
6802 }
6803 putback_lru_page(page);
6804 }
6805 put_page(page);
6806 }
6807 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006808 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006809 }
6810
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006811 if (pmd_trans_unstable(pmd))
6812 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006813retry:
6814 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6815 for (; addr != end; addr += PAGE_SIZE) {
6816 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006817 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006818
6819 if (!mc.precharge)
6820 break;
6821
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006822 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006823 case MC_TARGET_PAGE:
6824 page = target.page;
6825 if (isolate_lru_page(page))
6826 goto put;
6827 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006828 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006829 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006830 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006831 /* we uncharge from mc.from later. */
6832 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006833 }
6834 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006835put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006836 put_page(page);
6837 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006838 case MC_TARGET_SWAP:
6839 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006840 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006841 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006842 /* we fixup refcnts and charges later. */
6843 mc.moved_swap++;
6844 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006845 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006846 default:
6847 break;
6848 }
6849 }
6850 pte_unmap_unlock(pte - 1, ptl);
6851 cond_resched();
6852
6853 if (addr != end) {
6854 /*
6855 * We have consumed all precharges we got in can_attach().
6856 * We try charge one by one, but don't do any additional
6857 * charges to mc.to if we have failed in charge once in attach()
6858 * phase.
6859 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006860 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006861 if (!ret)
6862 goto retry;
6863 }
6864
6865 return ret;
6866}
6867
6868static void mem_cgroup_move_charge(struct mm_struct *mm)
6869{
6870 struct vm_area_struct *vma;
6871
6872 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006873retry:
6874 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6875 /*
6876 * Someone who are holding the mmap_sem might be waiting in
6877 * waitq. So we cancel all extra charges, wake up all waiters,
6878 * and retry. Because we cancel precharges, we might not be able
6879 * to move enough charges, but moving charge is a best-effort
6880 * feature anyway, so it wouldn't be a big problem.
6881 */
6882 __mem_cgroup_clear_mc();
6883 cond_resched();
6884 goto retry;
6885 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006886 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6887 int ret;
6888 struct mm_walk mem_cgroup_move_charge_walk = {
6889 .pmd_entry = mem_cgroup_move_charge_pte_range,
6890 .mm = mm,
6891 .private = vma,
6892 };
6893 if (is_vm_hugetlb_page(vma))
6894 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006895 ret = walk_page_range(vma->vm_start, vma->vm_end,
6896 &mem_cgroup_move_charge_walk);
6897 if (ret)
6898 /*
6899 * means we have consumed all precharges and failed in
6900 * doing additional charge. Just abandon here.
6901 */
6902 break;
6903 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006904 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006905}
6906
Li Zefan761b3ef2012-01-31 13:47:36 +08006907static void mem_cgroup_move_task(struct cgroup *cont,
6908 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006909{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006910 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006911 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006912
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006913 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006914 if (mc.to)
6915 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006916 mmput(mm);
6917 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006918 if (mc.to)
6919 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006920}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006921#else /* !CONFIG_MMU */
Li Zefan761b3ef2012-01-31 13:47:36 +08006922static int mem_cgroup_can_attach(struct cgroup *cgroup,
6923 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006924{
6925 return 0;
6926}
Li Zefan761b3ef2012-01-31 13:47:36 +08006927static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
6928 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006929{
6930}
Li Zefan761b3ef2012-01-31 13:47:36 +08006931static void mem_cgroup_move_task(struct cgroup *cont,
6932 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006933{
6934}
6935#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006936
Tejun Heof00baae2013-04-15 13:41:15 -07006937/*
6938 * Cgroup retains root cgroups across [un]mount cycles making it necessary
6939 * to verify sane_behavior flag on each mount attempt.
6940 */
6941static void mem_cgroup_bind(struct cgroup *root)
6942{
6943 /*
6944 * use_hierarchy is forced with sane_behavior. cgroup core
6945 * guarantees that @root doesn't have any children, so turning it
6946 * on for the root memcg is enough.
6947 */
6948 if (cgroup_sane_behavior(root))
6949 mem_cgroup_from_cont(root)->use_hierarchy = true;
6950}
6951
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006952struct cgroup_subsys mem_cgroup_subsys = {
6953 .name = "memory",
6954 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08006955 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006956 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006957 .css_offline = mem_cgroup_css_offline,
6958 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006959 .can_attach = mem_cgroup_can_attach,
6960 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08006961 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006962 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07006963 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006964 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006965 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006966};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006967
Andrew Mortonc255a452012-07-31 16:43:02 -07006968#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08006969static int __init enable_swap_account(char *s)
6970{
6971 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07006972 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006973 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07006974 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006975 really_do_swap_account = 0;
6976 return 1;
6977}
Michal Hockoa2c89902011-05-24 17:12:50 -07006978__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006979
Michal Hocko2d110852013-02-22 16:34:43 -08006980static void __init memsw_file_init(void)
6981{
Michal Hocko6acc8b02013-02-22 16:34:45 -08006982 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08006983}
Michal Hocko6acc8b02013-02-22 16:34:45 -08006984
6985static void __init enable_swap_cgroup(void)
6986{
6987 if (!mem_cgroup_disabled() && really_do_swap_account) {
6988 do_swap_account = 1;
6989 memsw_file_init();
6990 }
6991}
6992
Michal Hocko2d110852013-02-22 16:34:43 -08006993#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08006994static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08006995{
6996}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006997#endif
Michal Hocko2d110852013-02-22 16:34:43 -08006998
6999/*
Michal Hocko10813122013-02-22 16:35:41 -08007000 * subsys_initcall() for memory controller.
7001 *
7002 * Some parts like hotcpu_notifier() have to be initialized from this context
7003 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7004 * everything that doesn't depend on a specific mem_cgroup structure should
7005 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007006 */
7007static int __init mem_cgroup_init(void)
7008{
7009 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007010 enable_swap_cgroup();
Michal Hocko8787a1d2013-02-22 16:35:39 -08007011 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007012 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007013 return 0;
7014}
7015subsys_initcall(mem_cgroup_init);