blob: 34d3ca9572d6baed85499d099a0050af3b9bdf66 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080017 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
21 *
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
26 */
27
28#include <linux/res_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080031#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080032#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080033#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
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
Johannes Weineraf7c4b02012-05-29 15:07:08 -070088static const char * const mem_cgroup_stat_names[] = {
89 "cache",
90 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070091 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070092 "mapped_file",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070093 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070094 "swap",
95};
96
Johannes Weinere9f89742011-03-23 16:42:37 -070097enum mem_cgroup_events_index {
98 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
99 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700100 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700102 MEM_CGROUP_EVENTS_NSTATS,
103};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700104
105static const char * const mem_cgroup_events_names[] = {
106 "pgpgin",
107 "pgpgout",
108 "pgfault",
109 "pgmajfault",
110};
111
Sha Zhengju58cf1882013-02-22 16:32:05 -0800112static const char * const mem_cgroup_lru_names[] = {
113 "inactive_anon",
114 "active_anon",
115 "inactive_file",
116 "active_file",
117 "unevictable",
118};
119
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700120/*
121 * Per memcg event counter is incremented at every pagein/pageout. With THP,
122 * it will be incremated by the number of pages. This counter is used for
123 * for trigger some periodic events. This is straightforward and better
124 * than using jiffies etc. to handle periodic memcg event.
125 */
126enum mem_cgroup_events_target {
127 MEM_CGROUP_TARGET_THRESH,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700128 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700129 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700130 MEM_CGROUP_NTARGETS,
131};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700132#define THRESHOLDS_EVENTS_TARGET 128
133#define SOFTLIMIT_EVENTS_TARGET 1024
134#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700135
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800136struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700137 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700138 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700139 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700140 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800141};
142
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800143struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700144 /*
145 * last scanned hierarchy member. Valid only if last_dead_count
146 * matches memcg->dead_count of the hierarchy root group.
147 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700148 struct mem_cgroup *last_visited;
Michal Hocko5f578162013-04-29 15:07:17 -0700149 unsigned long last_dead_count;
150
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800151 /* scan generation, increased every round-trip */
152 unsigned int generation;
153};
154
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800155/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800156 * per-zone information in memory controller.
157 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800158struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800159 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700160 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800161
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800162 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
163
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700164 struct rb_node tree_node; /* RB tree node */
165 unsigned long long usage_in_excess;/* Set to the value by which */
166 /* the soft limit is exceeded*/
167 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700168 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700169 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800170};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800171
172struct mem_cgroup_per_node {
173 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
174};
175
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700176/*
177 * Cgroups above their limits are maintained in a RB-Tree, independent of
178 * their hierarchy representation
179 */
180
181struct mem_cgroup_tree_per_zone {
182 struct rb_root rb_root;
183 spinlock_t lock;
184};
185
186struct mem_cgroup_tree_per_node {
187 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
188};
189
190struct mem_cgroup_tree {
191 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
192};
193
194static struct mem_cgroup_tree soft_limit_tree __read_mostly;
195
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800196struct mem_cgroup_threshold {
197 struct eventfd_ctx *eventfd;
198 u64 threshold;
199};
200
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700201/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800202struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700203 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700204 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800205 /* Size of entries[] */
206 unsigned int size;
207 /* Array of thresholds */
208 struct mem_cgroup_threshold entries[0];
209};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700210
211struct mem_cgroup_thresholds {
212 /* Primary thresholds array */
213 struct mem_cgroup_threshold_ary *primary;
214 /*
215 * Spare threshold array.
216 * This is needed to make mem_cgroup_unregister_event() "never fail".
217 * It must be able to store at least primary->size - 1 entries.
218 */
219 struct mem_cgroup_threshold_ary *spare;
220};
221
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700222/* for OOM */
223struct mem_cgroup_eventfd_list {
224 struct list_head list;
225 struct eventfd_ctx *eventfd;
226};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800227
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700228static void mem_cgroup_threshold(struct mem_cgroup *memcg);
229static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800230
Balbir Singhf64c3f52009-09-23 15:56:37 -0700231/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800232 * The memory controller data structure. The memory controller controls both
233 * page cache and RSS per cgroup. We would eventually like to provide
234 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
235 * to help the administrator determine what knobs to tune.
236 *
237 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800238 * we hit the water mark. May be even add a low water mark, such that
239 * no reclaim occurs from a cgroup at it's low water mark, this is
240 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800241 */
242struct mem_cgroup {
243 struct cgroup_subsys_state css;
244 /*
245 * the counter to account for memory usage
246 */
247 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700248
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700249 /* vmpressure notifications */
250 struct vmpressure vmpressure;
251
Li Zefan465939a2013-07-08 16:00:38 -0700252 /*
253 * the counter to account for mem+swap usage.
254 */
255 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700256
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800257 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800258 * the counter to account for kernel memory usage.
259 */
260 struct res_counter kmem;
261 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800262 * Should the accounting and control be hierarchical, per subtree?
263 */
264 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800265 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700266
267 bool oom_lock;
268 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700269 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700270
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700271 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700272 /* OOM-Killer disable */
273 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800274
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700275 /* set when res.limit == memsw.limit */
276 bool memsw_is_minimum;
277
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800278 /* protect arrays of thresholds */
279 struct mutex thresholds_lock;
280
281 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700282 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700283
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800284 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700285 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700286
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700287 /* For oom notifier event fd */
288 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700289
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800290 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800291 * Should we move charges of a task when a task is moved into this
292 * mem_cgroup ? And what type of charges should we move ?
293 */
Andrew Mortonf894ffa2013-09-12 15:13:35 -0700294 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800295 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700296 * set > 0 if pages under this cgroup are moving to other cgroup.
297 */
298 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700299 /* taken only while moving_account > 0 */
300 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700301 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800302 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800303 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700304 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700305 /*
306 * used when a cpu is offlined or other synchronizations
307 * See mem_cgroup_read_stat().
308 */
309 struct mem_cgroup_stat_cpu nocpu_base;
310 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000311
Michal Hocko5f578162013-04-29 15:07:17 -0700312 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700313#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000314 struct tcp_memcontrol tcp_mem;
315#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800316#if defined(CONFIG_MEMCG_KMEM)
317 /* analogous to slab_common's slab_caches list. per-memcg */
318 struct list_head memcg_slab_caches;
319 /* Not a spinlock, we can take a lot of time walking the list */
320 struct mutex slab_caches_mutex;
321 /* Index in the kmem_cache->memcg_params->memcg_caches array */
322 int kmemcg_id;
323#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800324
325 int last_scanned_node;
326#if MAX_NUMNODES > 1
327 nodemask_t scan_nodes;
328 atomic_t numainfo_events;
329 atomic_t numainfo_updating;
330#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700331
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700332 struct mem_cgroup_per_node *nodeinfo[0];
333 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800334};
335
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800336static size_t memcg_size(void)
337{
338 return sizeof(struct mem_cgroup) +
339 nr_node_ids * sizeof(struct mem_cgroup_per_node);
340}
341
Glauber Costa510fc4e2012-12-18 14:21:47 -0800342/* internal only representation about the status of kmem accounting. */
343enum {
344 KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */
Glauber Costaa8964b92012-12-18 14:22:09 -0800345 KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */
Glauber Costa7de37682012-12-18 14:22:07 -0800346 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800347};
348
Glauber Costaa8964b92012-12-18 14:22:09 -0800349/* We account when limit is on, but only after call sites are patched */
350#define KMEM_ACCOUNTED_MASK \
351 ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED))
Glauber Costa510fc4e2012-12-18 14:21:47 -0800352
353#ifdef CONFIG_MEMCG_KMEM
354static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
355{
356 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
357}
Glauber Costa7de37682012-12-18 14:22:07 -0800358
359static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
360{
361 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
362}
363
Glauber Costaa8964b92012-12-18 14:22:09 -0800364static void memcg_kmem_set_activated(struct mem_cgroup *memcg)
365{
366 set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
367}
368
Glauber Costa55007d82012-12-18 14:22:38 -0800369static void memcg_kmem_clear_activated(struct mem_cgroup *memcg)
370{
371 clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
372}
373
Glauber Costa7de37682012-12-18 14:22:07 -0800374static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
375{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700376 /*
377 * Our caller must use css_get() first, because memcg_uncharge_kmem()
378 * will call css_put() if it sees the memcg is dead.
379 */
380 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800381 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
382 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
383}
384
385static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
386{
387 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
388 &memcg->kmem_account_flags);
389}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800390#endif
391
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800392/* Stuffs for move charges at task migration. */
393/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800394 * Types of charges to be moved. "move_charge_at_immitgrate" and
395 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800396 */
397enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800398 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700399 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800400 NR_MOVE_TYPE,
401};
402
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800403/* "mc" and its members are protected by cgroup_mutex */
404static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800405 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800406 struct mem_cgroup *from;
407 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800408 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800409 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800410 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800411 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800412 struct task_struct *moving_task; /* a task moving charges */
413 wait_queue_head_t waitq; /* a waitq for other context */
414} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700415 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800416 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
417};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800418
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700419static bool move_anon(void)
420{
Glauber Costaee5e8472013-02-22 16:34:50 -0800421 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700422}
423
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700424static bool move_file(void)
425{
Glauber Costaee5e8472013-02-22 16:34:50 -0800426 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700427}
428
Balbir Singh4e416952009-09-23 15:56:39 -0700429/*
430 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
431 * limit reclaim to prevent infinite loops, if they ever occur.
432 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700433#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700434#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700435
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800436enum charge_type {
437 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700438 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800439 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700440 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700441 NR_CHARGE_TYPE,
442};
443
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800444/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800445enum res_type {
446 _MEM,
447 _MEMSWAP,
448 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800449 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800450};
451
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700452#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
453#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800454#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700455/* Used for OOM nofiier */
456#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800457
Balbir Singh75822b42009-09-23 15:56:38 -0700458/*
459 * Reclaim flags for mem_cgroup_hierarchical_reclaim
460 */
461#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
462#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
463#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
464#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
465
Glauber Costa09998212013-02-22 16:34:55 -0800466/*
467 * The memcg_create_mutex will be held whenever a new cgroup is created.
468 * As a consequence, any change that needs to protect against new child cgroups
469 * appearing has to hold it as well.
470 */
471static DEFINE_MUTEX(memcg_create_mutex);
472
Wanpeng Lib2145142012-07-31 16:46:01 -0700473struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
474{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400475 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700476}
477
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700478/* Some nice accessors for the vmpressure. */
479struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
480{
481 if (!memcg)
482 memcg = root_mem_cgroup;
483 return &memcg->vmpressure;
484}
485
486struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
487{
488 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
489}
490
491struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
492{
493 return &mem_cgroup_from_css(css)->vmpressure;
494}
495
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700496static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
497{
498 return (memcg == root_mem_cgroup);
499}
500
Glauber Costae1aab162011-12-11 21:47:03 +0000501/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700502#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000503
Glauber Costae1aab162011-12-11 21:47:03 +0000504void sock_update_memcg(struct sock *sk)
505{
Glauber Costa376be5f2012-01-20 04:57:14 +0000506 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000507 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700508 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000509
510 BUG_ON(!sk->sk_prot->proto_cgroup);
511
Glauber Costaf3f511e2012-01-05 20:16:39 +0000512 /* Socket cloning can throw us here with sk_cgrp already
513 * filled. It won't however, necessarily happen from
514 * process context. So the test for root memcg given
515 * the current task's memcg won't help us in this case.
516 *
517 * Respecting the original socket's memcg is a better
518 * decision in this case.
519 */
520 if (sk->sk_cgrp) {
521 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700522 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000523 return;
524 }
525
Glauber Costae1aab162011-12-11 21:47:03 +0000526 rcu_read_lock();
527 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700528 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700529 if (!mem_cgroup_is_root(memcg) &&
530 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700531 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000532 }
533 rcu_read_unlock();
534 }
535}
536EXPORT_SYMBOL(sock_update_memcg);
537
538void sock_release_memcg(struct sock *sk)
539{
Glauber Costa376be5f2012-01-20 04:57:14 +0000540 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000541 struct mem_cgroup *memcg;
542 WARN_ON(!sk->sk_cgrp->memcg);
543 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700544 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000545 }
546}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000547
548struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
549{
550 if (!memcg || mem_cgroup_is_root(memcg))
551 return NULL;
552
553 return &memcg->tcp_mem.cg_proto;
554}
555EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000556
Glauber Costa3f134612012-05-29 15:07:11 -0700557static void disarm_sock_keys(struct mem_cgroup *memcg)
558{
559 if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
560 return;
561 static_key_slow_dec(&memcg_socket_limit_enabled);
562}
563#else
564static void disarm_sock_keys(struct mem_cgroup *memcg)
565{
566}
567#endif
568
Glauber Costaa8964b92012-12-18 14:22:09 -0800569#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800570/*
571 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
572 * There are two main reasons for not using the css_id for this:
573 * 1) this works better in sparse environments, where we have a lot of memcgs,
574 * but only a few kmem-limited. Or also, if we have, for instance, 200
575 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
576 * 200 entry array for that.
577 *
578 * 2) In order not to violate the cgroup API, we would like to do all memory
579 * allocation in ->create(). At that point, we haven't yet allocated the
580 * css_id. Having a separate index prevents us from messing with the cgroup
581 * core for this
582 *
583 * The current size of the caches array is stored in
584 * memcg_limited_groups_array_size. It will double each time we have to
585 * increase it.
586 */
587static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800588int memcg_limited_groups_array_size;
589
Glauber Costa55007d82012-12-18 14:22:38 -0800590/*
591 * MIN_SIZE is different than 1, because we would like to avoid going through
592 * the alloc/free process all the time. In a small machine, 4 kmem-limited
593 * cgroups is a reasonable guess. In the future, it could be a parameter or
594 * tunable, but that is strictly not necessary.
595 *
596 * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get
597 * this constant directly from cgroup, but it is understandable that this is
598 * better kept as an internal representation in cgroup.c. In any case, the
599 * css_id space is not getting any smaller, and we don't have to necessarily
600 * increase ours as well if it increases.
601 */
602#define MEMCG_CACHES_MIN_SIZE 4
603#define MEMCG_CACHES_MAX_SIZE 65535
604
Glauber Costad7f25f82012-12-18 14:22:40 -0800605/*
606 * A lot of the calls to the cache allocation functions are expected to be
607 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
608 * conditional to this static branch, we'll have to allow modules that does
609 * kmem_cache_alloc and the such to see this symbol as well
610 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800611struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800612EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800613
614static void disarm_kmem_keys(struct mem_cgroup *memcg)
615{
Glauber Costa55007d82012-12-18 14:22:38 -0800616 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800617 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800618 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
619 }
Glauber Costabea207c2012-12-18 14:22:11 -0800620 /*
621 * This check can't live in kmem destruction function,
622 * since the charges will outlive the cgroup
623 */
624 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800625}
626#else
627static void disarm_kmem_keys(struct mem_cgroup *memcg)
628{
629}
630#endif /* CONFIG_MEMCG_KMEM */
631
632static void disarm_static_keys(struct mem_cgroup *memcg)
633{
634 disarm_sock_keys(memcg);
635 disarm_kmem_keys(memcg);
636}
637
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700638static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800639
Balbir Singhf64c3f52009-09-23 15:56:37 -0700640static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700641mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700642{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800643 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700644 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700645}
646
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700647struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100648{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700649 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100650}
651
Balbir Singhf64c3f52009-09-23 15:56:37 -0700652static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700653page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700654{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700655 int nid = page_to_nid(page);
656 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700657
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700658 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700659}
660
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700661static struct mem_cgroup_tree_per_zone *
662soft_limit_tree_node_zone(int nid, int zid)
663{
664 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
665}
666
667static struct mem_cgroup_tree_per_zone *
668soft_limit_tree_from_page(struct page *page)
669{
670 int nid = page_to_nid(page);
671 int zid = page_zonenum(page);
672
673 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
674}
675
676static void
677__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
678 struct mem_cgroup_per_zone *mz,
679 struct mem_cgroup_tree_per_zone *mctz,
680 unsigned long long new_usage_in_excess)
681{
682 struct rb_node **p = &mctz->rb_root.rb_node;
683 struct rb_node *parent = NULL;
684 struct mem_cgroup_per_zone *mz_node;
685
686 if (mz->on_tree)
687 return;
688
689 mz->usage_in_excess = new_usage_in_excess;
690 if (!mz->usage_in_excess)
691 return;
692 while (*p) {
693 parent = *p;
694 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
695 tree_node);
696 if (mz->usage_in_excess < mz_node->usage_in_excess)
697 p = &(*p)->rb_left;
698 /*
699 * We can't avoid mem cgroups that are over their soft
700 * limit by the same amount
701 */
702 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
703 p = &(*p)->rb_right;
704 }
705 rb_link_node(&mz->tree_node, parent, p);
706 rb_insert_color(&mz->tree_node, &mctz->rb_root);
707 mz->on_tree = true;
708}
709
710static void
711__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
712 struct mem_cgroup_per_zone *mz,
713 struct mem_cgroup_tree_per_zone *mctz)
714{
715 if (!mz->on_tree)
716 return;
717 rb_erase(&mz->tree_node, &mctz->rb_root);
718 mz->on_tree = false;
719}
720
721static void
722mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
723 struct mem_cgroup_per_zone *mz,
724 struct mem_cgroup_tree_per_zone *mctz)
725{
726 spin_lock(&mctz->lock);
727 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
728 spin_unlock(&mctz->lock);
729}
730
731
732static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
733{
734 unsigned long long excess;
735 struct mem_cgroup_per_zone *mz;
736 struct mem_cgroup_tree_per_zone *mctz;
737 int nid = page_to_nid(page);
738 int zid = page_zonenum(page);
739 mctz = soft_limit_tree_from_page(page);
740
741 /*
742 * Necessary to update all ancestors when hierarchy is used.
743 * because their event counter is not touched.
744 */
745 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
746 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
747 excess = res_counter_soft_limit_excess(&memcg->res);
748 /*
749 * We have to update the tree if mz is on RB-tree or
750 * mem is over its softlimit.
751 */
752 if (excess || mz->on_tree) {
753 spin_lock(&mctz->lock);
754 /* if on-tree, remove it */
755 if (mz->on_tree)
756 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
757 /*
758 * Insert again. mz->usage_in_excess will be updated.
759 * If excess is 0, no tree ops.
760 */
761 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
762 spin_unlock(&mctz->lock);
763 }
764 }
765}
766
767static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
768{
769 int node, zone;
770 struct mem_cgroup_per_zone *mz;
771 struct mem_cgroup_tree_per_zone *mctz;
772
773 for_each_node(node) {
774 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
775 mz = mem_cgroup_zoneinfo(memcg, node, zone);
776 mctz = soft_limit_tree_node_zone(node, zone);
777 mem_cgroup_remove_exceeded(memcg, mz, mctz);
778 }
779 }
780}
781
782static struct mem_cgroup_per_zone *
783__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
784{
785 struct rb_node *rightmost = NULL;
786 struct mem_cgroup_per_zone *mz;
787
788retry:
789 mz = NULL;
790 rightmost = rb_last(&mctz->rb_root);
791 if (!rightmost)
792 goto done; /* Nothing to reclaim from */
793
794 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
795 /*
796 * Remove the node now but someone else can add it back,
797 * we will to add it back at the end of reclaim to its correct
798 * position in the tree.
799 */
800 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
801 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
802 !css_tryget(&mz->memcg->css))
803 goto retry;
804done:
805 return mz;
806}
807
808static struct mem_cgroup_per_zone *
809mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
810{
811 struct mem_cgroup_per_zone *mz;
812
813 spin_lock(&mctz->lock);
814 mz = __mem_cgroup_largest_soft_limit_node(mctz);
815 spin_unlock(&mctz->lock);
816 return mz;
817}
818
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700819/*
820 * Implementation Note: reading percpu statistics for memcg.
821 *
822 * Both of vmstat[] and percpu_counter has threshold and do periodic
823 * synchronization to implement "quick" read. There are trade-off between
824 * reading cost and precision of value. Then, we may have a chance to implement
825 * a periodic synchronizion of counter in memcg's counter.
826 *
827 * But this _read() function is used for user interface now. The user accounts
828 * memory usage by memory cgroup and he _always_ requires exact value because
829 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
830 * have to visit all online cpus and make sum. So, for now, unnecessary
831 * synchronization is not implemented. (just implemented for cpu hotplug)
832 *
833 * If there are kernel internal actions which can make use of some not-exact
834 * value, and reading all cpu value can be performance bottleneck in some
835 * common workload, threashold and synchonization as vmstat[] should be
836 * implemented.
837 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700838static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700839 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800840{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700841 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800842 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800843
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700844 get_online_cpus();
845 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700846 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700847#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700848 spin_lock(&memcg->pcp_counter_lock);
849 val += memcg->nocpu_base.count[idx];
850 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700851#endif
852 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800853 return val;
854}
855
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700856static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700857 bool charge)
858{
859 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700860 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700861}
862
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700863static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700864 enum mem_cgroup_events_index idx)
865{
866 unsigned long val = 0;
867 int cpu;
868
David Rientjes9c567512013-10-16 13:46:43 -0700869 get_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700870 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700871 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700872#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700873 spin_lock(&memcg->pcp_counter_lock);
874 val += memcg->nocpu_base.events[idx];
875 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700876#endif
David Rientjes9c567512013-10-16 13:46:43 -0700877 put_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700878 return val;
879}
880
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700881static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700882 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700883 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800884{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800885 preempt_disable();
886
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700887 /*
888 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
889 * counted as CACHE even if it's on ANON LRU.
890 */
891 if (anon)
892 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700893 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800894 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700895 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700896 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700897
David Rientjesb070e652013-05-07 16:18:09 -0700898 if (PageTransHuge(page))
899 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
900 nr_pages);
901
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800902 /* pagein of a big page is an event. So, ignore page size */
903 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700904 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800905 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700906 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800907 nr_pages = -nr_pages; /* for event */
908 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800909
Johannes Weiner13114712012-05-29 15:07:07 -0700910 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800911
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800912 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800913}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800914
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700915unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700916mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700917{
918 struct mem_cgroup_per_zone *mz;
919
920 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
921 return mz->lru_size[lru];
922}
923
924static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700925mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700926 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700927{
928 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700929 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700930 unsigned long ret = 0;
931
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700932 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700933
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700934 for_each_lru(lru) {
935 if (BIT(lru) & lru_mask)
936 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700937 }
938 return ret;
939}
940
941static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700942mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700943 int nid, unsigned int lru_mask)
944{
Ying Han889976d2011-05-26 16:25:33 -0700945 u64 total = 0;
946 int zid;
947
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700948 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700949 total += mem_cgroup_zone_nr_lru_pages(memcg,
950 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700951
Ying Han889976d2011-05-26 16:25:33 -0700952 return total;
953}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700954
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700955static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700956 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800957{
Ying Han889976d2011-05-26 16:25:33 -0700958 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800959 u64 total = 0;
960
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800961 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700962 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800963 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800964}
965
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800966static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
967 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800968{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700969 unsigned long val, next;
970
Johannes Weiner13114712012-05-29 15:07:07 -0700971 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700972 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700973 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800974 if ((long)next - (long)val < 0) {
975 switch (target) {
976 case MEM_CGROUP_TARGET_THRESH:
977 next = val + THRESHOLDS_EVENTS_TARGET;
978 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700979 case MEM_CGROUP_TARGET_SOFTLIMIT:
980 next = val + SOFTLIMIT_EVENTS_TARGET;
981 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800982 case MEM_CGROUP_TARGET_NUMAINFO:
983 next = val + NUMAINFO_EVENTS_TARGET;
984 break;
985 default:
986 break;
987 }
988 __this_cpu_write(memcg->stat->targets[target], next);
989 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700990 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800991 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800992}
993
994/*
995 * Check events in order.
996 *
997 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700998static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800999{
Steven Rostedt47994012011-11-02 13:38:33 -07001000 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001001 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001002 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1003 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001004 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001005 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001006
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001007 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1008 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001009#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001010 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1011 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001012#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001013 preempt_enable();
1014
1015 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001016 if (unlikely(do_softlimit))
1017 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001018#if MAX_NUMNODES > 1
1019 if (unlikely(do_numainfo))
1020 atomic_inc(&memcg->numainfo_events);
1021#endif
1022 } else
1023 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001024}
1025
Balbir Singhcf475ad2008-04-29 01:00:16 -07001026struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001027{
Balbir Singh31a78f22008-09-28 23:09:31 +01001028 /*
1029 * mm_update_next_owner() may clear mm->owner to NULL
1030 * if it races with swapoff, page migration, etc.
1031 * So this can be called with p == NULL.
1032 */
1033 if (unlikely(!p))
1034 return NULL;
1035
Tejun Heo8af01f52013-08-08 20:11:22 -04001036 return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001037}
1038
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001039struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001040{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001041 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001042
1043 if (!mm)
1044 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001045 /*
1046 * Because we have no locks, mm->owner's may be being moved to other
1047 * cgroup. We use css_tryget() here even if this looks
1048 * pessimistic (rather than adding locks here).
1049 */
1050 rcu_read_lock();
1051 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001052 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1053 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001054 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001055 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001056 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001057 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001058}
1059
Michal Hocko16248d82013-04-29 15:07:19 -07001060/*
1061 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1062 * ref. count) or NULL if the whole root's subtree has been visited.
1063 *
1064 * helper function to be used by mem_cgroup_iter
1065 */
1066static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001067 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001068{
Tejun Heo492eb212013-08-08 20:11:25 -04001069 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001070
Tejun Heobd8815a2013-08-08 20:11:27 -04001071 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001072skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001073 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001074
1075 /*
1076 * Even if we found a group we have to make sure it is
1077 * alive. css && !memcg means that the groups should be
1078 * skipped and we should continue the tree walk.
1079 * last_visited css is safe to use because it is
1080 * protected by css_get and the tree walk is rcu safe.
1081 */
Tejun Heo492eb212013-08-08 20:11:25 -04001082 if (next_css) {
1083 struct mem_cgroup *mem = mem_cgroup_from_css(next_css);
1084
Andrew Morton694fbc02013-09-24 15:27:37 -07001085 if (css_tryget(&mem->css))
1086 return mem;
1087 else {
Tejun Heo492eb212013-08-08 20:11:25 -04001088 prev_css = next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001089 goto skip_node;
1090 }
1091 }
1092
1093 return NULL;
1094}
1095
Johannes Weiner519ebea2013-07-03 15:04:51 -07001096static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1097{
1098 /*
1099 * When a group in the hierarchy below root is destroyed, the
1100 * hierarchy iterator can no longer be trusted since it might
1101 * have pointed to the destroyed group. Invalidate it.
1102 */
1103 atomic_inc(&root->dead_count);
1104}
1105
1106static struct mem_cgroup *
1107mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1108 struct mem_cgroup *root,
1109 int *sequence)
1110{
1111 struct mem_cgroup *position = NULL;
1112 /*
1113 * A cgroup destruction happens in two stages: offlining and
1114 * release. They are separated by a RCU grace period.
1115 *
1116 * If the iterator is valid, we may still race with an
1117 * offlining. The RCU lock ensures the object won't be
1118 * released, tryget will fail if we lost the race.
1119 */
1120 *sequence = atomic_read(&root->dead_count);
1121 if (iter->last_dead_count == *sequence) {
1122 smp_rmb();
1123 position = iter->last_visited;
1124 if (position && !css_tryget(&position->css))
1125 position = NULL;
1126 }
1127 return position;
1128}
1129
1130static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1131 struct mem_cgroup *last_visited,
1132 struct mem_cgroup *new_position,
1133 int sequence)
1134{
1135 if (last_visited)
1136 css_put(&last_visited->css);
1137 /*
1138 * We store the sequence count from the time @last_visited was
1139 * loaded successfully instead of rereading it here so that we
1140 * don't lose destruction events in between. We could have
1141 * raced with the destruction of @new_position after all.
1142 */
1143 iter->last_visited = new_position;
1144 smp_wmb();
1145 iter->last_dead_count = sequence;
1146}
1147
Johannes Weiner56600482012-01-12 17:17:59 -08001148/**
1149 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1150 * @root: hierarchy root
1151 * @prev: previously returned memcg, NULL on first invocation
1152 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1153 *
1154 * Returns references to children of the hierarchy below @root, or
1155 * @root itself, or %NULL after a full round-trip.
1156 *
1157 * Caller must pass the return value in @prev on subsequent
1158 * invocations for reference counting, or use mem_cgroup_iter_break()
1159 * to cancel a hierarchy walk before the round-trip is complete.
1160 *
1161 * Reclaimers can specify a zone and a priority level in @reclaim to
1162 * divide up the memcgs in the hierarchy among all concurrent
1163 * reclaimers operating on the same zone and priority.
1164 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001165struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001166 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001167 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001168{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001169 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001170 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001171
Andrew Morton694fbc02013-09-24 15:27:37 -07001172 if (mem_cgroup_disabled())
1173 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001174
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001175 if (!root)
1176 root = root_mem_cgroup;
1177
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001178 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001179 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001180
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001181 if (!root->use_hierarchy && root != root_mem_cgroup) {
1182 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001183 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001184 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001185 }
1186
Michal Hocko542f85f2013-04-29 15:07:15 -07001187 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001188 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001189 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001190 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001191
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001192 if (reclaim) {
1193 int nid = zone_to_nid(reclaim->zone);
1194 int zid = zone_idx(reclaim->zone);
1195 struct mem_cgroup_per_zone *mz;
1196
1197 mz = mem_cgroup_zoneinfo(root, nid, zid);
1198 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001199 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001200 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001201 goto out_unlock;
1202 }
Michal Hocko5f578162013-04-29 15:07:17 -07001203
Johannes Weiner519ebea2013-07-03 15:04:51 -07001204 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001205 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001206
Andrew Morton694fbc02013-09-24 15:27:37 -07001207 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001208
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001209 if (reclaim) {
Johannes Weiner519ebea2013-07-03 15:04:51 -07001210 mem_cgroup_iter_update(iter, last_visited, memcg, seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001211
Michal Hocko19f39402013-04-29 15:07:18 -07001212 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001213 iter->generation++;
1214 else if (!prev && memcg)
1215 reclaim->generation = iter->generation;
1216 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001217
Andrew Morton694fbc02013-09-24 15:27:37 -07001218 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001219 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001220 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001221out_unlock:
1222 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001223out_css_put:
1224 if (prev && prev != root)
1225 css_put(&prev->css);
1226
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001227 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001228}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001229
Johannes Weiner56600482012-01-12 17:17:59 -08001230/**
1231 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1232 * @root: hierarchy root
1233 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1234 */
1235void mem_cgroup_iter_break(struct mem_cgroup *root,
1236 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001237{
1238 if (!root)
1239 root = root_mem_cgroup;
1240 if (prev && prev != root)
1241 css_put(&prev->css);
1242}
1243
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001244/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001245 * Iteration constructs for visiting all cgroups (under a tree). If
1246 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1247 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001248 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001249#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001250 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001251 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001252 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001253
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001254#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001255 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001256 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001257 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001258
David Rientjes68ae5642012-12-12 13:51:57 -08001259void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001260{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001261 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001262
Ying Han456f9982011-05-26 16:25:38 -07001263 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001264 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1265 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001266 goto out;
1267
1268 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001269 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001270 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1271 break;
1272 case PGMAJFAULT:
1273 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001274 break;
1275 default:
1276 BUG();
1277 }
1278out:
1279 rcu_read_unlock();
1280}
David Rientjes68ae5642012-12-12 13:51:57 -08001281EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001282
Johannes Weiner925b7672012-01-12 17:18:15 -08001283/**
1284 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1285 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001286 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001287 *
1288 * Returns the lru list vector holding pages for the given @zone and
1289 * @mem. This can be the global zone lruvec, if the memory controller
1290 * is disabled.
1291 */
1292struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1293 struct mem_cgroup *memcg)
1294{
1295 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001296 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001297
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001298 if (mem_cgroup_disabled()) {
1299 lruvec = &zone->lruvec;
1300 goto out;
1301 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001302
1303 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001304 lruvec = &mz->lruvec;
1305out:
1306 /*
1307 * Since a node can be onlined after the mem_cgroup was created,
1308 * we have to be prepared to initialize lruvec->zone here;
1309 * and if offlined then reonlined, we need to reinitialize it.
1310 */
1311 if (unlikely(lruvec->zone != zone))
1312 lruvec->zone = zone;
1313 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001314}
1315
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001316/*
1317 * Following LRU functions are allowed to be used without PCG_LOCK.
1318 * Operations are called by routine of global LRU independently from memcg.
1319 * What we have to take care of here is validness of pc->mem_cgroup.
1320 *
1321 * Changes to pc->mem_cgroup happens when
1322 * 1. charge
1323 * 2. moving account
1324 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1325 * It is added to LRU before charge.
1326 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1327 * When moving account, the page is not on LRU. It's isolated.
1328 */
1329
Johannes Weiner925b7672012-01-12 17:18:15 -08001330/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001331 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001332 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001333 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001334 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001335struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001336{
1337 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001338 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001339 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001340 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001341
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001342 if (mem_cgroup_disabled()) {
1343 lruvec = &zone->lruvec;
1344 goto out;
1345 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001346
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001347 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001348 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001349
1350 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001351 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001352 * an uncharged page off lru does nothing to secure
1353 * its former mem_cgroup from sudden removal.
1354 *
1355 * Our caller holds lru_lock, and PageCgroupUsed is updated
1356 * under page_cgroup lock: between them, they make all uses
1357 * of pc->mem_cgroup safe.
1358 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001359 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001360 pc->mem_cgroup = memcg = root_mem_cgroup;
1361
Johannes Weiner925b7672012-01-12 17:18:15 -08001362 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001363 lruvec = &mz->lruvec;
1364out:
1365 /*
1366 * Since a node can be onlined after the mem_cgroup was created,
1367 * we have to be prepared to initialize lruvec->zone here;
1368 * and if offlined then reonlined, we need to reinitialize it.
1369 */
1370 if (unlikely(lruvec->zone != zone))
1371 lruvec->zone = zone;
1372 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001373}
1374
1375/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001376 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1377 * @lruvec: mem_cgroup per zone lru vector
1378 * @lru: index of lru list the page is sitting on
1379 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001380 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001381 * This function must be called when a page is added to or removed from an
1382 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001383 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001384void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1385 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001386{
1387 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001388 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001389
1390 if (mem_cgroup_disabled())
1391 return;
1392
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001393 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1394 lru_size = mz->lru_size + lru;
1395 *lru_size += nr_pages;
1396 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001397}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001398
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001399/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001400 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001401 * hierarchy subtree
1402 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001403bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1404 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001405{
Johannes Weiner91c637342012-05-29 15:06:24 -07001406 if (root_memcg == memcg)
1407 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001408 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001409 return false;
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001410 return css_is_ancestor(&memcg->css, &root_memcg->css);
1411}
1412
1413static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1414 struct mem_cgroup *memcg)
1415{
1416 bool ret;
1417
Johannes Weiner91c637342012-05-29 15:06:24 -07001418 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001419 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001420 rcu_read_unlock();
1421 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001422}
1423
David Rientjesffbdccf2013-07-03 15:01:23 -07001424bool task_in_mem_cgroup(struct task_struct *task,
1425 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001426{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001427 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001428 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001429 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001430
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001431 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001432 if (p) {
1433 curr = try_get_mem_cgroup_from_mm(p->mm);
1434 task_unlock(p);
1435 } else {
1436 /*
1437 * All threads may have already detached their mm's, but the oom
1438 * killer still needs to detect if they have already been oom
1439 * killed to prevent needlessly killing additional tasks.
1440 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001441 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001442 curr = mem_cgroup_from_task(task);
1443 if (curr)
1444 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001445 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001446 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001447 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001448 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001449 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001450 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001451 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001452 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1453 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001454 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001455 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001456 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001457 return ret;
1458}
1459
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001460int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001461{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001462 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001463 unsigned long inactive;
1464 unsigned long active;
1465 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001466
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001467 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1468 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001469
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001470 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1471 if (gb)
1472 inactive_ratio = int_sqrt(10 * gb);
1473 else
1474 inactive_ratio = 1;
1475
Johannes Weiner9b272972011-11-02 13:38:23 -07001476 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001477}
1478
Balbir Singh6d61ef42009-01-07 18:08:06 -08001479#define mem_cgroup_from_res_counter(counter, member) \
1480 container_of(counter, struct mem_cgroup, member)
1481
Johannes Weiner19942822011-02-01 15:52:43 -08001482/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001483 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001484 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001485 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001486 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001487 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001488 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001489static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001490{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001491 unsigned long long margin;
1492
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001493 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001494 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001495 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001496 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001497}
1498
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001499int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001500{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001501 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001502 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001503 return vm_swappiness;
1504
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001505 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001506}
1507
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001508/*
1509 * memcg->moving_account is used for checking possibility that some thread is
1510 * calling move_account(). When a thread on CPU-A starts moving pages under
1511 * a memcg, other threads should check memcg->moving_account under
1512 * rcu_read_lock(), like this:
1513 *
1514 * CPU-A CPU-B
1515 * rcu_read_lock()
1516 * memcg->moving_account+1 if (memcg->mocing_account)
1517 * take heavy locks.
1518 * synchronize_rcu() update something.
1519 * rcu_read_unlock()
1520 * start move here.
1521 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001522
1523/* for quick checking without looking up memcg */
1524atomic_t memcg_moving __read_mostly;
1525
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001526static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001527{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001528 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001529 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001530 synchronize_rcu();
1531}
1532
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001533static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001534{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001535 /*
1536 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1537 * We check NULL in callee rather than caller.
1538 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001539 if (memcg) {
1540 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001541 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001542 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001543}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001544
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001545/*
1546 * 2 routines for checking "mem" is under move_account() or not.
1547 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001548 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1549 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001550 * pc->mem_cgroup may be overwritten.
1551 *
1552 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1553 * under hierarchy of moving cgroups. This is for
1554 * waiting at hith-memory prressure caused by "move".
1555 */
1556
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001557static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001558{
1559 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001560 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001561}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001562
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001563static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001564{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001565 struct mem_cgroup *from;
1566 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001567 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001568 /*
1569 * Unlike task_move routines, we access mc.to, mc.from not under
1570 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1571 */
1572 spin_lock(&mc.lock);
1573 from = mc.from;
1574 to = mc.to;
1575 if (!from)
1576 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001577
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001578 ret = mem_cgroup_same_or_subtree(memcg, from)
1579 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001580unlock:
1581 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001582 return ret;
1583}
1584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001585static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001586{
1587 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001588 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001589 DEFINE_WAIT(wait);
1590 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1591 /* moving charge context might have finished. */
1592 if (mc.moving_task)
1593 schedule();
1594 finish_wait(&mc.waitq, &wait);
1595 return true;
1596 }
1597 }
1598 return false;
1599}
1600
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001601/*
1602 * Take this lock when
1603 * - a code tries to modify page's memcg while it's USED.
1604 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001605 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001606 */
1607static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1608 unsigned long *flags)
1609{
1610 spin_lock_irqsave(&memcg->move_lock, *flags);
1611}
1612
1613static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1614 unsigned long *flags)
1615{
1616 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1617}
1618
Sha Zhengju58cf1882013-02-22 16:32:05 -08001619#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001620/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001621 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001622 * @memcg: The memory cgroup that went over limit
1623 * @p: Task that is going to be killed
1624 *
1625 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1626 * enabled
1627 */
1628void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1629{
1630 struct cgroup *task_cgrp;
1631 struct cgroup *mem_cgrp;
1632 /*
1633 * Need a buffer in BSS, can't rely on allocations. The code relies
1634 * on the assumption that OOM is serialized for memory controller.
1635 * If this assumption is broken, revisit this code.
1636 */
1637 static char memcg_name[PATH_MAX];
1638 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001639 struct mem_cgroup *iter;
1640 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001641
Sha Zhengju58cf1882013-02-22 16:32:05 -08001642 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001643 return;
1644
Balbir Singhe2224322009-04-02 16:57:39 -07001645 rcu_read_lock();
1646
1647 mem_cgrp = memcg->css.cgroup;
1648 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1649
1650 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1651 if (ret < 0) {
1652 /*
1653 * Unfortunately, we are unable to convert to a useful name
1654 * But we'll still print out the usage information
1655 */
1656 rcu_read_unlock();
1657 goto done;
1658 }
1659 rcu_read_unlock();
1660
Andrew Mortond0451972013-02-22 16:32:06 -08001661 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001662
1663 rcu_read_lock();
1664 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1665 if (ret < 0) {
1666 rcu_read_unlock();
1667 goto done;
1668 }
1669 rcu_read_unlock();
1670
1671 /*
1672 * Continues from above, so we don't need an KERN_ level
1673 */
Andrew Mortond0451972013-02-22 16:32:06 -08001674 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001675done:
1676
Andrew Mortond0451972013-02-22 16:32:06 -08001677 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001678 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1679 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1680 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001681 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001682 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1683 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1684 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001685 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001686 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1687 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1688 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001689
1690 for_each_mem_cgroup_tree(iter, memcg) {
1691 pr_info("Memory cgroup stats");
1692
1693 rcu_read_lock();
1694 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1695 if (!ret)
1696 pr_cont(" for %s", memcg_name);
1697 rcu_read_unlock();
1698 pr_cont(":");
1699
1700 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1701 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1702 continue;
1703 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1704 K(mem_cgroup_read_stat(iter, i)));
1705 }
1706
1707 for (i = 0; i < NR_LRU_LISTS; i++)
1708 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1709 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1710
1711 pr_cont("\n");
1712 }
Balbir Singhe2224322009-04-02 16:57:39 -07001713}
1714
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001715/*
1716 * This function returns the number of memcg under hierarchy tree. Returns
1717 * 1(self count) if no children.
1718 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001719static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001720{
1721 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001722 struct mem_cgroup *iter;
1723
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001724 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001725 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001726 return num;
1727}
1728
Balbir Singh6d61ef42009-01-07 18:08:06 -08001729/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001730 * Return the memory (and swap, if configured) limit for a memcg.
1731 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001732static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001733{
1734 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001735
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001736 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001737
David Rientjesa63d83f2010-08-09 17:19:46 -07001738 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001739 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001740 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001741 if (mem_cgroup_swappiness(memcg)) {
1742 u64 memsw;
1743
1744 limit += total_swap_pages << PAGE_SHIFT;
1745 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1746
1747 /*
1748 * If memsw is finite and limits the amount of swap space
1749 * available to this memcg, return that limit.
1750 */
1751 limit = min(limit, memsw);
1752 }
1753
1754 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001755}
1756
David Rientjes19965462012-12-11 16:00:26 -08001757static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1758 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001759{
1760 struct mem_cgroup *iter;
1761 unsigned long chosen_points = 0;
1762 unsigned long totalpages;
1763 unsigned int points = 0;
1764 struct task_struct *chosen = NULL;
1765
David Rientjes876aafb2012-07-31 16:43:48 -07001766 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001767 * If current has a pending SIGKILL or is exiting, then automatically
1768 * select it. The goal is to allow it to allocate so that it may
1769 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001770 */
David Rientjes465adcf2013-04-29 15:08:45 -07001771 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001772 set_thread_flag(TIF_MEMDIE);
1773 return;
1774 }
1775
1776 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001777 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1778 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001779 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001780 struct task_struct *task;
1781
Tejun Heo72ec7022013-08-08 20:11:26 -04001782 css_task_iter_start(&iter->css, &it);
1783 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001784 switch (oom_scan_process_thread(task, totalpages, NULL,
1785 false)) {
1786 case OOM_SCAN_SELECT:
1787 if (chosen)
1788 put_task_struct(chosen);
1789 chosen = task;
1790 chosen_points = ULONG_MAX;
1791 get_task_struct(chosen);
1792 /* fall through */
1793 case OOM_SCAN_CONTINUE:
1794 continue;
1795 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001796 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001797 mem_cgroup_iter_break(memcg, iter);
1798 if (chosen)
1799 put_task_struct(chosen);
1800 return;
1801 case OOM_SCAN_OK:
1802 break;
1803 };
1804 points = oom_badness(task, memcg, NULL, totalpages);
1805 if (points > chosen_points) {
1806 if (chosen)
1807 put_task_struct(chosen);
1808 chosen = task;
1809 chosen_points = points;
1810 get_task_struct(chosen);
1811 }
1812 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001813 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001814 }
1815
1816 if (!chosen)
1817 return;
1818 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001819 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1820 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001821}
1822
Johannes Weiner56600482012-01-12 17:17:59 -08001823static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1824 gfp_t gfp_mask,
1825 unsigned long flags)
1826{
1827 unsigned long total = 0;
1828 bool noswap = false;
1829 int loop;
1830
1831 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1832 noswap = true;
1833 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1834 noswap = true;
1835
1836 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1837 if (loop)
1838 drain_all_stock_async(memcg);
1839 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1840 /*
1841 * Allow limit shrinkers, which are triggered directly
1842 * by userspace, to catch signals and stop reclaim
1843 * after minimal progress, regardless of the margin.
1844 */
1845 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1846 break;
1847 if (mem_cgroup_margin(memcg))
1848 break;
1849 /*
1850 * If nothing was reclaimed after two attempts, there
1851 * may be no reclaimable pages in this hierarchy.
1852 */
1853 if (loop && !total)
1854 break;
1855 }
1856 return total;
1857}
1858
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001859/**
1860 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001861 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001862 * @nid: the node ID to be checked.
1863 * @noswap : specify true here if the user wants flle only information.
1864 *
1865 * This function returns whether the specified memcg contains any
1866 * reclaimable pages on a node. Returns true if there are any reclaimable
1867 * pages in the node.
1868 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001869static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001870 int nid, bool noswap)
1871{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001872 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001873 return true;
1874 if (noswap || !total_swap_pages)
1875 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001876 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001877 return true;
1878 return false;
1879
1880}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001881#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001882
1883/*
1884 * Always updating the nodemask is not very good - even if we have an empty
1885 * list or the wrong list here, we can start from some node and traverse all
1886 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1887 *
1888 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001889static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001890{
1891 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001892 /*
1893 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1894 * pagein/pageout changes since the last update.
1895 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001896 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001897 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001898 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001899 return;
1900
Ying Han889976d2011-05-26 16:25:33 -07001901 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001902 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001903
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001904 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001905
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001906 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1907 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001908 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001909
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001910 atomic_set(&memcg->numainfo_events, 0);
1911 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001912}
1913
1914/*
1915 * Selecting a node where we start reclaim from. Because what we need is just
1916 * reducing usage counter, start from anywhere is O,K. Considering
1917 * memory reclaim from current node, there are pros. and cons.
1918 *
1919 * Freeing memory from current node means freeing memory from a node which
1920 * we'll use or we've used. So, it may make LRU bad. And if several threads
1921 * hit limits, it will see a contention on a node. But freeing from remote
1922 * node means more costs for memory reclaim because of memory latency.
1923 *
1924 * Now, we use round-robin. Better algorithm is welcomed.
1925 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001927{
1928 int node;
1929
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001930 mem_cgroup_may_update_nodemask(memcg);
1931 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001932
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001933 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001934 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001935 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001936 /*
1937 * We call this when we hit limit, not when pages are added to LRU.
1938 * No LRU may hold pages because all pages are UNEVICTABLE or
1939 * memcg is too small and all pages are not on LRU. In that case,
1940 * we use curret node.
1941 */
1942 if (unlikely(node == MAX_NUMNODES))
1943 node = numa_node_id();
1944
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001945 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001946 return node;
1947}
1948
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001949/*
1950 * Check all nodes whether it contains reclaimable pages or not.
1951 * For quick scan, we make use of scan_nodes. This will allow us to skip
1952 * unused nodes. But scan_nodes is lazily updated and may not cotain
1953 * enough new information. We need to do double check.
1954 */
1955static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1956{
1957 int nid;
1958
1959 /*
1960 * quick check...making use of scan_node.
1961 * We can skip unused nodes.
1962 */
1963 if (!nodes_empty(memcg->scan_nodes)) {
1964 for (nid = first_node(memcg->scan_nodes);
1965 nid < MAX_NUMNODES;
1966 nid = next_node(nid, memcg->scan_nodes)) {
1967
1968 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1969 return true;
1970 }
1971 }
1972 /*
1973 * Check rest of nodes.
1974 */
1975 for_each_node_state(nid, N_MEMORY) {
1976 if (node_isset(nid, memcg->scan_nodes))
1977 continue;
1978 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1979 return true;
1980 }
1981 return false;
1982}
1983
Ying Han889976d2011-05-26 16:25:33 -07001984#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001985int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001986{
1987 return 0;
1988}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001989
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001990static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1991{
1992 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
1993}
Ying Han889976d2011-05-26 16:25:33 -07001994#endif
1995
Andrew Morton0608f432013-09-24 15:27:41 -07001996static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1997 struct zone *zone,
1998 gfp_t gfp_mask,
1999 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002000{
Andrew Morton0608f432013-09-24 15:27:41 -07002001 struct mem_cgroup *victim = NULL;
2002 int total = 0;
2003 int loop = 0;
2004 unsigned long excess;
2005 unsigned long nr_scanned;
2006 struct mem_cgroup_reclaim_cookie reclaim = {
2007 .zone = zone,
2008 .priority = 0,
2009 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002010
Andrew Morton0608f432013-09-24 15:27:41 -07002011 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002012
Andrew Morton0608f432013-09-24 15:27:41 -07002013 while (1) {
2014 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
2015 if (!victim) {
2016 loop++;
2017 if (loop >= 2) {
2018 /*
2019 * If we have not been able to reclaim
2020 * anything, it might because there are
2021 * no reclaimable pages under this hierarchy
2022 */
2023 if (!total)
2024 break;
2025 /*
2026 * We want to do more targeted reclaim.
2027 * excess >> 2 is not to excessive so as to
2028 * reclaim too much, nor too less that we keep
2029 * coming back to reclaim from this cgroup
2030 */
2031 if (total >= (excess >> 2) ||
2032 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
2033 break;
2034 }
2035 continue;
2036 }
2037 if (!mem_cgroup_reclaimable(victim, false))
2038 continue;
2039 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2040 zone, &nr_scanned);
2041 *total_scanned += nr_scanned;
2042 if (!res_counter_soft_limit_excess(&root_memcg->res))
2043 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002044 }
Andrew Morton0608f432013-09-24 15:27:41 -07002045 mem_cgroup_iter_break(root_memcg, victim);
2046 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002047}
2048
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002049static DEFINE_SPINLOCK(memcg_oom_lock);
2050
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002051/*
2052 * Check OOM-Killer is already running under our hierarchy.
2053 * If someone is running, return false.
2054 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002055static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002056{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002057 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002058
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002059 spin_lock(&memcg_oom_lock);
2060
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002061 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002062 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002063 /*
2064 * this subtree of our hierarchy is already locked
2065 * so we cannot give a lock.
2066 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002067 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002068 mem_cgroup_iter_break(memcg, iter);
2069 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002070 } else
2071 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002072 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002073
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002074 if (failed) {
2075 /*
2076 * OK, we failed to lock the whole subtree so we have
2077 * to clean up what we set up to the failing subtree
2078 */
2079 for_each_mem_cgroup_tree(iter, memcg) {
2080 if (iter == failed) {
2081 mem_cgroup_iter_break(memcg, iter);
2082 break;
2083 }
2084 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002085 }
Michal Hocko79dfdac2011-07-26 16:08:23 -07002086 }
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002087
2088 spin_unlock(&memcg_oom_lock);
2089
2090 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002091}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002092
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002093static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002094{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002095 struct mem_cgroup *iter;
2096
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002097 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002098 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002099 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002100 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002101}
2102
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002103static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002104{
2105 struct mem_cgroup *iter;
2106
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002107 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002108 atomic_inc(&iter->under_oom);
2109}
2110
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002111static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002112{
2113 struct mem_cgroup *iter;
2114
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002115 /*
2116 * When a new child is created while the hierarchy is under oom,
2117 * mem_cgroup_oom_lock() may not be called. We have to use
2118 * atomic_add_unless() here.
2119 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002120 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002121 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002122}
2123
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002124static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2125
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002126struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002127 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002128 wait_queue_t wait;
2129};
2130
2131static int memcg_oom_wake_function(wait_queue_t *wait,
2132 unsigned mode, int sync, void *arg)
2133{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002134 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2135 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002136 struct oom_wait_info *oom_wait_info;
2137
2138 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002139 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002140
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002141 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002142 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002143 * Then we can use css_is_ancestor without taking care of RCU.
2144 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002145 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2146 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002147 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002148 return autoremove_wake_function(wait, mode, sync, arg);
2149}
2150
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002151static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002152{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002153 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002154 /* for filtering, pass "memcg" as argument. */
2155 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002156}
2157
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002158static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002159{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002160 if (memcg && atomic_read(&memcg->under_oom))
2161 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002162}
2163
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002164static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002165{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002166 if (!current->memcg_oom.may_oom)
2167 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002168 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002169 * We are in the middle of the charge context here, so we
2170 * don't want to block when potentially sitting on a callstack
2171 * that holds all kinds of filesystem and mm locks.
2172 *
2173 * Also, the caller may handle a failed allocation gracefully
2174 * (like optional page cache readahead) and so an OOM killer
2175 * invocation might not even be necessary.
2176 *
2177 * That's why we don't do anything here except remember the
2178 * OOM context and then deal with it at the end of the page
2179 * fault when the stack is unwound, the locks are released,
2180 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002181 */
Johannes Weiner49426422013-10-16 13:46:59 -07002182 css_get(&memcg->css);
2183 current->memcg_oom.memcg = memcg;
2184 current->memcg_oom.gfp_mask = mask;
2185 current->memcg_oom.order = order;
2186}
2187
2188/**
2189 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2190 * @handle: actually kill/wait or just clean up the OOM state
2191 *
2192 * This has to be called at the end of a page fault if the memcg OOM
2193 * handler was enabled.
2194 *
2195 * Memcg supports userspace OOM handling where failed allocations must
2196 * sleep on a waitqueue until the userspace task resolves the
2197 * situation. Sleeping directly in the charge context with all kinds
2198 * of locks held is not a good idea, instead we remember an OOM state
2199 * in the task and mem_cgroup_oom_synchronize() has to be called at
2200 * the end of the page fault to complete the OOM handling.
2201 *
2202 * Returns %true if an ongoing memcg OOM situation was detected and
2203 * completed, %false otherwise.
2204 */
2205bool mem_cgroup_oom_synchronize(bool handle)
2206{
2207 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2208 struct oom_wait_info owait;
2209 bool locked;
2210
2211 /* OOM is global, do not handle */
2212 if (!memcg)
2213 return false;
2214
2215 if (!handle)
2216 goto cleanup;
2217
2218 owait.memcg = memcg;
2219 owait.wait.flags = 0;
2220 owait.wait.func = memcg_oom_wake_function;
2221 owait.wait.private = current;
2222 INIT_LIST_HEAD(&owait.wait.task_list);
2223
2224 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002225 mem_cgroup_mark_under_oom(memcg);
2226
2227 locked = mem_cgroup_oom_trylock(memcg);
2228
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002229 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002230 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002231
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002232 if (locked && !memcg->oom_kill_disable) {
2233 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002234 finish_wait(&memcg_oom_waitq, &owait.wait);
2235 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2236 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002237 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002238 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002239 mem_cgroup_unmark_under_oom(memcg);
2240 finish_wait(&memcg_oom_waitq, &owait.wait);
2241 }
2242
2243 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002244 mem_cgroup_oom_unlock(memcg);
2245 /*
2246 * There is no guarantee that an OOM-lock contender
2247 * sees the wakeups triggered by the OOM kill
2248 * uncharges. Wake any sleepers explicitely.
2249 */
2250 memcg_oom_recover(memcg);
2251 }
Johannes Weiner49426422013-10-16 13:46:59 -07002252cleanup:
2253 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002254 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002255 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002256}
2257
Balbir Singhd69b0422009-06-17 16:26:34 -07002258/*
2259 * Currently used to update mapped file statistics, but the routine can be
2260 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002261 *
2262 * Notes: Race condition
2263 *
2264 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2265 * it tends to be costly. But considering some conditions, we doesn't need
2266 * to do so _always_.
2267 *
2268 * Considering "charge", lock_page_cgroup() is not required because all
2269 * file-stat operations happen after a page is attached to radix-tree. There
2270 * are no race with "charge".
2271 *
2272 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2273 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2274 * if there are race with "uncharge". Statistics itself is properly handled
2275 * by flags.
2276 *
2277 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002278 * small, we check mm->moving_account and detect there are possibility of race
2279 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002280 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002281
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002282void __mem_cgroup_begin_update_page_stat(struct page *page,
2283 bool *locked, unsigned long *flags)
2284{
2285 struct mem_cgroup *memcg;
2286 struct page_cgroup *pc;
2287
2288 pc = lookup_page_cgroup(page);
2289again:
2290 memcg = pc->mem_cgroup;
2291 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2292 return;
2293 /*
2294 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002295 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002296 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002297 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002298 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002299 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002300 return;
2301
2302 move_lock_mem_cgroup(memcg, flags);
2303 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2304 move_unlock_mem_cgroup(memcg, flags);
2305 goto again;
2306 }
2307 *locked = true;
2308}
2309
2310void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2311{
2312 struct page_cgroup *pc = lookup_page_cgroup(page);
2313
2314 /*
2315 * It's guaranteed that pc->mem_cgroup never changes while
2316 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002317 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002318 */
2319 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2320}
2321
Greg Thelen2a7106f2011-01-13 15:47:37 -08002322void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002323 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002324{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002325 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002326 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002327 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002328
Johannes Weinercfa44942012-01-12 17:18:38 -08002329 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002330 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002331
Sha Zhengju658b72c2013-09-12 15:13:52 -07002332 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002333 memcg = pc->mem_cgroup;
2334 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002335 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002336
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002337 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002338}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002339
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002340/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002341 * size of first charge trial. "32" comes from vmscan.c's magic value.
2342 * TODO: maybe necessary to use big numbers in big irons.
2343 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002344#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002345struct memcg_stock_pcp {
2346 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002347 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002348 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002349 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002350#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002351};
2352static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002353static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002354
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002355/**
2356 * consume_stock: Try to consume stocked charge on this cpu.
2357 * @memcg: memcg to consume from.
2358 * @nr_pages: how many pages to charge.
2359 *
2360 * The charges will only happen if @memcg matches the current cpu's memcg
2361 * stock, and at least @nr_pages are available in that stock. Failure to
2362 * service an allocation will refill the stock.
2363 *
2364 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002365 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002366static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002367{
2368 struct memcg_stock_pcp *stock;
2369 bool ret = true;
2370
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002371 if (nr_pages > CHARGE_BATCH)
2372 return false;
2373
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002374 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002375 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2376 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002377 else /* need to call res_counter_charge */
2378 ret = false;
2379 put_cpu_var(memcg_stock);
2380 return ret;
2381}
2382
2383/*
2384 * Returns stocks cached in percpu to res_counter and reset cached information.
2385 */
2386static void drain_stock(struct memcg_stock_pcp *stock)
2387{
2388 struct mem_cgroup *old = stock->cached;
2389
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002390 if (stock->nr_pages) {
2391 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2392
2393 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002394 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002395 res_counter_uncharge(&old->memsw, bytes);
2396 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002397 }
2398 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002399}
2400
2401/*
2402 * This must be called under preempt disabled or must be called by
2403 * a thread which is pinned to local cpu.
2404 */
2405static void drain_local_stock(struct work_struct *dummy)
2406{
2407 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2408 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002409 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002410}
2411
Michal Hockoe4777492013-02-22 16:35:40 -08002412static void __init memcg_stock_init(void)
2413{
2414 int cpu;
2415
2416 for_each_possible_cpu(cpu) {
2417 struct memcg_stock_pcp *stock =
2418 &per_cpu(memcg_stock, cpu);
2419 INIT_WORK(&stock->work, drain_local_stock);
2420 }
2421}
2422
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002423/*
2424 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002425 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002426 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002427static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002428{
2429 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2430
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002431 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002432 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002433 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002434 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002435 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002436 put_cpu_var(memcg_stock);
2437}
2438
2439/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002440 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002441 * of the hierarchy under it. sync flag says whether we should block
2442 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002443 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002444static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002445{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002446 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002447
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002448 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002449 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002450 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002451 for_each_online_cpu(cpu) {
2452 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002453 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002454
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002455 memcg = stock->cached;
2456 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002457 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002458 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002459 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002460 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2461 if (cpu == curcpu)
2462 drain_local_stock(&stock->work);
2463 else
2464 schedule_work_on(cpu, &stock->work);
2465 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002466 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002467 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002468
2469 if (!sync)
2470 goto out;
2471
2472 for_each_online_cpu(cpu) {
2473 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002474 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002475 flush_work(&stock->work);
2476 }
2477out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002478 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002479}
2480
2481/*
2482 * Tries to drain stocked charges in other cpus. This function is asynchronous
2483 * and just put a work per cpu for draining localy on each cpu. Caller can
2484 * expects some charges will be back to res_counter later but cannot wait for
2485 * it.
2486 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002487static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002488{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002489 /*
2490 * If someone calls draining, avoid adding more kworker runs.
2491 */
2492 if (!mutex_trylock(&percpu_charge_mutex))
2493 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002494 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002495 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002496}
2497
2498/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002499static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002500{
2501 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002502 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002503 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002504 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002505}
2506
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002507/*
2508 * This function drains percpu counter value from DEAD cpu and
2509 * move it to local cpu. Note that this function can be preempted.
2510 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002511static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002512{
2513 int i;
2514
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002515 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002516 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002517 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002519 per_cpu(memcg->stat->count[i], cpu) = 0;
2520 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002521 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002522 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002523 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002524
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002525 per_cpu(memcg->stat->events[i], cpu) = 0;
2526 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002527 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002528 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002529}
2530
Paul Gortmaker0db06282013-06-19 14:53:51 -04002531static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002532 unsigned long action,
2533 void *hcpu)
2534{
2535 int cpu = (unsigned long)hcpu;
2536 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002537 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002538
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002539 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002540 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002541
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002542 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002543 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002544
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002545 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002546 mem_cgroup_drain_pcp_counter(iter, cpu);
2547
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002548 stock = &per_cpu(memcg_stock, cpu);
2549 drain_stock(stock);
2550 return NOTIFY_OK;
2551}
2552
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002553
2554/* See __mem_cgroup_try_charge() for details */
2555enum {
2556 CHARGE_OK, /* success */
2557 CHARGE_RETRY, /* need to retry but retry is not bad */
2558 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2559 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002560};
2561
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002562static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002563 unsigned int nr_pages, unsigned int min_pages,
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002564 bool invoke_oom)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002565{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002566 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002567 struct mem_cgroup *mem_over_limit;
2568 struct res_counter *fail_res;
2569 unsigned long flags = 0;
2570 int ret;
2571
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002572 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002573
2574 if (likely(!ret)) {
2575 if (!do_swap_account)
2576 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002577 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002578 if (likely(!ret))
2579 return CHARGE_OK;
2580
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002581 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002582 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2583 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2584 } else
2585 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002586 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002587 * Never reclaim on behalf of optional batching, retry with a
2588 * single page instead.
2589 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002590 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002591 return CHARGE_RETRY;
2592
2593 if (!(gfp_mask & __GFP_WAIT))
2594 return CHARGE_WOULDBLOCK;
2595
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002596 if (gfp_mask & __GFP_NORETRY)
2597 return CHARGE_NOMEM;
2598
Johannes Weiner56600482012-01-12 17:17:59 -08002599 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002600 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002601 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002602 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002603 * Even though the limit is exceeded at this point, reclaim
2604 * may have been able to free some pages. Retry the charge
2605 * before killing the task.
2606 *
2607 * Only for regular pages, though: huge pages are rather
2608 * unlikely to succeed so close to the limit, and we fall back
2609 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002610 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002611 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002612 return CHARGE_RETRY;
2613
2614 /*
2615 * At task move, charge accounts can be doubly counted. So, it's
2616 * better to wait until the end of task_move if something is going on.
2617 */
2618 if (mem_cgroup_wait_acct_move(mem_over_limit))
2619 return CHARGE_RETRY;
2620
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002621 if (invoke_oom)
2622 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002623
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002624 return CHARGE_NOMEM;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002625}
2626
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002627/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002628 * __mem_cgroup_try_charge() does
2629 * 1. detect memcg to be charged against from passed *mm and *ptr,
2630 * 2. update res_counter
2631 * 3. call memory reclaim if necessary.
2632 *
2633 * In some special case, if the task is fatal, fatal_signal_pending() or
2634 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2635 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2636 * as possible without any hazards. 2: all pages should have a valid
2637 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2638 * pointer, that is treated as a charge to root_mem_cgroup.
2639 *
2640 * So __mem_cgroup_try_charge() will return
2641 * 0 ... on success, filling *ptr with a valid memcg pointer.
2642 * -ENOMEM ... charge failure because of resource limits.
2643 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2644 *
2645 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2646 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002647 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002648static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002649 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002650 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002651 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002652 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002653{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002654 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002655 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002656 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002657 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002658
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002659 /*
2660 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2661 * in system level. So, allow to go ahead dying process in addition to
2662 * MEMDIE process.
2663 */
2664 if (unlikely(test_thread_flag(TIF_MEMDIE)
2665 || fatal_signal_pending(current)))
2666 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002667
Johannes Weiner49426422013-10-16 13:46:59 -07002668 if (unlikely(task_in_memcg_oom(current)))
2669 goto bypass;
2670
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002671 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002672 * We always charge the cgroup the mm_struct belongs to.
2673 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002674 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002675 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002676 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002677 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002678 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002679again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002680 if (*ptr) { /* css should be a valid one */
2681 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002682 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002683 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002684 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002685 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002686 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002687 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002688 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002689
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002690 rcu_read_lock();
2691 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002692 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002693 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002694 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002695 * race with swapoff. Then, we have small risk of mis-accouning.
2696 * But such kind of mis-account by race always happens because
2697 * we don't have cgroup_mutex(). It's overkill and we allo that
2698 * small race, here.
2699 * (*) swapoff at el will charge against mm-struct not against
2700 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002701 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002702 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002703 if (!memcg)
2704 memcg = root_mem_cgroup;
2705 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002706 rcu_read_unlock();
2707 goto done;
2708 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002709 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002710 /*
2711 * It seems dagerous to access memcg without css_get().
2712 * But considering how consume_stok works, it's not
2713 * necessary. If consume_stock success, some charges
2714 * from this memcg are cached on this cpu. So, we
2715 * don't need to call css_get()/css_tryget() before
2716 * calling consume_stock().
2717 */
2718 rcu_read_unlock();
2719 goto done;
2720 }
2721 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002723 rcu_read_unlock();
2724 goto again;
2725 }
2726 rcu_read_unlock();
2727 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002728
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002729 do {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002730 bool invoke_oom = oom && !nr_oom_retries;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002731
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002732 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002733 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002734 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002735 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002736 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002737
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002738 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
2739 nr_pages, invoke_oom);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002740 switch (ret) {
2741 case CHARGE_OK:
2742 break;
2743 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002744 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002745 css_put(&memcg->css);
2746 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002747 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002748 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002749 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002750 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002751 case CHARGE_NOMEM: /* OOM routine works */
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002752 if (!oom || invoke_oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002753 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002754 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002755 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002756 nr_oom_retries--;
2757 break;
Balbir Singh66e17072008-02-07 00:13:56 -08002758 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002759 } while (ret != CHARGE_OK);
2760
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002761 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002762 refill_stock(memcg, batch - nr_pages);
2763 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002764done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002765 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002766 return 0;
2767nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002768 *ptr = NULL;
Johannes Weiner84235de2013-10-16 13:47:00 -07002769 if (gfp_mask & __GFP_NOFAIL)
2770 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002771 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002772bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002773 *ptr = root_mem_cgroup;
2774 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002775}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002776
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002777/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002778 * Somemtimes we have to undo a charge we got by try_charge().
2779 * This function is for that and do uncharge, put css's refcnt.
2780 * gotten by try_charge().
2781 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002782static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002783 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002784{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002785 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002786 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002787
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002788 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002789 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002790 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002791 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002792}
2793
2794/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002795 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2796 * This is useful when moving usage to parent cgroup.
2797 */
2798static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2799 unsigned int nr_pages)
2800{
2801 unsigned long bytes = nr_pages * PAGE_SIZE;
2802
2803 if (mem_cgroup_is_root(memcg))
2804 return;
2805
2806 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2807 if (do_swap_account)
2808 res_counter_uncharge_until(&memcg->memsw,
2809 memcg->memsw.parent, bytes);
2810}
2811
2812/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002813 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002814 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2815 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2816 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002817 */
2818static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2819{
2820 struct cgroup_subsys_state *css;
2821
2822 /* ID 0 is unused ID */
2823 if (!id)
2824 return NULL;
2825 css = css_lookup(&mem_cgroup_subsys, id);
2826 if (!css)
2827 return NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -07002828 return mem_cgroup_from_css(css);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002829}
2830
Wu Fengguange42d9d52009-12-16 12:19:59 +01002831struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002832{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002833 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002834 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002835 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002836 swp_entry_t ent;
2837
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002838 VM_BUG_ON(!PageLocked(page));
2839
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002840 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002841 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002842 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002843 memcg = pc->mem_cgroup;
2844 if (memcg && !css_tryget(&memcg->css))
2845 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002846 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002847 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002848 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002849 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002850 memcg = mem_cgroup_lookup(id);
2851 if (memcg && !css_tryget(&memcg->css))
2852 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002853 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002854 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002855 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002856 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002857}
2858
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002859static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002860 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002861 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002862 enum charge_type ctype,
2863 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002864{
Johannes Weinerce587e62012-04-24 20:22:33 +02002865 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002866 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002867 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002868 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002869 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002870
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002871 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002872 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002873 /*
2874 * we don't need page_cgroup_lock about tail pages, becase they are not
2875 * accessed by any other context at this point.
2876 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002877
2878 /*
2879 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2880 * may already be on some other mem_cgroup's LRU. Take care of it.
2881 */
2882 if (lrucare) {
2883 zone = page_zone(page);
2884 spin_lock_irq(&zone->lru_lock);
2885 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002886 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002887 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002888 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002889 was_on_lru = true;
2890 }
2891 }
2892
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002893 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002894 /*
2895 * We access a page_cgroup asynchronously without lock_page_cgroup().
2896 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2897 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2898 * before USED bit, we need memory barrier here.
2899 * See mem_cgroup_add_lru_list(), etc.
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002900 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002901 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002902 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002903
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002904 if (lrucare) {
2905 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002906 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002907 VM_BUG_ON(PageLRU(page));
2908 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002909 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002910 }
2911 spin_unlock_irq(&zone->lru_lock);
2912 }
2913
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002914 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002915 anon = true;
2916 else
2917 anon = false;
2918
David Rientjesb070e652013-05-07 16:18:09 -07002919 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002920 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002921
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002922 /*
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002923 * "charge_statistics" updated event counter. Then, check it.
2924 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2925 * if they exceeds softlimit.
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002926 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002927 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002928}
2929
Glauber Costa7cf27982012-12-18 14:22:55 -08002930static DEFINE_MUTEX(set_limit_mutex);
2931
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002932#ifdef CONFIG_MEMCG_KMEM
2933static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2934{
2935 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
2936 (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK);
2937}
2938
Glauber Costa1f458cb2012-12-18 14:22:50 -08002939/*
2940 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2941 * in the memcg_cache_params struct.
2942 */
2943static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2944{
2945 struct kmem_cache *cachep;
2946
2947 VM_BUG_ON(p->is_root_cache);
2948 cachep = p->root_cache;
2949 return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)];
2950}
2951
Glauber Costa749c5412012-12-18 14:23:01 -08002952#ifdef CONFIG_SLABINFO
Tejun Heo182446d2013-08-08 20:11:24 -04002953static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css,
2954 struct cftype *cft, struct seq_file *m)
Glauber Costa749c5412012-12-18 14:23:01 -08002955{
Tejun Heo182446d2013-08-08 20:11:24 -04002956 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa749c5412012-12-18 14:23:01 -08002957 struct memcg_cache_params *params;
2958
2959 if (!memcg_can_account_kmem(memcg))
2960 return -EIO;
2961
2962 print_slabinfo_header(m);
2963
2964 mutex_lock(&memcg->slab_caches_mutex);
2965 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2966 cache_show(memcg_params_to_cache(params), m);
2967 mutex_unlock(&memcg->slab_caches_mutex);
2968
2969 return 0;
2970}
2971#endif
2972
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002973static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
2974{
2975 struct res_counter *fail_res;
2976 struct mem_cgroup *_memcg;
2977 int ret = 0;
2978 bool may_oom;
2979
2980 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
2981 if (ret)
2982 return ret;
2983
2984 /*
2985 * Conditions under which we can wait for the oom_killer. Those are
2986 * the same conditions tested by the core page allocator
2987 */
2988 may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY);
2989
2990 _memcg = memcg;
2991 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
2992 &_memcg, may_oom);
2993
2994 if (ret == -EINTR) {
2995 /*
2996 * __mem_cgroup_try_charge() chosed to bypass to root due to
2997 * OOM kill or fatal signal. Since our only options are to
2998 * either fail the allocation or charge it to this cgroup, do
2999 * it as a temporary condition. But we can't fail. From a
3000 * kmem/slab perspective, the cache has already been selected,
3001 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3002 * our minds.
3003 *
3004 * This condition will only trigger if the task entered
3005 * memcg_charge_kmem in a sane state, but was OOM-killed during
3006 * __mem_cgroup_try_charge() above. Tasks that were already
3007 * dying when the allocation triggers should have been already
3008 * directed to the root cgroup in memcontrol.h
3009 */
3010 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3011 if (do_swap_account)
3012 res_counter_charge_nofail(&memcg->memsw, size,
3013 &fail_res);
3014 ret = 0;
3015 } else if (ret)
3016 res_counter_uncharge(&memcg->kmem, size);
3017
3018 return ret;
3019}
3020
3021static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3022{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003023 res_counter_uncharge(&memcg->res, size);
3024 if (do_swap_account)
3025 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003026
3027 /* Not down to 0 */
3028 if (res_counter_uncharge(&memcg->kmem, size))
3029 return;
3030
Li Zefan10d5ebf2013-07-08 16:00:33 -07003031 /*
3032 * Releases a reference taken in kmem_cgroup_css_offline in case
3033 * this last uncharge is racing with the offlining code or it is
3034 * outliving the memcg existence.
3035 *
3036 * The memory barrier imposed by test&clear is paired with the
3037 * explicit one in memcg_kmem_mark_dead().
3038 */
Glauber Costa7de37682012-12-18 14:22:07 -08003039 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003040 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003041}
3042
Glauber Costa2633d7a2012-12-18 14:22:34 -08003043void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
3044{
3045 if (!memcg)
3046 return;
3047
3048 mutex_lock(&memcg->slab_caches_mutex);
3049 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
3050 mutex_unlock(&memcg->slab_caches_mutex);
3051}
3052
3053/*
3054 * helper for acessing a memcg's index. It will be used as an index in the
3055 * child cache array in kmem_cache, and also to derive its name. This function
3056 * will return -1 when this is not a kmem-limited memcg.
3057 */
3058int memcg_cache_id(struct mem_cgroup *memcg)
3059{
3060 return memcg ? memcg->kmemcg_id : -1;
3061}
3062
Glauber Costa55007d82012-12-18 14:22:38 -08003063/*
3064 * This ends up being protected by the set_limit mutex, during normal
3065 * operation, because that is its main call site.
3066 *
3067 * But when we create a new cache, we can call this as well if its parent
3068 * is kmem-limited. That will have to hold set_limit_mutex as well.
3069 */
3070int memcg_update_cache_sizes(struct mem_cgroup *memcg)
3071{
3072 int num, ret;
3073
3074 num = ida_simple_get(&kmem_limited_groups,
3075 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
3076 if (num < 0)
3077 return num;
3078 /*
3079 * After this point, kmem_accounted (that we test atomically in
3080 * the beginning of this conditional), is no longer 0. This
3081 * guarantees only one process will set the following boolean
3082 * to true. We don't need test_and_set because we're protected
3083 * by the set_limit_mutex anyway.
3084 */
3085 memcg_kmem_set_activated(memcg);
3086
3087 ret = memcg_update_all_caches(num+1);
3088 if (ret) {
3089 ida_simple_remove(&kmem_limited_groups, num);
3090 memcg_kmem_clear_activated(memcg);
3091 return ret;
3092 }
3093
3094 memcg->kmemcg_id = num;
3095 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
3096 mutex_init(&memcg->slab_caches_mutex);
3097 return 0;
3098}
3099
3100static size_t memcg_caches_array_size(int num_groups)
3101{
3102 ssize_t size;
3103 if (num_groups <= 0)
3104 return 0;
3105
3106 size = 2 * num_groups;
3107 if (size < MEMCG_CACHES_MIN_SIZE)
3108 size = MEMCG_CACHES_MIN_SIZE;
3109 else if (size > MEMCG_CACHES_MAX_SIZE)
3110 size = MEMCG_CACHES_MAX_SIZE;
3111
3112 return size;
3113}
3114
3115/*
3116 * We should update the current array size iff all caches updates succeed. This
3117 * can only be done from the slab side. The slab mutex needs to be held when
3118 * calling this.
3119 */
3120void memcg_update_array_size(int num)
3121{
3122 if (num > memcg_limited_groups_array_size)
3123 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3124}
3125
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003126static void kmem_cache_destroy_work_func(struct work_struct *w);
3127
Glauber Costa55007d82012-12-18 14:22:38 -08003128int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3129{
3130 struct memcg_cache_params *cur_params = s->memcg_params;
3131
3132 VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache);
3133
3134 if (num_groups > memcg_limited_groups_array_size) {
3135 int i;
3136 ssize_t size = memcg_caches_array_size(num_groups);
3137
3138 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003139 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003140
3141 s->memcg_params = kzalloc(size, GFP_KERNEL);
3142 if (!s->memcg_params) {
3143 s->memcg_params = cur_params;
3144 return -ENOMEM;
3145 }
3146
3147 s->memcg_params->is_root_cache = true;
3148
3149 /*
3150 * There is the chance it will be bigger than
3151 * memcg_limited_groups_array_size, if we failed an allocation
3152 * in a cache, in which case all caches updated before it, will
3153 * have a bigger array.
3154 *
3155 * But if that is the case, the data after
3156 * memcg_limited_groups_array_size is certainly unused
3157 */
3158 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3159 if (!cur_params->memcg_caches[i])
3160 continue;
3161 s->memcg_params->memcg_caches[i] =
3162 cur_params->memcg_caches[i];
3163 }
3164
3165 /*
3166 * Ideally, we would wait until all caches succeed, and only
3167 * then free the old one. But this is not worth the extra
3168 * pointer per-cache we'd have to have for this.
3169 *
3170 * It is not a big deal if some caches are left with a size
3171 * bigger than the others. And all updates will reset this
3172 * anyway.
3173 */
3174 kfree(cur_params);
3175 }
3176 return 0;
3177}
3178
Glauber Costa943a4512012-12-18 14:23:03 -08003179int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
3180 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003181{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003182 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003183
3184 if (!memcg_kmem_enabled())
3185 return 0;
3186
Andrey Vagin90c7a792013-09-11 14:22:18 -07003187 if (!memcg) {
3188 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003189 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003190 } else
3191 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003192
Glauber Costa2633d7a2012-12-18 14:22:34 -08003193 s->memcg_params = kzalloc(size, GFP_KERNEL);
3194 if (!s->memcg_params)
3195 return -ENOMEM;
3196
Glauber Costa943a4512012-12-18 14:23:03 -08003197 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003198 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003199 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003200 INIT_WORK(&s->memcg_params->destroy,
3201 kmem_cache_destroy_work_func);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003202 } else
3203 s->memcg_params->is_root_cache = true;
3204
Glauber Costa2633d7a2012-12-18 14:22:34 -08003205 return 0;
3206}
3207
3208void memcg_release_cache(struct kmem_cache *s)
3209{
Glauber Costad7f25f82012-12-18 14:22:40 -08003210 struct kmem_cache *root;
3211 struct mem_cgroup *memcg;
3212 int id;
3213
3214 /*
3215 * This happens, for instance, when a root cache goes away before we
3216 * add any memcg.
3217 */
3218 if (!s->memcg_params)
3219 return;
3220
3221 if (s->memcg_params->is_root_cache)
3222 goto out;
3223
3224 memcg = s->memcg_params->memcg;
3225 id = memcg_cache_id(memcg);
3226
3227 root = s->memcg_params->root_cache;
3228 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08003229
3230 mutex_lock(&memcg->slab_caches_mutex);
3231 list_del(&s->memcg_params->list);
3232 mutex_unlock(&memcg->slab_caches_mutex);
3233
Li Zefan20f05312013-07-08 16:00:31 -07003234 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003235out:
Glauber Costa2633d7a2012-12-18 14:22:34 -08003236 kfree(s->memcg_params);
3237}
3238
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003239/*
3240 * During the creation a new cache, we need to disable our accounting mechanism
3241 * altogether. This is true even if we are not creating, but rather just
3242 * enqueing new caches to be created.
3243 *
3244 * This is because that process will trigger allocations; some visible, like
3245 * explicit kmallocs to auxiliary data structures, name strings and internal
3246 * cache structures; some well concealed, like INIT_WORK() that can allocate
3247 * objects during debug.
3248 *
3249 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3250 * to it. This may not be a bounded recursion: since the first cache creation
3251 * failed to complete (waiting on the allocation), we'll just try to create the
3252 * cache again, failing at the same point.
3253 *
3254 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3255 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3256 * inside the following two functions.
3257 */
3258static inline void memcg_stop_kmem_account(void)
3259{
3260 VM_BUG_ON(!current->mm);
3261 current->memcg_kmem_skip_account++;
3262}
3263
3264static inline void memcg_resume_kmem_account(void)
3265{
3266 VM_BUG_ON(!current->mm);
3267 current->memcg_kmem_skip_account--;
3268}
3269
Glauber Costa1f458cb2012-12-18 14:22:50 -08003270static void kmem_cache_destroy_work_func(struct work_struct *w)
3271{
3272 struct kmem_cache *cachep;
3273 struct memcg_cache_params *p;
3274
3275 p = container_of(w, struct memcg_cache_params, destroy);
3276
3277 cachep = memcg_params_to_cache(p);
3278
Glauber Costa22933152012-12-18 14:22:59 -08003279 /*
3280 * If we get down to 0 after shrink, we could delete right away.
3281 * However, memcg_release_pages() already puts us back in the workqueue
3282 * in that case. If we proceed deleting, we'll get a dangling
3283 * reference, and removing the object from the workqueue in that case
3284 * is unnecessary complication. We are not a fast path.
3285 *
3286 * Note that this case is fundamentally different from racing with
3287 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3288 * kmem_cache_shrink, not only we would be reinserting a dead cache
3289 * into the queue, but doing so from inside the worker racing to
3290 * destroy it.
3291 *
3292 * So if we aren't down to zero, we'll just schedule a worker and try
3293 * again
3294 */
3295 if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
3296 kmem_cache_shrink(cachep);
3297 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
3298 return;
3299 } else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003300 kmem_cache_destroy(cachep);
3301}
3302
3303void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3304{
3305 if (!cachep->memcg_params->dead)
3306 return;
3307
3308 /*
Glauber Costa22933152012-12-18 14:22:59 -08003309 * There are many ways in which we can get here.
3310 *
3311 * We can get to a memory-pressure situation while the delayed work is
3312 * still pending to run. The vmscan shrinkers can then release all
3313 * cache memory and get us to destruction. If this is the case, we'll
3314 * be executed twice, which is a bug (the second time will execute over
3315 * bogus data). In this case, cancelling the work should be fine.
3316 *
3317 * But we can also get here from the worker itself, if
3318 * kmem_cache_shrink is enough to shake all the remaining objects and
3319 * get the page count to 0. In this case, we'll deadlock if we try to
3320 * cancel the work (the worker runs with an internal lock held, which
3321 * is the same lock we would hold for cancel_work_sync().)
3322 *
3323 * Since we can't possibly know who got us here, just refrain from
3324 * running if there is already work pending
3325 */
3326 if (work_pending(&cachep->memcg_params->destroy))
3327 return;
3328 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003329 * We have to defer the actual destroying to a workqueue, because
3330 * we might currently be in a context that cannot sleep.
3331 */
3332 schedule_work(&cachep->memcg_params->destroy);
3333}
3334
Glauber Costad7f25f82012-12-18 14:22:40 -08003335/*
3336 * This lock protects updaters, not readers. We want readers to be as fast as
3337 * they can, and they will either see NULL or a valid cache value. Our model
3338 * allow them to see NULL, in which case the root memcg will be selected.
3339 *
3340 * We need this lock because multiple allocations to the same cache from a non
3341 * will span more than one worker. Only one of them can create the cache.
3342 */
3343static DEFINE_MUTEX(memcg_cache_mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003344
3345/*
3346 * Called with memcg_cache_mutex held
3347 */
3348static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
3349 struct kmem_cache *s)
3350{
3351 struct kmem_cache *new;
3352 static char *tmp_name = NULL;
3353
3354 lockdep_assert_held(&memcg_cache_mutex);
3355
3356 /*
3357 * kmem_cache_create_memcg duplicates the given name and
3358 * cgroup_name for this name requires RCU context.
3359 * This static temporary buffer is used to prevent from
3360 * pointless shortliving allocation.
3361 */
3362 if (!tmp_name) {
3363 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3364 if (!tmp_name)
3365 return NULL;
3366 }
3367
3368 rcu_read_lock();
3369 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3370 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3371 rcu_read_unlock();
3372
3373 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3374 (s->flags & ~SLAB_PANIC), s->ctor, s);
3375
3376 if (new)
3377 new->allocflags |= __GFP_KMEMCG;
3378
3379 return new;
3380}
3381
Glauber Costad7f25f82012-12-18 14:22:40 -08003382static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3383 struct kmem_cache *cachep)
3384{
3385 struct kmem_cache *new_cachep;
3386 int idx;
3387
3388 BUG_ON(!memcg_can_account_kmem(memcg));
3389
3390 idx = memcg_cache_id(memcg);
3391
3392 mutex_lock(&memcg_cache_mutex);
3393 new_cachep = cachep->memcg_params->memcg_caches[idx];
Li Zefan20f05312013-07-08 16:00:31 -07003394 if (new_cachep) {
3395 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003396 goto out;
Li Zefan20f05312013-07-08 16:00:31 -07003397 }
Glauber Costad7f25f82012-12-18 14:22:40 -08003398
3399 new_cachep = kmem_cache_dup(memcg, cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003400 if (new_cachep == NULL) {
3401 new_cachep = cachep;
Li Zefan20f05312013-07-08 16:00:31 -07003402 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003403 goto out;
3404 }
3405
Glauber Costa1f458cb2012-12-18 14:22:50 -08003406 atomic_set(&new_cachep->memcg_params->nr_pages , 0);
Glauber Costad7f25f82012-12-18 14:22:40 -08003407
3408 cachep->memcg_params->memcg_caches[idx] = new_cachep;
3409 /*
3410 * the readers won't lock, make sure everybody sees the updated value,
3411 * so they won't put stuff in the queue again for no reason
3412 */
3413 wmb();
3414out:
3415 mutex_unlock(&memcg_cache_mutex);
3416 return new_cachep;
3417}
3418
Glauber Costa7cf27982012-12-18 14:22:55 -08003419void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3420{
3421 struct kmem_cache *c;
3422 int i;
3423
3424 if (!s->memcg_params)
3425 return;
3426 if (!s->memcg_params->is_root_cache)
3427 return;
3428
3429 /*
3430 * If the cache is being destroyed, we trust that there is no one else
3431 * requesting objects from it. Even if there are, the sanity checks in
3432 * kmem_cache_destroy should caught this ill-case.
3433 *
3434 * Still, we don't want anyone else freeing memcg_caches under our
3435 * noses, which can happen if a new memcg comes to life. As usual,
3436 * we'll take the set_limit_mutex to protect ourselves against this.
3437 */
3438 mutex_lock(&set_limit_mutex);
3439 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3440 c = s->memcg_params->memcg_caches[i];
3441 if (!c)
3442 continue;
3443
3444 /*
3445 * We will now manually delete the caches, so to avoid races
3446 * we need to cancel all pending destruction workers and
3447 * proceed with destruction ourselves.
3448 *
3449 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3450 * and that could spawn the workers again: it is likely that
3451 * the cache still have active pages until this very moment.
3452 * This would lead us back to mem_cgroup_destroy_cache.
3453 *
3454 * But that will not execute at all if the "dead" flag is not
3455 * set, so flip it down to guarantee we are in control.
3456 */
3457 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003458 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003459 kmem_cache_destroy(c);
3460 }
3461 mutex_unlock(&set_limit_mutex);
3462}
3463
Glauber Costad7f25f82012-12-18 14:22:40 -08003464struct create_work {
3465 struct mem_cgroup *memcg;
3466 struct kmem_cache *cachep;
3467 struct work_struct work;
3468};
3469
Glauber Costa1f458cb2012-12-18 14:22:50 -08003470static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3471{
3472 struct kmem_cache *cachep;
3473 struct memcg_cache_params *params;
3474
3475 if (!memcg_kmem_is_active(memcg))
3476 return;
3477
3478 mutex_lock(&memcg->slab_caches_mutex);
3479 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3480 cachep = memcg_params_to_cache(params);
3481 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003482 schedule_work(&cachep->memcg_params->destroy);
3483 }
3484 mutex_unlock(&memcg->slab_caches_mutex);
3485}
3486
Glauber Costad7f25f82012-12-18 14:22:40 -08003487static void memcg_create_cache_work_func(struct work_struct *w)
3488{
3489 struct create_work *cw;
3490
3491 cw = container_of(w, struct create_work, work);
3492 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003493 kfree(cw);
3494}
3495
3496/*
3497 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003498 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003499static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3500 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003501{
3502 struct create_work *cw;
3503
3504 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003505 if (cw == NULL) {
3506 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003507 return;
3508 }
3509
3510 cw->memcg = memcg;
3511 cw->cachep = cachep;
3512
3513 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3514 schedule_work(&cw->work);
3515}
3516
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003517static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3518 struct kmem_cache *cachep)
3519{
3520 /*
3521 * We need to stop accounting when we kmalloc, because if the
3522 * corresponding kmalloc cache is not yet created, the first allocation
3523 * in __memcg_create_cache_enqueue will recurse.
3524 *
3525 * However, it is better to enclose the whole function. Depending on
3526 * the debugging options enabled, INIT_WORK(), for instance, can
3527 * trigger an allocation. This too, will make us recurse. Because at
3528 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3529 * the safest choice is to do it like this, wrapping the whole function.
3530 */
3531 memcg_stop_kmem_account();
3532 __memcg_create_cache_enqueue(memcg, cachep);
3533 memcg_resume_kmem_account();
3534}
Glauber Costad7f25f82012-12-18 14:22:40 -08003535/*
3536 * Return the kmem_cache we're supposed to use for a slab allocation.
3537 * We try to use the current memcg's version of the cache.
3538 *
3539 * If the cache does not exist yet, if we are the first user of it,
3540 * we either create it immediately, if possible, or create it asynchronously
3541 * in a workqueue.
3542 * In the latter case, we will let the current allocation go through with
3543 * the original cache.
3544 *
3545 * Can't be called in interrupt context or from kernel threads.
3546 * This function needs to be called with rcu_read_lock() held.
3547 */
3548struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3549 gfp_t gfp)
3550{
3551 struct mem_cgroup *memcg;
3552 int idx;
3553
3554 VM_BUG_ON(!cachep->memcg_params);
3555 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3556
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003557 if (!current->mm || current->memcg_kmem_skip_account)
3558 return cachep;
3559
Glauber Costad7f25f82012-12-18 14:22:40 -08003560 rcu_read_lock();
3561 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003562
3563 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003564 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003565
3566 idx = memcg_cache_id(memcg);
3567
3568 /*
3569 * barrier to mare sure we're always seeing the up to date value. The
3570 * code updating memcg_caches will issue a write barrier to match this.
3571 */
3572 read_barrier_depends();
Li Zefanca0dde92013-04-29 15:08:57 -07003573 if (likely(cachep->memcg_params->memcg_caches[idx])) {
3574 cachep = cachep->memcg_params->memcg_caches[idx];
3575 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003576 }
3577
Li Zefanca0dde92013-04-29 15:08:57 -07003578 /* The corresponding put will be done in the workqueue. */
3579 if (!css_tryget(&memcg->css))
3580 goto out;
3581 rcu_read_unlock();
3582
3583 /*
3584 * If we are in a safe context (can wait, and not in interrupt
3585 * context), we could be be predictable and return right away.
3586 * This would guarantee that the allocation being performed
3587 * already belongs in the new cache.
3588 *
3589 * However, there are some clashes that can arrive from locking.
3590 * For instance, because we acquire the slab_mutex while doing
3591 * kmem_cache_dup, this means no further allocation could happen
3592 * with the slab_mutex held.
3593 *
3594 * Also, because cache creation issue get_online_cpus(), this
3595 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3596 * that ends up reversed during cpu hotplug. (cpuset allocates
3597 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3598 * better to defer everything.
3599 */
3600 memcg_create_cache_enqueue(memcg, cachep);
3601 return cachep;
3602out:
3603 rcu_read_unlock();
3604 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003605}
3606EXPORT_SYMBOL(__memcg_kmem_get_cache);
3607
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003608/*
3609 * We need to verify if the allocation against current->mm->owner's memcg is
3610 * possible for the given order. But the page is not allocated yet, so we'll
3611 * need a further commit step to do the final arrangements.
3612 *
3613 * It is possible for the task to switch cgroups in this mean time, so at
3614 * commit time, we can't rely on task conversion any longer. We'll then use
3615 * the handle argument to return to the caller which cgroup we should commit
3616 * against. We could also return the memcg directly and avoid the pointer
3617 * passing, but a boolean return value gives better semantics considering
3618 * the compiled-out case as well.
3619 *
3620 * Returning true means the allocation is possible.
3621 */
3622bool
3623__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3624{
3625 struct mem_cgroup *memcg;
3626 int ret;
3627
3628 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003629
3630 /*
3631 * Disabling accounting is only relevant for some specific memcg
3632 * internal allocations. Therefore we would initially not have such
3633 * check here, since direct calls to the page allocator that are marked
3634 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3635 * concerned with cache allocations, and by having this test at
3636 * memcg_kmem_get_cache, we are already able to relay the allocation to
3637 * the root cache and bypass the memcg cache altogether.
3638 *
3639 * There is one exception, though: the SLUB allocator does not create
3640 * large order caches, but rather service large kmallocs directly from
3641 * the page allocator. Therefore, the following sequence when backed by
3642 * the SLUB allocator:
3643 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003644 * memcg_stop_kmem_account();
3645 * kmalloc(<large_number>)
3646 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003647 *
3648 * would effectively ignore the fact that we should skip accounting,
3649 * since it will drive us directly to this function without passing
3650 * through the cache selector memcg_kmem_get_cache. Such large
3651 * allocations are extremely rare but can happen, for instance, for the
3652 * cache arrays. We bring this test here.
3653 */
3654 if (!current->mm || current->memcg_kmem_skip_account)
3655 return true;
3656
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003657 memcg = try_get_mem_cgroup_from_mm(current->mm);
3658
3659 /*
3660 * very rare case described in mem_cgroup_from_task. Unfortunately there
3661 * isn't much we can do without complicating this too much, and it would
3662 * be gfp-dependent anyway. Just let it go
3663 */
3664 if (unlikely(!memcg))
3665 return true;
3666
3667 if (!memcg_can_account_kmem(memcg)) {
3668 css_put(&memcg->css);
3669 return true;
3670 }
3671
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003672 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3673 if (!ret)
3674 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003675
3676 css_put(&memcg->css);
3677 return (ret == 0);
3678}
3679
3680void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3681 int order)
3682{
3683 struct page_cgroup *pc;
3684
3685 VM_BUG_ON(mem_cgroup_is_root(memcg));
3686
3687 /* The page allocation failed. Revert */
3688 if (!page) {
3689 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003690 return;
3691 }
3692
3693 pc = lookup_page_cgroup(page);
3694 lock_page_cgroup(pc);
3695 pc->mem_cgroup = memcg;
3696 SetPageCgroupUsed(pc);
3697 unlock_page_cgroup(pc);
3698}
3699
3700void __memcg_kmem_uncharge_pages(struct page *page, int order)
3701{
3702 struct mem_cgroup *memcg = NULL;
3703 struct page_cgroup *pc;
3704
3705
3706 pc = lookup_page_cgroup(page);
3707 /*
3708 * Fast unlocked return. Theoretically might have changed, have to
3709 * check again after locking.
3710 */
3711 if (!PageCgroupUsed(pc))
3712 return;
3713
3714 lock_page_cgroup(pc);
3715 if (PageCgroupUsed(pc)) {
3716 memcg = pc->mem_cgroup;
3717 ClearPageCgroupUsed(pc);
3718 }
3719 unlock_page_cgroup(pc);
3720
3721 /*
3722 * We trust that only if there is a memcg associated with the page, it
3723 * is a valid allocation
3724 */
3725 if (!memcg)
3726 return;
3727
3728 VM_BUG_ON(mem_cgroup_is_root(memcg));
3729 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003730}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003731#else
3732static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3733{
3734}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003735#endif /* CONFIG_MEMCG_KMEM */
3736
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003737#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3738
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003739#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003740/*
3741 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003742 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3743 * charge/uncharge will be never happen and move_account() is done under
3744 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003745 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003746void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003747{
3748 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003749 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003750 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003751 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003752
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003753 if (mem_cgroup_disabled())
3754 return;
David Rientjesb070e652013-05-07 16:18:09 -07003755
3756 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003757 for (i = 1; i < HPAGE_PMD_NR; i++) {
3758 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003759 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003760 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003761 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3762 }
David Rientjesb070e652013-05-07 16:18:09 -07003763 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3764 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003765}
Hugh Dickins12d27102012-01-12 17:19:52 -08003766#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003767
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003768static inline
3769void mem_cgroup_move_account_page_stat(struct mem_cgroup *from,
3770 struct mem_cgroup *to,
3771 unsigned int nr_pages,
3772 enum mem_cgroup_stat_index idx)
3773{
3774 /* Update stat data for mem_cgroup */
3775 preempt_disable();
3776 WARN_ON_ONCE(from->stat->count[idx] < nr_pages);
3777 __this_cpu_add(from->stat->count[idx], -nr_pages);
3778 __this_cpu_add(to->stat->count[idx], nr_pages);
3779 preempt_enable();
3780}
3781
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003782/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003783 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003784 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003785 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003786 * @pc: page_cgroup of the page.
3787 * @from: mem_cgroup which the page is moved from.
3788 * @to: mem_cgroup which the page is moved to. @from != @to.
3789 *
3790 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003791 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003792 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003793 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003794 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3795 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003796 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003797static int mem_cgroup_move_account(struct page *page,
3798 unsigned int nr_pages,
3799 struct page_cgroup *pc,
3800 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003801 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003802{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003803 unsigned long flags;
3804 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003805 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003806
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003807 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07003808 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07003809 /*
3810 * The page is isolated from LRU. So, collapse function
3811 * will not handle this page. But page splitting can happen.
3812 * Do this check under compound_page_lock(). The caller should
3813 * hold it.
3814 */
3815 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003816 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003817 goto out;
3818
3819 lock_page_cgroup(pc);
3820
3821 ret = -EINVAL;
3822 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3823 goto unlock;
3824
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003825 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003826
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003827 if (!anon && page_mapped(page))
3828 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3829 MEM_CGROUP_STAT_FILE_MAPPED);
3830
3831 if (PageWriteback(page))
3832 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3833 MEM_CGROUP_STAT_WRITEBACK);
3834
David Rientjesb070e652013-05-07 16:18:09 -07003835 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003836
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003837 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003838 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003839 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003840 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003841 ret = 0;
3842unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003843 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003844 /*
3845 * check events
3846 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003847 memcg_check_events(to, page);
3848 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003849out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003850 return ret;
3851}
3852
Michal Hocko2ef37d32012-10-26 13:37:30 +02003853/**
3854 * mem_cgroup_move_parent - moves page to the parent group
3855 * @page: the page to move
3856 * @pc: page_cgroup of the page
3857 * @child: page's cgroup
3858 *
3859 * move charges to its parent or the root cgroup if the group has no
3860 * parent (aka use_hierarchy==0).
3861 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3862 * mem_cgroup_move_account fails) the failure is always temporary and
3863 * it signals a race with a page removal/uncharge or migration. In the
3864 * first case the page is on the way out and it will vanish from the LRU
3865 * on the next attempt and the call should be retried later.
3866 * Isolation from the LRU fails only if page has been isolated from
3867 * the LRU since we looked at it and that usually means either global
3868 * reclaim or migration going on. The page will either get back to the
3869 * LRU or vanish.
3870 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3871 * (!PageCgroupUsed) or moved to a different group. The page will
3872 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003873 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003874static int mem_cgroup_move_parent(struct page *page,
3875 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003876 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003877{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003878 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003879 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003880 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003881 int ret;
3882
Michal Hockod8423012012-10-26 13:37:29 +02003883 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003884
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003885 ret = -EBUSY;
3886 if (!get_page_unless_zero(page))
3887 goto out;
3888 if (isolate_lru_page(page))
3889 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003890
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003891 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003892
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003893 parent = parent_mem_cgroup(child);
3894 /*
3895 * If no parent, move charges to root cgroup.
3896 */
3897 if (!parent)
3898 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003899
Michal Hocko2ef37d32012-10-26 13:37:30 +02003900 if (nr_pages > 1) {
3901 VM_BUG_ON(!PageTransHuge(page));
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003902 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003903 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003904
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003905 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003906 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003907 if (!ret)
3908 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003909
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003910 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003911 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003912 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003913put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003914 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003915out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003916 return ret;
3917}
3918
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003919/*
3920 * Charge the memory controller for page usage.
3921 * Return
3922 * 0 if the charge was successful
3923 * < 0 if the cgroup is over its limit
3924 */
3925static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003926 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003927{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003928 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003929 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003930 bool oom = true;
3931 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003932
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003933 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003934 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003935 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08003936 /*
3937 * Never OOM-kill a process for a huge page. The
3938 * fault handler will fall back to regular pages.
3939 */
3940 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003941 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003942
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003943 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003944 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003945 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003946 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003947 return 0;
3948}
3949
3950int mem_cgroup_newpage_charge(struct page *page,
3951 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003952{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003953 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003954 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003955 VM_BUG_ON(page_mapped(page));
3956 VM_BUG_ON(page->mapping && !PageAnon(page));
3957 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003958 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003959 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003960}
3961
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003962/*
3963 * While swap-in, try_charge -> commit or cancel, the page is locked.
3964 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003965 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003966 * "commit()" or removed by "cancel()"
3967 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003968static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3969 struct page *page,
3970 gfp_t mask,
3971 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003972{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003973 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003974 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003975 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003976
Johannes Weiner90deb782012-07-31 16:45:47 -07003977 pc = lookup_page_cgroup(page);
3978 /*
3979 * Every swap fault against a single page tries to charge the
3980 * page, bail as early as possible. shmem_unuse() encounters
3981 * already charged pages, too. The USED bit is protected by
3982 * the page lock, which serializes swap cache removal, which
3983 * in turn serializes uncharging.
3984 */
3985 if (PageCgroupUsed(pc))
3986 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003987 if (!do_swap_account)
3988 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003989 memcg = try_get_mem_cgroup_from_page(page);
3990 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003991 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08003992 *memcgp = memcg;
3993 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003994 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003995 if (ret == -EINTR)
3996 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003997 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003998charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003999 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
4000 if (ret == -EINTR)
4001 ret = 0;
4002 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004003}
4004
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004005int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
4006 gfp_t gfp_mask, struct mem_cgroup **memcgp)
4007{
4008 *memcgp = NULL;
4009 if (mem_cgroup_disabled())
4010 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07004011 /*
4012 * A racing thread's fault, or swapoff, may have already
4013 * updated the pte, and even removed page from swap cache: in
4014 * those cases unuse_pte()'s pte_same() test will fail; but
4015 * there's also a KSM case which does need to charge the page.
4016 */
4017 if (!PageSwapCache(page)) {
4018 int ret;
4019
4020 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
4021 if (ret == -EINTR)
4022 ret = 0;
4023 return ret;
4024 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004025 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
4026}
4027
Johannes Weiner827a03d2012-07-31 16:45:36 -07004028void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
4029{
4030 if (mem_cgroup_disabled())
4031 return;
4032 if (!memcg)
4033 return;
4034 __mem_cgroup_cancel_charge(memcg, 1);
4035}
4036
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004037static void
Johannes Weiner72835c82012-01-12 17:18:32 -08004038__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004039 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004040{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004041 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004042 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08004043 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004044 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07004045
Johannes Weinerce587e62012-04-24 20:22:33 +02004046 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004047 /*
4048 * Now swap is on-memory. This means this page may be
4049 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004050 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
4051 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
4052 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004053 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004054 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004055 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004056 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004057 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004058}
4059
Johannes Weiner72835c82012-01-12 17:18:32 -08004060void mem_cgroup_commit_charge_swapin(struct page *page,
4061 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004062{
Johannes Weiner72835c82012-01-12 17:18:32 -08004063 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004064 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004065}
4066
Johannes Weiner827a03d2012-07-31 16:45:36 -07004067int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4068 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004069{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004070 struct mem_cgroup *memcg = NULL;
4071 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4072 int ret;
4073
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004074 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004075 return 0;
4076 if (PageCompound(page))
4077 return 0;
4078
Johannes Weiner827a03d2012-07-31 16:45:36 -07004079 if (!PageSwapCache(page))
4080 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4081 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004082 ret = __mem_cgroup_try_charge_swapin(mm, page,
4083 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004084 if (!ret)
4085 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4086 }
4087 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004088}
4089
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004090static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004091 unsigned int nr_pages,
4092 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004093{
4094 struct memcg_batch_info *batch = NULL;
4095 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004096
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004097 /* If swapout, usage of swap doesn't decrease */
4098 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4099 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004100
4101 batch = &current->memcg_batch;
4102 /*
4103 * In usual, we do css_get() when we remember memcg pointer.
4104 * But in this case, we keep res->usage until end of a series of
4105 * uncharges. Then, it's ok to ignore memcg's refcnt.
4106 */
4107 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004108 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004109 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004110 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004111 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004112 * the same cgroup and we have chance to coalesce uncharges.
4113 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4114 * because we want to do uncharge as soon as possible.
4115 */
4116
4117 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4118 goto direct_uncharge;
4119
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004120 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004121 goto direct_uncharge;
4122
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004123 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004124 * In typical case, batch->memcg == mem. This means we can
4125 * merge a series of uncharges to an uncharge of res_counter.
4126 * If not, we uncharge res_counter ony by one.
4127 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004128 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004129 goto direct_uncharge;
4130 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004131 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004132 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004133 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004134 return;
4135direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004136 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004137 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004138 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4139 if (unlikely(batch->memcg != memcg))
4140 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004141}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004142
Balbir Singh8697d332008-02-07 00:13:59 -08004143/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004144 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004145 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004146static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004147__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4148 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004149{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004150 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004151 unsigned int nr_pages = 1;
4152 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004153 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004154
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004155 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004156 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004157
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004158 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004159 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004160 VM_BUG_ON(!PageTransHuge(page));
4161 }
Balbir Singh8697d332008-02-07 00:13:59 -08004162 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004163 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004164 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004165 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004166 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004167 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004168
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004169 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004170
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004171 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004172
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004173 if (!PageCgroupUsed(pc))
4174 goto unlock_out;
4175
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004176 anon = PageAnon(page);
4177
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004178 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004179 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004180 /*
4181 * Generally PageAnon tells if it's the anon statistics to be
4182 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4183 * used before page reached the stage of being marked PageAnon.
4184 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004185 anon = true;
4186 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004187 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004188 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004189 if (page_mapped(page))
4190 goto unlock_out;
4191 /*
4192 * Pages under migration may not be uncharged. But
4193 * end_migration() /must/ be the one uncharging the
4194 * unused post-migration page and so it has to call
4195 * here with the migration bit still set. See the
4196 * res_counter handling below.
4197 */
4198 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004199 goto unlock_out;
4200 break;
4201 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4202 if (!PageAnon(page)) { /* Shared memory */
4203 if (page->mapping && !page_is_file_cache(page))
4204 goto unlock_out;
4205 } else if (page_mapped(page)) /* Anon */
4206 goto unlock_out;
4207 break;
4208 default:
4209 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004210 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004211
David Rientjesb070e652013-05-07 16:18:09 -07004212 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004213
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004214 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004215 /*
4216 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4217 * freed from LRU. This is safe because uncharged page is expected not
4218 * to be reused (freed soon). Exception is SwapCache, it's handled by
4219 * special functions.
4220 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004221
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004222 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004223 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004224 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004225 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004226 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004227 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004228 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004229 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004230 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004231 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004232 /*
4233 * Migration does not charge the res_counter for the
4234 * replacement page, so leave it alone when phasing out the
4235 * page that is unused after the migration.
4236 */
4237 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004238 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004239
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004240 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004241
4242unlock_out:
4243 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004244 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004245}
4246
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004247void mem_cgroup_uncharge_page(struct page *page)
4248{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004249 /* early check. */
4250 if (page_mapped(page))
4251 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08004252 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004253 /*
4254 * If the page is in swap cache, uncharge should be deferred
4255 * to the swap path, which also properly accounts swap usage
4256 * and handles memcg lifetime.
4257 *
4258 * Note that this check is not stable and reclaim may add the
4259 * page to swap cache at any time after this. However, if the
4260 * page is not in swap cache by the time page->mapcount hits
4261 * 0, there won't be any page table references to the swap
4262 * slot, and reclaim will free it and not actually write the
4263 * page to disk.
4264 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004265 if (PageSwapCache(page))
4266 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004267 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004268}
4269
4270void mem_cgroup_uncharge_cache_page(struct page *page)
4271{
4272 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07004273 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07004274 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004275}
4276
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004277/*
4278 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4279 * In that cases, pages are freed continuously and we can expect pages
4280 * are in the same memcg. All these calls itself limits the number of
4281 * pages freed at once, then uncharge_start/end() is called properly.
4282 * This may be called prural(2) times in a context,
4283 */
4284
4285void mem_cgroup_uncharge_start(void)
4286{
4287 current->memcg_batch.do_batch++;
4288 /* We can do nest. */
4289 if (current->memcg_batch.do_batch == 1) {
4290 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004291 current->memcg_batch.nr_pages = 0;
4292 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004293 }
4294}
4295
4296void mem_cgroup_uncharge_end(void)
4297{
4298 struct memcg_batch_info *batch = &current->memcg_batch;
4299
4300 if (!batch->do_batch)
4301 return;
4302
4303 batch->do_batch--;
4304 if (batch->do_batch) /* If stacked, do nothing. */
4305 return;
4306
4307 if (!batch->memcg)
4308 return;
4309 /*
4310 * This "batch->memcg" is valid without any css_get/put etc...
4311 * bacause we hide charges behind us.
4312 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004313 if (batch->nr_pages)
4314 res_counter_uncharge(&batch->memcg->res,
4315 batch->nr_pages * PAGE_SIZE);
4316 if (batch->memsw_nr_pages)
4317 res_counter_uncharge(&batch->memcg->memsw,
4318 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004319 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004320 /* forget this pointer (for sanity check) */
4321 batch->memcg = NULL;
4322}
4323
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004324#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004325/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004326 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004327 * memcg information is recorded to swap_cgroup of "ent"
4328 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004329void
4330mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004331{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004332 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004333 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004334
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004335 if (!swapout) /* this was a swap cache but the swap is unused ! */
4336 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4337
Johannes Weiner0030f532012-07-31 16:45:25 -07004338 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004339
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004340 /*
4341 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004342 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004343 */
4344 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004345 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004346}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004347#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004348
Andrew Mortonc255a452012-07-31 16:43:02 -07004349#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004350/*
4351 * called from swap_entry_free(). remove record in swap_cgroup and
4352 * uncharge "memsw" account.
4353 */
4354void mem_cgroup_uncharge_swap(swp_entry_t ent)
4355{
4356 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004357 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004358
4359 if (!do_swap_account)
4360 return;
4361
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004362 id = swap_cgroup_record(ent, 0);
4363 rcu_read_lock();
4364 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004365 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004366 /*
4367 * We uncharge this because swap is freed.
4368 * This memcg can be obsolete one. We avoid calling css_tryget
4369 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004370 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004371 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004372 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004373 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004374 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004375 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004376}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004377
4378/**
4379 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4380 * @entry: swap entry to be moved
4381 * @from: mem_cgroup which the entry is moved from
4382 * @to: mem_cgroup which the entry is moved to
4383 *
4384 * It succeeds only when the swap_cgroup's record for this entry is the same
4385 * as the mem_cgroup's id of @from.
4386 *
4387 * Returns 0 on success, -EINVAL on failure.
4388 *
4389 * The caller must have charged to @to, IOW, called res_counter_charge() about
4390 * both res and memsw, and called css_get().
4391 */
4392static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004393 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004394{
4395 unsigned short old_id, new_id;
4396
4397 old_id = css_id(&from->css);
4398 new_id = css_id(&to->css);
4399
4400 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004401 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004402 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004403 /*
4404 * This function is only called from task migration context now.
4405 * It postpones res_counter and refcount handling till the end
4406 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004407 * improvement. But we cannot postpone css_get(to) because if
4408 * the process that has been moved to @to does swap-in, the
4409 * refcount of @to might be decreased to 0.
4410 *
4411 * We are in attach() phase, so the cgroup is guaranteed to be
4412 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004413 */
Li Zefan40503772013-07-08 16:00:34 -07004414 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004415 return 0;
4416 }
4417 return -EINVAL;
4418}
4419#else
4420static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004421 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004422{
4423 return -EINVAL;
4424}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004425#endif
4426
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004427/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004428 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4429 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004430 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004431void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4432 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004433{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004434 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004435 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004436 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004437 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004438
Johannes Weiner72835c82012-01-12 17:18:32 -08004439 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004440
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004441 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004442 return;
Balbir Singh40779602008-04-04 14:29:59 -07004443
Mel Gormanb32967f2012-11-19 12:35:47 +00004444 if (PageTransHuge(page))
4445 nr_pages <<= compound_order(page);
4446
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004447 pc = lookup_page_cgroup(page);
4448 lock_page_cgroup(pc);
4449 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004450 memcg = pc->mem_cgroup;
4451 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004452 /*
4453 * At migrating an anonymous page, its mapcount goes down
4454 * to 0 and uncharge() will be called. But, even if it's fully
4455 * unmapped, migration may fail and this page has to be
4456 * charged again. We set MIGRATION flag here and delay uncharge
4457 * until end_migration() is called
4458 *
4459 * Corner Case Thinking
4460 * A)
4461 * When the old page was mapped as Anon and it's unmap-and-freed
4462 * while migration was ongoing.
4463 * If unmap finds the old page, uncharge() of it will be delayed
4464 * until end_migration(). If unmap finds a new page, it's
4465 * uncharged when it make mapcount to be 1->0. If unmap code
4466 * finds swap_migration_entry, the new page will not be mapped
4467 * and end_migration() will find it(mapcount==0).
4468 *
4469 * B)
4470 * When the old page was mapped but migraion fails, the kernel
4471 * remaps it. A charge for it is kept by MIGRATION flag even
4472 * if mapcount goes down to 0. We can do remap successfully
4473 * without charging it again.
4474 *
4475 * C)
4476 * The "old" page is under lock_page() until the end of
4477 * migration, so, the old page itself will not be swapped-out.
4478 * If the new page is swapped out before end_migraton, our
4479 * hook to usual swap-out path will catch the event.
4480 */
4481 if (PageAnon(page))
4482 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004483 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004484 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004485 /*
4486 * If the page is not charged at this point,
4487 * we return here.
4488 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004489 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004490 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004491
Johannes Weiner72835c82012-01-12 17:18:32 -08004492 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004493 /*
4494 * We charge new page before it's used/mapped. So, even if unlock_page()
4495 * is called before end_migration, we can catch all events on this new
4496 * page. In the case new page is migrated but not remapped, new page's
4497 * mapcount will be finally 0 and we call uncharge in end_migration().
4498 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004499 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004500 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004501 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004502 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004503 /*
4504 * The page is committed to the memcg, but it's not actually
4505 * charged to the res_counter since we plan on replacing the
4506 * old one and only one page is going to be left afterwards.
4507 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004508 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004509}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004510
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004511/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004512void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004513 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004514{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004515 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004516 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004517 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004519 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004520 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004521
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004522 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004523 used = oldpage;
4524 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004525 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004526 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004527 unused = oldpage;
4528 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004529 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004530 __mem_cgroup_uncharge_common(unused,
4531 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4532 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4533 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004534 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004535 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004536 * We disallowed uncharge of pages under migration because mapcount
4537 * of the page goes down to zero, temporarly.
4538 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004539 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004540 pc = lookup_page_cgroup(oldpage);
4541 lock_page_cgroup(pc);
4542 ClearPageCgroupMigration(pc);
4543 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004544
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004545 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004546 * If a page is a file cache, radix-tree replacement is very atomic
4547 * and we can skip this check. When it was an Anon page, its mapcount
4548 * goes down to 0. But because we added MIGRATION flage, it's not
4549 * uncharged yet. There are several case but page->mapcount check
4550 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4551 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004552 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004553 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004554 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004555}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004556
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004557/*
4558 * At replace page cache, newpage is not under any memcg but it's on
4559 * LRU. So, this function doesn't touch res_counter but handles LRU
4560 * in correct way. Both pages are locked so we cannot race with uncharge.
4561 */
4562void mem_cgroup_replace_page_cache(struct page *oldpage,
4563 struct page *newpage)
4564{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004565 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004566 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004567 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004568
4569 if (mem_cgroup_disabled())
4570 return;
4571
4572 pc = lookup_page_cgroup(oldpage);
4573 /* fix accounting on old pages */
4574 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004575 if (PageCgroupUsed(pc)) {
4576 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004577 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004578 ClearPageCgroupUsed(pc);
4579 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004580 unlock_page_cgroup(pc);
4581
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004582 /*
4583 * When called from shmem_replace_page(), in some cases the
4584 * oldpage has already been charged, and in some cases not.
4585 */
4586 if (!memcg)
4587 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004588 /*
4589 * Even if newpage->mapping was NULL before starting replacement,
4590 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4591 * LRU while we overwrite pc->mem_cgroup.
4592 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004593 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004594}
4595
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004596#ifdef CONFIG_DEBUG_VM
4597static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4598{
4599 struct page_cgroup *pc;
4600
4601 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004602 /*
4603 * Can be NULL while feeding pages into the page allocator for
4604 * the first time, i.e. during boot or memory hotplug;
4605 * or when mem_cgroup_disabled().
4606 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004607 if (likely(pc) && PageCgroupUsed(pc))
4608 return pc;
4609 return NULL;
4610}
4611
4612bool mem_cgroup_bad_page_check(struct page *page)
4613{
4614 if (mem_cgroup_disabled())
4615 return false;
4616
4617 return lookup_page_cgroup_used(page) != NULL;
4618}
4619
4620void mem_cgroup_print_bad_page(struct page *page)
4621{
4622 struct page_cgroup *pc;
4623
4624 pc = lookup_page_cgroup_used(page);
4625 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004626 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4627 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004628 }
4629}
4630#endif
4631
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004632static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004633 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004634{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004635 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004636 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004637 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004638 int children = mem_cgroup_count_children(memcg);
4639 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004640 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004641
4642 /*
4643 * For keeping hierarchical_reclaim simple, how long we should retry
4644 * is depends on callers. We set our retry-count to be function
4645 * of # of children which we should visit in this loop.
4646 */
4647 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4648
4649 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004650
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004651 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004652 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004653 if (signal_pending(current)) {
4654 ret = -EINTR;
4655 break;
4656 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004657 /*
4658 * Rather than hide all in some function, I do this in
4659 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004660 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004661 */
4662 mutex_lock(&set_limit_mutex);
4663 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4664 if (memswlimit < val) {
4665 ret = -EINVAL;
4666 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004667 break;
4668 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004669
4670 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4671 if (memlimit < val)
4672 enlarge = 1;
4673
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004674 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004675 if (!ret) {
4676 if (memswlimit == val)
4677 memcg->memsw_is_minimum = true;
4678 else
4679 memcg->memsw_is_minimum = false;
4680 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004681 mutex_unlock(&set_limit_mutex);
4682
4683 if (!ret)
4684 break;
4685
Johannes Weiner56600482012-01-12 17:17:59 -08004686 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4687 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004688 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4689 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004690 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004691 retry_count--;
4692 else
4693 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004694 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004695 if (!ret && enlarge)
4696 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004697
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004698 return ret;
4699}
4700
Li Zefan338c8432009-06-17 16:27:15 -07004701static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4702 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004703{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004704 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004705 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004706 int children = mem_cgroup_count_children(memcg);
4707 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004708 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004709
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004710 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004711 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004712 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004713 while (retry_count) {
4714 if (signal_pending(current)) {
4715 ret = -EINTR;
4716 break;
4717 }
4718 /*
4719 * Rather than hide all in some function, I do this in
4720 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004721 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004722 */
4723 mutex_lock(&set_limit_mutex);
4724 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4725 if (memlimit > val) {
4726 ret = -EINVAL;
4727 mutex_unlock(&set_limit_mutex);
4728 break;
4729 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004730 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4731 if (memswlimit < val)
4732 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004733 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004734 if (!ret) {
4735 if (memlimit == val)
4736 memcg->memsw_is_minimum = true;
4737 else
4738 memcg->memsw_is_minimum = false;
4739 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004740 mutex_unlock(&set_limit_mutex);
4741
4742 if (!ret)
4743 break;
4744
Johannes Weiner56600482012-01-12 17:17:59 -08004745 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4746 MEM_CGROUP_RECLAIM_NOSWAP |
4747 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004748 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004749 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004750 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004751 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004752 else
4753 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004754 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004755 if (!ret && enlarge)
4756 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004757 return ret;
4758}
4759
Andrew Morton0608f432013-09-24 15:27:41 -07004760unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
4761 gfp_t gfp_mask,
4762 unsigned long *total_scanned)
4763{
4764 unsigned long nr_reclaimed = 0;
4765 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4766 unsigned long reclaimed;
4767 int loop = 0;
4768 struct mem_cgroup_tree_per_zone *mctz;
4769 unsigned long long excess;
4770 unsigned long nr_scanned;
4771
4772 if (order > 0)
4773 return 0;
4774
4775 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
4776 /*
4777 * This loop can run a while, specially if mem_cgroup's continuously
4778 * keep exceeding their soft limit and putting the system under
4779 * pressure
4780 */
4781 do {
4782 if (next_mz)
4783 mz = next_mz;
4784 else
4785 mz = mem_cgroup_largest_soft_limit_node(mctz);
4786 if (!mz)
4787 break;
4788
4789 nr_scanned = 0;
4790 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
4791 gfp_mask, &nr_scanned);
4792 nr_reclaimed += reclaimed;
4793 *total_scanned += nr_scanned;
4794 spin_lock(&mctz->lock);
4795
4796 /*
4797 * If we failed to reclaim anything from this memory cgroup
4798 * it is time to move on to the next cgroup
4799 */
4800 next_mz = NULL;
4801 if (!reclaimed) {
4802 do {
4803 /*
4804 * Loop until we find yet another one.
4805 *
4806 * By the time we get the soft_limit lock
4807 * again, someone might have aded the
4808 * group back on the RB tree. Iterate to
4809 * make sure we get a different mem.
4810 * mem_cgroup_largest_soft_limit_node returns
4811 * NULL if no other cgroup is present on
4812 * the tree
4813 */
4814 next_mz =
4815 __mem_cgroup_largest_soft_limit_node(mctz);
4816 if (next_mz == mz)
4817 css_put(&next_mz->memcg->css);
4818 else /* next_mz == NULL or other memcg */
4819 break;
4820 } while (1);
4821 }
4822 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4823 excess = res_counter_soft_limit_excess(&mz->memcg->res);
4824 /*
4825 * One school of thought says that we should not add
4826 * back the node to the tree if reclaim returns 0.
4827 * But our reclaim could return 0, simply because due
4828 * to priority we are exposing a smaller subset of
4829 * memory to reclaim from. Consider this as a longer
4830 * term TODO.
4831 */
4832 /* If excess == 0, no tree ops */
4833 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
4834 spin_unlock(&mctz->lock);
4835 css_put(&mz->memcg->css);
4836 loop++;
4837 /*
4838 * Could not reclaim anything and there are no more
4839 * mem cgroups to try or we seem to be looping without
4840 * reclaiming anything.
4841 */
4842 if (!nr_reclaimed &&
4843 (next_mz == NULL ||
4844 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4845 break;
4846 } while (!nr_reclaimed);
4847 if (next_mz)
4848 css_put(&next_mz->memcg->css);
4849 return nr_reclaimed;
4850}
4851
Michal Hocko2ef37d32012-10-26 13:37:30 +02004852/**
4853 * mem_cgroup_force_empty_list - clears LRU of a group
4854 * @memcg: group to clear
4855 * @node: NUMA node
4856 * @zid: zone id
4857 * @lru: lru to to clear
4858 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004859 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004860 * reclaim the pages page themselves - pages are moved to the parent (or root)
4861 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004862 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004863static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004864 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004865{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004866 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004867 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004868 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004869 struct page *busy;
4870 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004871
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004872 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004873 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4874 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004875
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004876 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004877 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004878 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004879 struct page *page;
4880
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004881 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004882 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004883 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004884 break;
4885 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004886 page = list_entry(list->prev, struct page, lru);
4887 if (busy == page) {
4888 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004889 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004890 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004891 continue;
4892 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004893 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004894
Johannes Weiner925b7672012-01-12 17:18:15 -08004895 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004896
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004897 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004898 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004899 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004900 cond_resched();
4901 } else
4902 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004903 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004904}
4905
4906/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004907 * make mem_cgroup's charge to be 0 if there is no task by moving
4908 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004909 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004910 *
4911 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004912 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004913static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004914{
Michal Hockoc26251f2012-10-26 13:37:28 +02004915 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004916 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004917
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004918 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004919 /* This is for making all *used* pages to be on LRU. */
4920 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004921 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004922 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004923 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004924 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004925 enum lru_list lru;
4926 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004927 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004928 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004929 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004930 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004931 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004932 mem_cgroup_end_move(memcg);
4933 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004934 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004935
Michal Hocko2ef37d32012-10-26 13:37:30 +02004936 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004937 * Kernel memory may not necessarily be trackable to a specific
4938 * process. So they are not migrated, and therefore we can't
4939 * expect their value to drop to 0 here.
4940 * Having res filled up with kmem only is enough.
4941 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004942 * This is a safety check because mem_cgroup_force_empty_list
4943 * could have raced with mem_cgroup_replace_page_cache callers
4944 * so the lru seemed empty but the page could have been added
4945 * right after the check. RES_USAGE should be safe as we always
4946 * charge before adding to the LRU.
4947 */
Glauber Costabea207c2012-12-18 14:22:11 -08004948 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4949 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4950 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004951}
4952
4953/*
Glauber Costab5f99b52013-02-22 16:34:53 -08004954 * This mainly exists for tests during the setting of set of use_hierarchy.
4955 * Since this is the very setting we are changing, the current hierarchy value
4956 * is meaningless
4957 */
4958static inline bool __memcg_has_children(struct mem_cgroup *memcg)
4959{
Tejun Heo492eb212013-08-08 20:11:25 -04004960 struct cgroup_subsys_state *pos;
Glauber Costab5f99b52013-02-22 16:34:53 -08004961
4962 /* bounce at first found */
Tejun Heo492eb212013-08-08 20:11:25 -04004963 css_for_each_child(pos, &memcg->css)
Glauber Costab5f99b52013-02-22 16:34:53 -08004964 return true;
4965 return false;
4966}
4967
4968/*
Glauber Costa09998212013-02-22 16:34:55 -08004969 * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed
4970 * to be already dead (as in mem_cgroup_force_empty, for instance). This is
Glauber Costab5f99b52013-02-22 16:34:53 -08004971 * from mem_cgroup_count_children(), in the sense that we don't really care how
4972 * many children we have; we only need to know if we have any. It also counts
4973 * any memcg without hierarchy as infertile.
4974 */
4975static inline bool memcg_has_children(struct mem_cgroup *memcg)
4976{
4977 return memcg->use_hierarchy && __memcg_has_children(memcg);
4978}
4979
4980/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004981 * Reclaims as many pages from the given memcg as possible and moves
4982 * the rest to the parent.
4983 *
4984 * Caller is responsible for holding css reference for memcg.
4985 */
4986static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4987{
4988 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4989 struct cgroup *cgrp = memcg->css.cgroup;
4990
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004991 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004992 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4993 return -EBUSY;
4994
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004995 /* we call try-to-free pages for make this cgroup empty */
4996 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004997 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07004998 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004999 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005000
Michal Hockoc26251f2012-10-26 13:37:28 +02005001 if (signal_pending(current))
5002 return -EINTR;
5003
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005004 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07005005 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005006 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005007 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005008 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02005009 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005010 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005011
5012 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08005013 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02005014 mem_cgroup_reparent_charges(memcg);
5015
5016 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08005017}
5018
Tejun Heo182446d2013-08-08 20:11:24 -04005019static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
5020 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005021{
Tejun Heo182446d2013-08-08 20:11:24 -04005022 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02005023
Michal Hockod8423012012-10-26 13:37:29 +02005024 if (mem_cgroup_is_root(memcg))
5025 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07005026 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005027}
5028
Tejun Heo182446d2013-08-08 20:11:24 -04005029static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
5030 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005031{
Tejun Heo182446d2013-08-08 20:11:24 -04005032 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005033}
5034
Tejun Heo182446d2013-08-08 20:11:24 -04005035static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
5036 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005037{
5038 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04005039 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005040 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08005041
Glauber Costa09998212013-02-22 16:34:55 -08005042 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07005043
5044 if (memcg->use_hierarchy == val)
5045 goto out;
5046
Balbir Singh18f59ea2009-01-07 18:08:07 -08005047 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005048 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08005049 * in the child subtrees. If it is unset, then the change can
5050 * occur, provided the current cgroup has no children.
5051 *
5052 * For the root cgroup, parent_mem is NULL, we allow value to be
5053 * set if there are no children.
5054 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005055 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08005056 (val == 1 || val == 0)) {
Glauber Costab5f99b52013-02-22 16:34:53 -08005057 if (!__memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005058 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005059 else
5060 retval = -EBUSY;
5061 } else
5062 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005063
5064out:
Glauber Costa09998212013-02-22 16:34:55 -08005065 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005066
5067 return retval;
5068}
5069
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005070
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005071static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005072 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005073{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005074 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005075 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005076
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005077 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005078 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005079 val += mem_cgroup_read_stat(iter, idx);
5080
5081 if (val < 0) /* race ? */
5082 val = 0;
5083 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005084}
5085
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005086static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005087{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005088 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005089
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005090 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005091 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005092 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005093 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005094 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005095 }
5096
David Rientjesb070e652013-05-07 16:18:09 -07005097 /*
5098 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5099 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5100 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005101 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5102 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005103
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005104 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005105 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005106
5107 return val << PAGE_SHIFT;
5108}
5109
Tejun Heo182446d2013-08-08 20:11:24 -04005110static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css,
5111 struct cftype *cft, struct file *file,
5112 char __user *buf, size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005113{
Tejun Heo182446d2013-08-08 20:11:24 -04005114 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heoaf36f902012-04-01 12:09:55 -07005115 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005116 u64 val;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005117 int name, len;
5118 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005119
5120 type = MEMFILE_TYPE(cft->private);
5121 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005122
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005123 switch (type) {
5124 case _MEM:
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, false);
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->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005129 break;
5130 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005131 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005132 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005133 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005134 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005135 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005136 case _KMEM:
5137 val = res_counter_read_u64(&memcg->kmem, name);
5138 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005139 default:
5140 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005141 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005142
5143 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
5144 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005145}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005146
Tejun Heo182446d2013-08-08 20:11:24 -04005147static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005148{
5149 int ret = -EINVAL;
5150#ifdef CONFIG_MEMCG_KMEM
Tejun Heo182446d2013-08-08 20:11:24 -04005151 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005152 /*
5153 * For simplicity, we won't allow this to be disabled. It also can't
5154 * be changed if the cgroup has children already, or if tasks had
5155 * already joined.
5156 *
5157 * If tasks join before we set the limit, a person looking at
5158 * kmem.usage_in_bytes will have no way to determine when it took
5159 * place, which makes the value quite meaningless.
5160 *
5161 * After it first became limited, changes in the value of the limit are
5162 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005163 */
Glauber Costa09998212013-02-22 16:34:55 -08005164 mutex_lock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005165 mutex_lock(&set_limit_mutex);
Sha Zhengju6de5a8b2013-09-12 15:13:47 -07005166 if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) {
Tejun Heo182446d2013-08-08 20:11:24 -04005167 if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005168 ret = -EBUSY;
5169 goto out;
5170 }
5171 ret = res_counter_set_limit(&memcg->kmem, val);
5172 VM_BUG_ON(ret);
5173
Glauber Costa55007d82012-12-18 14:22:38 -08005174 ret = memcg_update_cache_sizes(memcg);
5175 if (ret) {
Sha Zhengju6de5a8b2013-09-12 15:13:47 -07005176 res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX);
Glauber Costa55007d82012-12-18 14:22:38 -08005177 goto out;
5178 }
Glauber Costa692e89a2013-02-22 16:34:56 -08005179 static_key_slow_inc(&memcg_kmem_enabled_key);
5180 /*
5181 * setting the active bit after the inc will guarantee no one
5182 * starts accounting before all call sites are patched
5183 */
5184 memcg_kmem_set_active(memcg);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005185 } else
5186 ret = res_counter_set_limit(&memcg->kmem, val);
5187out:
5188 mutex_unlock(&set_limit_mutex);
Glauber Costa09998212013-02-22 16:34:55 -08005189 mutex_unlock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005190#endif
5191 return ret;
5192}
5193
Hugh Dickins6d0439902013-02-22 16:35:50 -08005194#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -08005195static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005196{
Glauber Costa55007d82012-12-18 14:22:38 -08005197 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005198 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5199 if (!parent)
Glauber Costa55007d82012-12-18 14:22:38 -08005200 goto out;
5201
Glauber Costa510fc4e2012-12-18 14:21:47 -08005202 memcg->kmem_account_flags = parent->kmem_account_flags;
Glauber Costaa8964b92012-12-18 14:22:09 -08005203 /*
5204 * When that happen, we need to disable the static branch only on those
5205 * memcgs that enabled it. To achieve this, we would be forced to
5206 * complicate the code by keeping track of which memcgs were the ones
5207 * that actually enabled limits, and which ones got it from its
5208 * parents.
5209 *
5210 * It is a lot simpler just to do static_key_slow_inc() on every child
5211 * that is accounted.
5212 */
Glauber Costa55007d82012-12-18 14:22:38 -08005213 if (!memcg_kmem_is_active(memcg))
5214 goto out;
5215
5216 /*
Li Zefan10d5ebf2013-07-08 16:00:33 -07005217 * __mem_cgroup_free() will issue static_key_slow_dec() because this
5218 * memcg is active already. If the later initialization fails then the
5219 * cgroup core triggers the cleanup so we do not have to do it here.
Glauber Costa55007d82012-12-18 14:22:38 -08005220 */
Glauber Costa55007d82012-12-18 14:22:38 -08005221 static_key_slow_inc(&memcg_kmem_enabled_key);
5222
5223 mutex_lock(&set_limit_mutex);
Glauber Costa425c5982013-07-08 16:00:01 -07005224 memcg_stop_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005225 ret = memcg_update_cache_sizes(memcg);
Glauber Costa425c5982013-07-08 16:00:01 -07005226 memcg_resume_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005227 mutex_unlock(&set_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005228out:
5229 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005230}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005231#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005232
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005233/*
5234 * The user of this function is...
5235 * RES_LIMIT.
5236 */
Tejun Heo182446d2013-08-08 20:11:24 -04005237static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Paul Menage856c13a2008-07-25 01:47:04 -07005238 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005239{
Tejun Heo182446d2013-08-08 20:11:24 -04005240 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005241 enum res_type type;
5242 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005243 unsigned long long val;
5244 int ret;
5245
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005246 type = MEMFILE_TYPE(cft->private);
5247 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005248
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005249 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005250 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005251 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5252 ret = -EINVAL;
5253 break;
5254 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005255 /* This function does all necessary parse...reuse it */
5256 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005257 if (ret)
5258 break;
5259 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005260 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005261 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005262 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005263 else if (type == _KMEM)
Tejun Heo182446d2013-08-08 20:11:24 -04005264 ret = memcg_update_kmem_limit(css, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005265 else
5266 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005267 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005268 case RES_SOFT_LIMIT:
5269 ret = res_counter_memparse_write_strategy(buffer, &val);
5270 if (ret)
5271 break;
5272 /*
5273 * For memsw, soft limits are hard to implement in terms
5274 * of semantics, for now, we support soft limits for
5275 * control without swap
5276 */
5277 if (type == _MEM)
5278 ret = res_counter_set_soft_limit(&memcg->res, val);
5279 else
5280 ret = -EINVAL;
5281 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005282 default:
5283 ret = -EINVAL; /* should be BUG() ? */
5284 break;
5285 }
5286 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005287}
5288
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005289static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5290 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5291{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005292 unsigned long long min_limit, min_memsw_limit, tmp;
5293
5294 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5295 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005296 if (!memcg->use_hierarchy)
5297 goto out;
5298
Tejun Heo63876982013-08-08 20:11:23 -04005299 while (css_parent(&memcg->css)) {
5300 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005301 if (!memcg->use_hierarchy)
5302 break;
5303 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5304 min_limit = min(min_limit, tmp);
5305 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5306 min_memsw_limit = min(min_memsw_limit, tmp);
5307 }
5308out:
5309 *mem_limit = min_limit;
5310 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005311}
5312
Tejun Heo182446d2013-08-08 20:11:24 -04005313static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005314{
Tejun Heo182446d2013-08-08 20:11:24 -04005315 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005316 int name;
5317 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005318
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005319 type = MEMFILE_TYPE(event);
5320 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005321
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005322 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005323 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005324 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005325 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005326 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005327 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005328 else if (type == _KMEM)
5329 res_counter_reset_max(&memcg->kmem);
5330 else
5331 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005332 break;
5333 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005334 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005335 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005336 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005337 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005338 else if (type == _KMEM)
5339 res_counter_reset_failcnt(&memcg->kmem);
5340 else
5341 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005342 break;
5343 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005344
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005345 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005346}
5347
Tejun Heo182446d2013-08-08 20:11:24 -04005348static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005349 struct cftype *cft)
5350{
Tejun Heo182446d2013-08-08 20:11:24 -04005351 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005352}
5353
Daisuke Nishimura02491442010-03-10 15:22:17 -08005354#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005355static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005356 struct cftype *cft, u64 val)
5357{
Tejun Heo182446d2013-08-08 20:11:24 -04005358 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005359
5360 if (val >= (1 << NR_MOVE_TYPE))
5361 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005362
Glauber Costaee5e8472013-02-22 16:34:50 -08005363 /*
5364 * No kind of locking is needed in here, because ->can_attach() will
5365 * check this value once in the beginning of the process, and then carry
5366 * on with stale data. This means that changes to this value will only
5367 * affect task migrations starting after the change.
5368 */
5369 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005370 return 0;
5371}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005372#else
Tejun Heo182446d2013-08-08 20:11:24 -04005373static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005374 struct cftype *cft, u64 val)
5375{
5376 return -ENOSYS;
5377}
5378#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005379
Ying Han406eb0c2011-05-26 16:25:37 -07005380#ifdef CONFIG_NUMA
Tejun Heo182446d2013-08-08 20:11:24 -04005381static int memcg_numa_stat_show(struct cgroup_subsys_state *css,
5382 struct cftype *cft, struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07005383{
5384 int nid;
5385 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
5386 unsigned long node_nr;
Tejun Heo182446d2013-08-08 20:11:24 -04005387 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Ying Han406eb0c2011-05-26 16:25:37 -07005388
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005389 total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005390 seq_printf(m, "total=%lu", total_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005391 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005392 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07005393 seq_printf(m, " N%d=%lu", nid, node_nr);
5394 }
5395 seq_putc(m, '\n');
5396
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005397 file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005398 seq_printf(m, "file=%lu", file_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005399 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005400 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005401 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07005402 seq_printf(m, " N%d=%lu", nid, node_nr);
5403 }
5404 seq_putc(m, '\n');
5405
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005406 anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005407 seq_printf(m, "anon=%lu", anon_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005408 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005409 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005410 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07005411 seq_printf(m, " N%d=%lu", nid, node_nr);
5412 }
5413 seq_putc(m, '\n');
5414
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005415 unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005416 seq_printf(m, "unevictable=%lu", unevictable_nr);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08005417 for_each_node_state(nid, N_MEMORY) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005418 node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07005419 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07005420 seq_printf(m, " N%d=%lu", nid, node_nr);
5421 }
5422 seq_putc(m, '\n');
5423 return 0;
5424}
5425#endif /* CONFIG_NUMA */
5426
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005427static inline void mem_cgroup_lru_names_not_uptodate(void)
5428{
5429 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5430}
5431
Tejun Heo182446d2013-08-08 20:11:24 -04005432static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005433 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005434{
Tejun Heo182446d2013-08-08 20:11:24 -04005435 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005436 struct mem_cgroup *mi;
5437 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005438
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005439 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005440 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005441 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005442 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5443 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005444 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005445
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005446 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5447 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5448 mem_cgroup_read_events(memcg, i));
5449
5450 for (i = 0; i < NR_LRU_LISTS; i++)
5451 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5452 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5453
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005454 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005455 {
5456 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005457 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005458 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005459 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005460 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5461 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005462 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005463
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005464 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5465 long long val = 0;
5466
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005467 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005468 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005469 for_each_mem_cgroup_tree(mi, memcg)
5470 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5471 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5472 }
5473
5474 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5475 unsigned long long val = 0;
5476
5477 for_each_mem_cgroup_tree(mi, memcg)
5478 val += mem_cgroup_read_events(mi, i);
5479 seq_printf(m, "total_%s %llu\n",
5480 mem_cgroup_events_names[i], val);
5481 }
5482
5483 for (i = 0; i < NR_LRU_LISTS; i++) {
5484 unsigned long long val = 0;
5485
5486 for_each_mem_cgroup_tree(mi, memcg)
5487 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5488 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005489 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005490
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005491#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005492 {
5493 int nid, zid;
5494 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005495 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005496 unsigned long recent_rotated[2] = {0, 0};
5497 unsigned long recent_scanned[2] = {0, 0};
5498
5499 for_each_online_node(nid)
5500 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005501 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005502 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005503
Hugh Dickins89abfab2012-05-29 15:06:53 -07005504 recent_rotated[0] += rstat->recent_rotated[0];
5505 recent_rotated[1] += rstat->recent_rotated[1];
5506 recent_scanned[0] += rstat->recent_scanned[0];
5507 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005508 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005509 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5510 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5511 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5512 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005513 }
5514#endif
5515
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005516 return 0;
5517}
5518
Tejun Heo182446d2013-08-08 20:11:24 -04005519static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5520 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005521{
Tejun Heo182446d2013-08-08 20:11:24 -04005522 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005523
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005524 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005525}
5526
Tejun Heo182446d2013-08-08 20:11:24 -04005527static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5528 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005529{
Tejun Heo182446d2013-08-08 20:11:24 -04005530 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005531 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005532
Tejun Heo63876982013-08-08 20:11:23 -04005533 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005534 return -EINVAL;
5535
Glauber Costa09998212013-02-22 16:34:55 -08005536 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005537
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005538 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005539 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005540 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005541 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005542 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005543
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005544 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005545
Glauber Costa09998212013-02-22 16:34:55 -08005546 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005547
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005548 return 0;
5549}
5550
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005551static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5552{
5553 struct mem_cgroup_threshold_ary *t;
5554 u64 usage;
5555 int i;
5556
5557 rcu_read_lock();
5558 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005559 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005560 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005561 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005562
5563 if (!t)
5564 goto unlock;
5565
5566 usage = mem_cgroup_usage(memcg, swap);
5567
5568 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005569 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005570 * If it's not true, a threshold was crossed after last
5571 * call of __mem_cgroup_threshold().
5572 */
Phil Carmody5407a562010-05-26 14:42:42 -07005573 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005574
5575 /*
5576 * Iterate backward over array of thresholds starting from
5577 * current_threshold and check if a threshold is crossed.
5578 * If none of thresholds below usage is crossed, we read
5579 * only one element of the array here.
5580 */
5581 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5582 eventfd_signal(t->entries[i].eventfd, 1);
5583
5584 /* i = current_threshold + 1 */
5585 i++;
5586
5587 /*
5588 * Iterate forward over array of thresholds starting from
5589 * current_threshold+1 and check if a threshold is crossed.
5590 * If none of thresholds above usage is crossed, we read
5591 * only one element of the array here.
5592 */
5593 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5594 eventfd_signal(t->entries[i].eventfd, 1);
5595
5596 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005597 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005598unlock:
5599 rcu_read_unlock();
5600}
5601
5602static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5603{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005604 while (memcg) {
5605 __mem_cgroup_threshold(memcg, false);
5606 if (do_swap_account)
5607 __mem_cgroup_threshold(memcg, true);
5608
5609 memcg = parent_mem_cgroup(memcg);
5610 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005611}
5612
5613static int compare_thresholds(const void *a, const void *b)
5614{
5615 const struct mem_cgroup_threshold *_a = a;
5616 const struct mem_cgroup_threshold *_b = b;
5617
Greg Thelen2bff24a2013-09-11 14:23:08 -07005618 if (_a->threshold > _b->threshold)
5619 return 1;
5620
5621 if (_a->threshold < _b->threshold)
5622 return -1;
5623
5624 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005625}
5626
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005627static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005628{
5629 struct mem_cgroup_eventfd_list *ev;
5630
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005631 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005632 eventfd_signal(ev->eventfd, 1);
5633 return 0;
5634}
5635
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005636static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005637{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005638 struct mem_cgroup *iter;
5639
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005640 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005641 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005642}
5643
Tejun Heo81eeaf02013-08-08 20:11:26 -04005644static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005645 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005646{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005647 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005648 struct mem_cgroup_thresholds *thresholds;
5649 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005650 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005651 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005652 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005653
5654 ret = res_counter_memparse_write_strategy(args, &threshold);
5655 if (ret)
5656 return ret;
5657
5658 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005659
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005660 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005661 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005662 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005663 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005664 else
5665 BUG();
5666
5667 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5668
5669 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005670 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005671 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5672
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005673 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005674
5675 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005676 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005677 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005678 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005679 ret = -ENOMEM;
5680 goto unlock;
5681 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005682 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005683
5684 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005685 if (thresholds->primary) {
5686 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005687 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005688 }
5689
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005690 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005691 new->entries[size - 1].eventfd = eventfd;
5692 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005693
5694 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005695 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005696 compare_thresholds, NULL);
5697
5698 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005699 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005700 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005701 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005702 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005703 * new->current_threshold will not be used until
5704 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005705 * it here.
5706 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005707 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005708 } else
5709 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005710 }
5711
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005712 /* Free old spare buffer and save old primary buffer as spare */
5713 kfree(thresholds->spare);
5714 thresholds->spare = thresholds->primary;
5715
5716 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005717
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005718 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005719 synchronize_rcu();
5720
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005721unlock:
5722 mutex_unlock(&memcg->thresholds_lock);
5723
5724 return ret;
5725}
5726
Tejun Heo81eeaf02013-08-08 20:11:26 -04005727static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005728 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005729{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005730 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005731 struct mem_cgroup_thresholds *thresholds;
5732 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005733 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005734 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005735 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005736
5737 mutex_lock(&memcg->thresholds_lock);
5738 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005739 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005740 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005741 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005742 else
5743 BUG();
5744
Anton Vorontsov371528c2012-02-24 05:14:46 +04005745 if (!thresholds->primary)
5746 goto unlock;
5747
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005748 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5749
5750 /* Check if a threshold crossed before removing */
5751 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5752
5753 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005754 size = 0;
5755 for (i = 0; i < thresholds->primary->size; i++) {
5756 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005757 size++;
5758 }
5759
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005760 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005761
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005762 /* Set thresholds array to NULL if we don't have thresholds */
5763 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005764 kfree(new);
5765 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005766 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005767 }
5768
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005769 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005770
5771 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005772 new->current_threshold = -1;
5773 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5774 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005775 continue;
5776
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005777 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005778 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005779 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005780 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005781 * until rcu_assign_pointer(), so it's safe to increment
5782 * it here.
5783 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005784 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005785 }
5786 j++;
5787 }
5788
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005789swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005790 /* Swap primary and spare array */
5791 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005792 /* If all events are unregistered, free the spare array */
5793 if (!new) {
5794 kfree(thresholds->spare);
5795 thresholds->spare = NULL;
5796 }
5797
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005798 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005799
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005800 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005801 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005802unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005803 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005804}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005805
Tejun Heo81eeaf02013-08-08 20:11:26 -04005806static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005807 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5808{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005809 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005810 struct mem_cgroup_eventfd_list *event;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005811 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005812
5813 BUG_ON(type != _OOM_TYPE);
5814 event = kmalloc(sizeof(*event), GFP_KERNEL);
5815 if (!event)
5816 return -ENOMEM;
5817
Michal Hocko1af8efe2011-07-26 16:08:24 -07005818 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005819
5820 event->eventfd = eventfd;
5821 list_add(&event->list, &memcg->oom_notify);
5822
5823 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005824 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005825 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005826 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005827
5828 return 0;
5829}
5830
Tejun Heo81eeaf02013-08-08 20:11:26 -04005831static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005832 struct cftype *cft, struct eventfd_ctx *eventfd)
5833{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005834 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005835 struct mem_cgroup_eventfd_list *ev, *tmp;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005836 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005837
5838 BUG_ON(type != _OOM_TYPE);
5839
Michal Hocko1af8efe2011-07-26 16:08:24 -07005840 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005841
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005842 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005843 if (ev->eventfd == eventfd) {
5844 list_del(&ev->list);
5845 kfree(ev);
5846 }
5847 }
5848
Michal Hocko1af8efe2011-07-26 16:08:24 -07005849 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005850}
5851
Tejun Heo182446d2013-08-08 20:11:24 -04005852static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005853 struct cftype *cft, struct cgroup_map_cb *cb)
5854{
Tejun Heo182446d2013-08-08 20:11:24 -04005855 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005856
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005857 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005858
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005859 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005860 cb->fill(cb, "under_oom", 1);
5861 else
5862 cb->fill(cb, "under_oom", 0);
5863 return 0;
5864}
5865
Tejun Heo182446d2013-08-08 20:11:24 -04005866static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005867 struct cftype *cft, u64 val)
5868{
Tejun Heo182446d2013-08-08 20:11:24 -04005869 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005870 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005871
5872 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005873 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005874 return -EINVAL;
5875
Glauber Costa09998212013-02-22 16:34:55 -08005876 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005877 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005878 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005879 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005880 return -EINVAL;
5881 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005882 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005883 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005884 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005885 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005886 return 0;
5887}
5888
Andrew Mortonc255a452012-07-31 16:43:02 -07005889#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005890static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005891{
Glauber Costa55007d82012-12-18 14:22:38 -08005892 int ret;
5893
Glauber Costa2633d7a2012-12-18 14:22:34 -08005894 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005895 ret = memcg_propagate_kmem(memcg);
5896 if (ret)
5897 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005898
Glauber Costa1d62e432012-04-09 19:36:33 -03005899 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005900}
Glauber Costae5671df2011-12-11 21:47:01 +00005901
Li Zefan10d5ebf2013-07-08 16:00:33 -07005902static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005903{
Glauber Costa1d62e432012-04-09 19:36:33 -03005904 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005905}
5906
5907static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5908{
5909 if (!memcg_kmem_is_active(memcg))
5910 return;
5911
5912 /*
5913 * kmem charges can outlive the cgroup. In the case of slab
5914 * pages, for instance, a page contain objects from various
5915 * processes. As we prevent from taking a reference for every
5916 * such allocation we have to be careful when doing uncharge
5917 * (see memcg_uncharge_kmem) and here during offlining.
5918 *
5919 * The idea is that that only the _last_ uncharge which sees
5920 * the dead memcg will drop the last reference. An additional
5921 * reference is taken here before the group is marked dead
5922 * which is then paired with css_put during uncharge resp. here.
5923 *
5924 * Although this might sound strange as this path is called from
5925 * css_offline() when the referencemight have dropped down to 0
5926 * and shouldn't be incremented anymore (css_tryget would fail)
5927 * we do not have other options because of the kmem allocations
5928 * lifetime.
5929 */
5930 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005931
5932 memcg_kmem_mark_dead(memcg);
5933
5934 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5935 return;
5936
Glauber Costa7de37682012-12-18 14:22:07 -08005937 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005938 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005939}
Glauber Costae5671df2011-12-11 21:47:01 +00005940#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005941static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005942{
5943 return 0;
5944}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005945
Li Zefan10d5ebf2013-07-08 16:00:33 -07005946static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5947{
5948}
5949
5950static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005951{
5952}
Glauber Costae5671df2011-12-11 21:47:01 +00005953#endif
5954
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005955static struct cftype mem_cgroup_files[] = {
5956 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005957 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005958 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07005959 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005960 .register_event = mem_cgroup_usage_register_event,
5961 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005962 },
5963 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005964 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005965 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005966 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005967 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005968 },
5969 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005970 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005971 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07005972 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005973 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005974 },
5975 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005976 .name = "soft_limit_in_bytes",
5977 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
5978 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005979 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07005980 },
5981 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005982 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005983 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005984 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005985 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005986 },
Balbir Singh8697d332008-02-07 00:13:59 -08005987 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005988 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07005989 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005990 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005991 {
5992 .name = "force_empty",
5993 .trigger = mem_cgroup_force_empty_write,
5994 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005995 {
5996 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07005997 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08005998 .write_u64 = mem_cgroup_hierarchy_write,
5999 .read_u64 = mem_cgroup_hierarchy_read,
6000 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006001 {
6002 .name = "swappiness",
6003 .read_u64 = mem_cgroup_swappiness_read,
6004 .write_u64 = mem_cgroup_swappiness_write,
6005 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006006 {
6007 .name = "move_charge_at_immigrate",
6008 .read_u64 = mem_cgroup_move_charge_read,
6009 .write_u64 = mem_cgroup_move_charge_write,
6010 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006011 {
6012 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006013 .read_map = mem_cgroup_oom_control_read,
6014 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006015 .register_event = mem_cgroup_oom_register_event,
6016 .unregister_event = mem_cgroup_oom_unregister_event,
6017 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6018 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006019 {
6020 .name = "pressure_level",
6021 .register_event = vmpressure_register_event,
6022 .unregister_event = vmpressure_unregister_event,
6023 },
Ying Han406eb0c2011-05-26 16:25:37 -07006024#ifdef CONFIG_NUMA
6025 {
6026 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07006027 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006028 },
6029#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006030#ifdef CONFIG_MEMCG_KMEM
6031 {
6032 .name = "kmem.limit_in_bytes",
6033 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6034 .write_string = mem_cgroup_write,
6035 .read = mem_cgroup_read,
6036 },
6037 {
6038 .name = "kmem.usage_in_bytes",
6039 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
6040 .read = mem_cgroup_read,
6041 },
6042 {
6043 .name = "kmem.failcnt",
6044 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6045 .trigger = mem_cgroup_reset,
6046 .read = mem_cgroup_read,
6047 },
6048 {
6049 .name = "kmem.max_usage_in_bytes",
6050 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6051 .trigger = mem_cgroup_reset,
6052 .read = mem_cgroup_read,
6053 },
Glauber Costa749c5412012-12-18 14:23:01 -08006054#ifdef CONFIG_SLABINFO
6055 {
6056 .name = "kmem.slabinfo",
6057 .read_seq_string = mem_cgroup_slabinfo_read,
6058 },
6059#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006060#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006061 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006062};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006063
Michal Hocko2d110852013-02-22 16:34:43 -08006064#ifdef CONFIG_MEMCG_SWAP
6065static struct cftype memsw_cgroup_files[] = {
6066 {
6067 .name = "memsw.usage_in_bytes",
6068 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6069 .read = mem_cgroup_read,
6070 .register_event = mem_cgroup_usage_register_event,
6071 .unregister_event = mem_cgroup_usage_unregister_event,
6072 },
6073 {
6074 .name = "memsw.max_usage_in_bytes",
6075 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6076 .trigger = mem_cgroup_reset,
6077 .read = mem_cgroup_read,
6078 },
6079 {
6080 .name = "memsw.limit_in_bytes",
6081 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6082 .write_string = mem_cgroup_write,
6083 .read = mem_cgroup_read,
6084 },
6085 {
6086 .name = "memsw.failcnt",
6087 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6088 .trigger = mem_cgroup_reset,
6089 .read = mem_cgroup_read,
6090 },
6091 { }, /* terminate */
6092};
6093#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006094static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006095{
6096 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006097 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006098 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006099 /*
6100 * This routine is called against possible nodes.
6101 * But it's BUG to call kmalloc() against offline node.
6102 *
6103 * TODO: this routine can waste much memory for nodes which will
6104 * never be onlined. It's better to use memory hotplug callback
6105 * function.
6106 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006107 if (!node_state(node, N_NORMAL_MEMORY))
6108 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006109 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006110 if (!pn)
6111 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006112
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006113 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6114 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006115 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006116 mz->usage_in_excess = 0;
6117 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006118 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006119 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006120 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006121 return 0;
6122}
6123
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006124static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006125{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006126 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006127}
6128
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006129static struct mem_cgroup *mem_cgroup_alloc(void)
6130{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006131 struct mem_cgroup *memcg;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006132 size_t size = memcg_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006133
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006134 /* Can be very big if nr_node_ids is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08006135 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006136 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006137 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006138 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006139
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006140 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006141 return NULL;
6142
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006143 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6144 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006145 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006146 spin_lock_init(&memcg->pcp_counter_lock);
6147 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006148
6149out_free:
6150 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006151 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006152 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006153 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006154 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006155}
6156
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006157/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006158 * At destroying mem_cgroup, references from swap_cgroup can remain.
6159 * (scanning all at force_empty is too costly...)
6160 *
6161 * Instead of clearing all references at force_empty, we remember
6162 * the number of reference from swap_cgroup and free mem_cgroup when
6163 * it goes down to 0.
6164 *
6165 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006166 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006167
6168static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006169{
Glauber Costac8b2a362012-12-18 14:22:13 -08006170 int node;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006171 size_t size = memcg_size();
Hugh Dickins59927fb2012-03-15 15:17:07 -07006172
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006173 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006174 free_css_id(&mem_cgroup_subsys, &memcg->css);
6175
6176 for_each_node(node)
6177 free_mem_cgroup_per_zone_info(memcg, node);
6178
6179 free_percpu(memcg->stat);
6180
Glauber Costa3f134612012-05-29 15:07:11 -07006181 /*
6182 * We need to make sure that (at least for now), the jump label
6183 * destruction code runs outside of the cgroup lock. This is because
6184 * get_online_cpus(), which is called from the static_branch update,
6185 * can't be called inside the cgroup_lock. cpusets are the ones
6186 * enforcing this dependency, so if they ever change, we might as well.
6187 *
6188 * schedule_work() will guarantee this happens. Be careful if you need
6189 * to move this code around, and make sure it is outside
6190 * the cgroup_lock.
6191 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006192 disarm_static_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006193 if (size < PAGE_SIZE)
6194 kfree(memcg);
6195 else
6196 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006197}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006198
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006199/*
6200 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6201 */
Glauber Costae1aab162011-12-11 21:47:03 +00006202struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006203{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006204 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006205 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006206 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006207}
Glauber Costae1aab162011-12-11 21:47:03 +00006208EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006209
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006210static void __init mem_cgroup_soft_limit_tree_init(void)
6211{
6212 struct mem_cgroup_tree_per_node *rtpn;
6213 struct mem_cgroup_tree_per_zone *rtpz;
6214 int tmp, node, zone;
6215
6216 for_each_node(node) {
6217 tmp = node;
6218 if (!node_state(node, N_NORMAL_MEMORY))
6219 tmp = -1;
6220 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
6221 BUG_ON(!rtpn);
6222
6223 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6224
6225 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6226 rtpz = &rtpn->rb_tree_per_zone[zone];
6227 rtpz->rb_root = RB_ROOT;
6228 spin_lock_init(&rtpz->lock);
6229 }
6230 }
6231}
6232
Li Zefan0eb253e2009-01-15 13:51:25 -08006233static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006234mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006235{
Glauber Costad142e3e2013-02-22 16:34:52 -08006236 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006237 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006238 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006239
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006240 memcg = mem_cgroup_alloc();
6241 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006242 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006243
Bob Liu3ed28fa2012-01-12 17:19:04 -08006244 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006245 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006246 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006247
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006248 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006249 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006250 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006251 res_counter_init(&memcg->res, NULL);
6252 res_counter_init(&memcg->memsw, NULL);
6253 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006254 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006255
Glauber Costad142e3e2013-02-22 16:34:52 -08006256 memcg->last_scanned_node = MAX_NUMNODES;
6257 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006258 memcg->move_charge_at_immigrate = 0;
6259 mutex_init(&memcg->thresholds_lock);
6260 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006261 vmpressure_init(&memcg->vmpressure);
Glauber Costad142e3e2013-02-22 16:34:52 -08006262
6263 return &memcg->css;
6264
6265free_out:
6266 __mem_cgroup_free(memcg);
6267 return ERR_PTR(error);
6268}
6269
6270static int
Tejun Heoeb954192013-08-08 20:11:23 -04006271mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006272{
Tejun Heoeb954192013-08-08 20:11:23 -04006273 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6274 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006275 int error = 0;
6276
Tejun Heo63876982013-08-08 20:11:23 -04006277 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006278 return 0;
6279
Glauber Costa09998212013-02-22 16:34:55 -08006280 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006281
6282 memcg->use_hierarchy = parent->use_hierarchy;
6283 memcg->oom_kill_disable = parent->oom_kill_disable;
6284 memcg->swappiness = mem_cgroup_swappiness(parent);
6285
6286 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006287 res_counter_init(&memcg->res, &parent->res);
6288 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006289 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006290
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006291 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006292 * No need to take a reference to the parent because cgroup
6293 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006294 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006295 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006296 res_counter_init(&memcg->res, NULL);
6297 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006298 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006299 /*
6300 * Deeper hierachy with use_hierarchy == false doesn't make
6301 * much sense so let cgroup subsystem know about this
6302 * unfortunate state in our controller.
6303 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006304 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006305 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006306 }
Glauber Costacbe128e32012-04-09 19:36:34 -03006307
6308 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
Glauber Costa09998212013-02-22 16:34:55 -08006309 mutex_unlock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006310 return error;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006311}
6312
Michal Hocko5f578162013-04-29 15:07:17 -07006313/*
6314 * Announce all parents that a group from their hierarchy is gone.
6315 */
6316static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6317{
6318 struct mem_cgroup *parent = memcg;
6319
6320 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006321 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006322
6323 /*
6324 * if the root memcg is not hierarchical we have to check it
6325 * explicitely.
6326 */
6327 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006328 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006329}
6330
Tejun Heoeb954192013-08-08 20:11:23 -04006331static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006332{
Tejun Heoeb954192013-08-08 20:11:23 -04006333 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006334
Li Zefan10d5ebf2013-07-08 16:00:33 -07006335 kmem_cgroup_css_offline(memcg);
6336
Michal Hocko5f578162013-04-29 15:07:17 -07006337 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006338 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006339 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006340 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006341}
6342
Tejun Heoeb954192013-08-08 20:11:23 -04006343static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006344{
Tejun Heoeb954192013-08-08 20:11:23 -04006345 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006346
Li Zefan10d5ebf2013-07-08 16:00:33 -07006347 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006348 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006349}
6350
Daisuke Nishimura02491442010-03-10 15:22:17 -08006351#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006352/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006353#define PRECHARGE_COUNT_AT_ONCE 256
6354static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006355{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006356 int ret = 0;
6357 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006358 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006359
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006360 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006361 mc.precharge += count;
6362 /* we don't need css_get for root */
6363 return ret;
6364 }
6365 /* try to charge at once */
6366 if (count > 1) {
6367 struct res_counter *dummy;
6368 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006369 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006370 * by cgroup_lock_live_cgroup() that it is not removed and we
6371 * are still under the same cgroup_mutex. So we can postpone
6372 * css_get().
6373 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006374 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006375 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006376 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006377 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006378 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006379 goto one_by_one;
6380 }
6381 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006382 return ret;
6383 }
6384one_by_one:
6385 /* fall back to one by one charge */
6386 while (count--) {
6387 if (signal_pending(current)) {
6388 ret = -EINTR;
6389 break;
6390 }
6391 if (!batch_count--) {
6392 batch_count = PRECHARGE_COUNT_AT_ONCE;
6393 cond_resched();
6394 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006395 ret = __mem_cgroup_try_charge(NULL,
6396 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006397 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006398 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006399 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006400 mc.precharge++;
6401 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006402 return ret;
6403}
6404
6405/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006406 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006407 * @vma: the vma the pte to be checked belongs
6408 * @addr: the address corresponding to the pte to be checked
6409 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006410 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006411 *
6412 * Returns
6413 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6414 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6415 * move charge. if @target is not NULL, the page is stored in target->page
6416 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006417 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6418 * target for charge migration. if @target is not NULL, the entry is stored
6419 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006420 *
6421 * Called with pte lock held.
6422 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006423union mc_target {
6424 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006425 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006426};
6427
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006428enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006429 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006430 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006431 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006432};
6433
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006434static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6435 unsigned long addr, pte_t ptent)
6436{
6437 struct page *page = vm_normal_page(vma, addr, ptent);
6438
6439 if (!page || !page_mapped(page))
6440 return NULL;
6441 if (PageAnon(page)) {
6442 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006443 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006444 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006445 } else if (!move_file())
6446 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006447 return NULL;
6448 if (!get_page_unless_zero(page))
6449 return NULL;
6450
6451 return page;
6452}
6453
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006454#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006455static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6456 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6457{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006458 struct page *page = NULL;
6459 swp_entry_t ent = pte_to_swp_entry(ptent);
6460
6461 if (!move_anon() || non_swap_entry(ent))
6462 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006463 /*
6464 * Because lookup_swap_cache() updates some statistics counter,
6465 * we call find_get_page() with swapper_space directly.
6466 */
Shaohua Li33806f02013-02-22 16:34:37 -08006467 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006468 if (do_swap_account)
6469 entry->val = ent.val;
6470
6471 return page;
6472}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006473#else
6474static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6475 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6476{
6477 return NULL;
6478}
6479#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006480
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006481static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6482 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6483{
6484 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006485 struct address_space *mapping;
6486 pgoff_t pgoff;
6487
6488 if (!vma->vm_file) /* anonymous vma */
6489 return NULL;
6490 if (!move_file())
6491 return NULL;
6492
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006493 mapping = vma->vm_file->f_mapping;
6494 if (pte_none(ptent))
6495 pgoff = linear_page_index(vma, addr);
6496 else /* pte_file(ptent) is true */
6497 pgoff = pte_to_pgoff(ptent);
6498
6499 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006500 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006501
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006502#ifdef CONFIG_SWAP
6503 /* shmem/tmpfs may report page out on swap: account for that too. */
6504 if (radix_tree_exceptional_entry(page)) {
6505 swp_entry_t swap = radix_to_swp_entry(page);
6506 if (do_swap_account)
6507 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006508 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006509 }
6510#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006511 return page;
6512}
6513
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006514static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006515 unsigned long addr, pte_t ptent, union mc_target *target)
6516{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006517 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006518 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006519 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006520 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006521
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006522 if (pte_present(ptent))
6523 page = mc_handle_present_pte(vma, addr, ptent);
6524 else if (is_swap_pte(ptent))
6525 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006526 else if (pte_none(ptent) || pte_file(ptent))
6527 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006528
6529 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006530 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006531 if (page) {
6532 pc = lookup_page_cgroup(page);
6533 /*
6534 * Do only loose check w/o page_cgroup lock.
6535 * mem_cgroup_move_account() checks the pc is valid or not under
6536 * the lock.
6537 */
6538 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6539 ret = MC_TARGET_PAGE;
6540 if (target)
6541 target->page = page;
6542 }
6543 if (!ret || !target)
6544 put_page(page);
6545 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006546 /* There is a swap entry and a page doesn't exist or isn't charged */
6547 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08006548 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006549 ret = MC_TARGET_SWAP;
6550 if (target)
6551 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006552 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006553 return ret;
6554}
6555
Naoya Horiguchi12724852012-03-21 16:34:28 -07006556#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6557/*
6558 * We don't consider swapping or file mapped pages because THP does not
6559 * support them for now.
6560 * Caller should make sure that pmd_trans_huge(pmd) is true.
6561 */
6562static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6563 unsigned long addr, pmd_t pmd, union mc_target *target)
6564{
6565 struct page *page = NULL;
6566 struct page_cgroup *pc;
6567 enum mc_target_type ret = MC_TARGET_NONE;
6568
6569 page = pmd_page(pmd);
6570 VM_BUG_ON(!page || !PageHead(page));
6571 if (!move_anon())
6572 return ret;
6573 pc = lookup_page_cgroup(page);
6574 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6575 ret = MC_TARGET_PAGE;
6576 if (target) {
6577 get_page(page);
6578 target->page = page;
6579 }
6580 }
6581 return ret;
6582}
6583#else
6584static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6585 unsigned long addr, pmd_t pmd, union mc_target *target)
6586{
6587 return MC_TARGET_NONE;
6588}
6589#endif
6590
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006591static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6592 unsigned long addr, unsigned long end,
6593 struct mm_walk *walk)
6594{
6595 struct vm_area_struct *vma = walk->private;
6596 pte_t *pte;
6597 spinlock_t *ptl;
6598
Naoya Horiguchi12724852012-03-21 16:34:28 -07006599 if (pmd_trans_huge_lock(pmd, vma) == 1) {
6600 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6601 mc.precharge += HPAGE_PMD_NR;
6602 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006603 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006604 }
Dave Hansen03319322011-03-22 16:32:56 -07006605
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006606 if (pmd_trans_unstable(pmd))
6607 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006608 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6609 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006610 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006611 mc.precharge++; /* increment precharge temporarily */
6612 pte_unmap_unlock(pte - 1, ptl);
6613 cond_resched();
6614
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006615 return 0;
6616}
6617
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006618static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6619{
6620 unsigned long precharge;
6621 struct vm_area_struct *vma;
6622
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006623 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006624 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6625 struct mm_walk mem_cgroup_count_precharge_walk = {
6626 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6627 .mm = mm,
6628 .private = vma,
6629 };
6630 if (is_vm_hugetlb_page(vma))
6631 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006632 walk_page_range(vma->vm_start, vma->vm_end,
6633 &mem_cgroup_count_precharge_walk);
6634 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006635 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006636
6637 precharge = mc.precharge;
6638 mc.precharge = 0;
6639
6640 return precharge;
6641}
6642
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006643static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6644{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006645 unsigned long precharge = mem_cgroup_count_precharge(mm);
6646
6647 VM_BUG_ON(mc.moving_task);
6648 mc.moving_task = current;
6649 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006650}
6651
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006652/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6653static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006654{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006655 struct mem_cgroup *from = mc.from;
6656 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006657 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006658
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006659 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006660 if (mc.precharge) {
6661 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6662 mc.precharge = 0;
6663 }
6664 /*
6665 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6666 * we must uncharge here.
6667 */
6668 if (mc.moved_charge) {
6669 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6670 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006671 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006672 /* we must fixup refcnts and charges */
6673 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006674 /* uncharge swap account from the old cgroup */
6675 if (!mem_cgroup_is_root(mc.from))
6676 res_counter_uncharge(&mc.from->memsw,
6677 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006678
6679 for (i = 0; i < mc.moved_swap; i++)
6680 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006681
6682 if (!mem_cgroup_is_root(mc.to)) {
6683 /*
6684 * we charged both to->res and to->memsw, so we should
6685 * uncharge to->res.
6686 */
6687 res_counter_uncharge(&mc.to->res,
6688 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006689 }
Li Zefan40503772013-07-08 16:00:34 -07006690 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006691 mc.moved_swap = 0;
6692 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006693 memcg_oom_recover(from);
6694 memcg_oom_recover(to);
6695 wake_up_all(&mc.waitq);
6696}
6697
6698static void mem_cgroup_clear_mc(void)
6699{
6700 struct mem_cgroup *from = mc.from;
6701
6702 /*
6703 * we must clear moving_task before waking up waiters at the end of
6704 * task migration.
6705 */
6706 mc.moving_task = NULL;
6707 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006708 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006709 mc.from = NULL;
6710 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006711 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006712 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006713}
6714
Tejun Heoeb954192013-08-08 20:11:23 -04006715static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006716 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006717{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006718 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006719 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04006720 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08006721 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006722
Glauber Costaee5e8472013-02-22 16:34:50 -08006723 /*
6724 * We are now commited to this value whatever it is. Changes in this
6725 * tunable will only affect upcoming migrations, not the current one.
6726 * So we need to save it, and keep it going.
6727 */
6728 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6729 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006730 struct mm_struct *mm;
6731 struct mem_cgroup *from = mem_cgroup_from_task(p);
6732
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006733 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006734
6735 mm = get_task_mm(p);
6736 if (!mm)
6737 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006738 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006739 if (mm->owner == p) {
6740 VM_BUG_ON(mc.from);
6741 VM_BUG_ON(mc.to);
6742 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006743 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006744 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006745 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006746 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006747 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006748 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006749 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006750 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006751 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006752
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006753 ret = mem_cgroup_precharge_mc(mm);
6754 if (ret)
6755 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006756 }
6757 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006758 }
6759 return ret;
6760}
6761
Tejun Heoeb954192013-08-08 20:11:23 -04006762static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006763 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006764{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006765 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006766}
6767
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006768static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6769 unsigned long addr, unsigned long end,
6770 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006771{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006772 int ret = 0;
6773 struct vm_area_struct *vma = walk->private;
6774 pte_t *pte;
6775 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006776 enum mc_target_type target_type;
6777 union mc_target target;
6778 struct page *page;
6779 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006780
Naoya Horiguchi12724852012-03-21 16:34:28 -07006781 /*
6782 * We don't take compound_lock() here but no race with splitting thp
6783 * happens because:
6784 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6785 * under splitting, which means there's no concurrent thp split,
6786 * - if another thread runs into split_huge_page() just after we
6787 * entered this if-block, the thread must wait for page table lock
6788 * to be unlocked in __split_huge_page_splitting(), where the main
6789 * part of thp split is not executed yet.
6790 */
6791 if (pmd_trans_huge_lock(pmd, vma) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006792 if (mc.precharge < HPAGE_PMD_NR) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006793 spin_unlock(&vma->vm_mm->page_table_lock);
6794 return 0;
6795 }
6796 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6797 if (target_type == MC_TARGET_PAGE) {
6798 page = target.page;
6799 if (!isolate_lru_page(page)) {
6800 pc = lookup_page_cgroup(page);
6801 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006802 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006803 mc.precharge -= HPAGE_PMD_NR;
6804 mc.moved_charge += HPAGE_PMD_NR;
6805 }
6806 putback_lru_page(page);
6807 }
6808 put_page(page);
6809 }
6810 spin_unlock(&vma->vm_mm->page_table_lock);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006811 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006812 }
6813
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006814 if (pmd_trans_unstable(pmd))
6815 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006816retry:
6817 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6818 for (; addr != end; addr += PAGE_SIZE) {
6819 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006820 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006821
6822 if (!mc.precharge)
6823 break;
6824
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006825 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006826 case MC_TARGET_PAGE:
6827 page = target.page;
6828 if (isolate_lru_page(page))
6829 goto put;
6830 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006831 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006832 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006833 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006834 /* we uncharge from mc.from later. */
6835 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006836 }
6837 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006838put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006839 put_page(page);
6840 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006841 case MC_TARGET_SWAP:
6842 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006843 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006844 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006845 /* we fixup refcnts and charges later. */
6846 mc.moved_swap++;
6847 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006848 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006849 default:
6850 break;
6851 }
6852 }
6853 pte_unmap_unlock(pte - 1, ptl);
6854 cond_resched();
6855
6856 if (addr != end) {
6857 /*
6858 * We have consumed all precharges we got in can_attach().
6859 * We try charge one by one, but don't do any additional
6860 * charges to mc.to if we have failed in charge once in attach()
6861 * phase.
6862 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006863 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006864 if (!ret)
6865 goto retry;
6866 }
6867
6868 return ret;
6869}
6870
6871static void mem_cgroup_move_charge(struct mm_struct *mm)
6872{
6873 struct vm_area_struct *vma;
6874
6875 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006876retry:
6877 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6878 /*
6879 * Someone who are holding the mmap_sem might be waiting in
6880 * waitq. So we cancel all extra charges, wake up all waiters,
6881 * and retry. Because we cancel precharges, we might not be able
6882 * to move enough charges, but moving charge is a best-effort
6883 * feature anyway, so it wouldn't be a big problem.
6884 */
6885 __mem_cgroup_clear_mc();
6886 cond_resched();
6887 goto retry;
6888 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006889 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6890 int ret;
6891 struct mm_walk mem_cgroup_move_charge_walk = {
6892 .pmd_entry = mem_cgroup_move_charge_pte_range,
6893 .mm = mm,
6894 .private = vma,
6895 };
6896 if (is_vm_hugetlb_page(vma))
6897 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006898 ret = walk_page_range(vma->vm_start, vma->vm_end,
6899 &mem_cgroup_move_charge_walk);
6900 if (ret)
6901 /*
6902 * means we have consumed all precharges and failed in
6903 * doing additional charge. Just abandon here.
6904 */
6905 break;
6906 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006907 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006908}
6909
Tejun Heoeb954192013-08-08 20:11:23 -04006910static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006911 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006912{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006913 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006914 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006915
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006916 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006917 if (mc.to)
6918 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006919 mmput(mm);
6920 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006921 if (mc.to)
6922 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006923}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006924#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04006925static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006926 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006927{
6928 return 0;
6929}
Tejun Heoeb954192013-08-08 20:11:23 -04006930static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006931 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006932{
6933}
Tejun Heoeb954192013-08-08 20:11:23 -04006934static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006935 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006936{
6937}
6938#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006939
Tejun Heof00baae2013-04-15 13:41:15 -07006940/*
6941 * Cgroup retains root cgroups across [un]mount cycles making it necessary
6942 * to verify sane_behavior flag on each mount attempt.
6943 */
Tejun Heoeb954192013-08-08 20:11:23 -04006944static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006945{
6946 /*
6947 * use_hierarchy is forced with sane_behavior. cgroup core
6948 * guarantees that @root doesn't have any children, so turning it
6949 * on for the root memcg is enough.
6950 */
Tejun Heoeb954192013-08-08 20:11:23 -04006951 if (cgroup_sane_behavior(root_css->cgroup))
6952 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07006953}
6954
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006955struct cgroup_subsys mem_cgroup_subsys = {
6956 .name = "memory",
6957 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08006958 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006959 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006960 .css_offline = mem_cgroup_css_offline,
6961 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006962 .can_attach = mem_cgroup_can_attach,
6963 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08006964 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006965 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07006966 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006967 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006968 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006969};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006970
Andrew Mortonc255a452012-07-31 16:43:02 -07006971#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08006972static int __init enable_swap_account(char *s)
6973{
Michal Hockoa2c89902011-05-24 17:12:50 -07006974 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006975 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07006976 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08006977 really_do_swap_account = 0;
6978 return 1;
6979}
Michal Hockoa2c89902011-05-24 17:12:50 -07006980__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006981
Michal Hocko2d110852013-02-22 16:34:43 -08006982static void __init memsw_file_init(void)
6983{
Michal Hocko6acc8b02013-02-22 16:34:45 -08006984 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08006985}
Michal Hocko6acc8b02013-02-22 16:34:45 -08006986
6987static void __init enable_swap_cgroup(void)
6988{
6989 if (!mem_cgroup_disabled() && really_do_swap_account) {
6990 do_swap_account = 1;
6991 memsw_file_init();
6992 }
6993}
6994
Michal Hocko2d110852013-02-22 16:34:43 -08006995#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08006996static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08006997{
6998}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006999#endif
Michal Hocko2d110852013-02-22 16:34:43 -08007000
7001/*
Michal Hocko10813122013-02-22 16:35:41 -08007002 * subsys_initcall() for memory controller.
7003 *
7004 * Some parts like hotcpu_notifier() have to be initialized from this context
7005 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7006 * everything that doesn't depend on a specific mem_cgroup structure should
7007 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007008 */
7009static int __init mem_cgroup_init(void)
7010{
7011 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007012 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07007013 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007014 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007015 return 0;
7016}
7017subsys_initcall(mem_cgroup_init);