blob: 3065fa80251db17353ae4543266e7b38713e1d9e [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>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070057#include <linux/lockdep.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080058#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000059#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070060#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000061#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080062#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
Balbir Singh8697d332008-02-07 00:13:59 -080064#include <asm/uaccess.h>
65
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070066#include <trace/events/vmscan.h>
67
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070068struct cgroup_subsys mem_cgroup_subsys __read_mostly;
David Rientjes68ae5642012-12-12 13:51:57 -080069EXPORT_SYMBOL(mem_cgroup_subsys);
70
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070071#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070072static struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080073
Andrew Mortonc255a452012-07-31 16:43:02 -070074#ifdef CONFIG_MEMCG_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070075/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080076int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080077
78/* for remember boot option*/
Andrew Mortonc255a452012-07-31 16:43:02 -070079#ifdef CONFIG_MEMCG_SWAP_ENABLED
Michal Hockoa42c3902010-11-24 12:57:08 -080080static int really_do_swap_account __initdata = 1;
81#else
82static int really_do_swap_account __initdata = 0;
83#endif
84
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070086#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080087#endif
88
89
Johannes Weineraf7c4b02012-05-29 15:07:08 -070090static const char * const mem_cgroup_stat_names[] = {
91 "cache",
92 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070093 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070094 "mapped_file",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070095 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070096 "swap",
97};
98
Johannes Weinere9f89742011-03-23 16:42:37 -070099enum mem_cgroup_events_index {
100 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
101 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
Ying Han456f9982011-05-26 16:25:38 -0700102 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
103 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700104 MEM_CGROUP_EVENTS_NSTATS,
105};
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700106
107static const char * const mem_cgroup_events_names[] = {
108 "pgpgin",
109 "pgpgout",
110 "pgfault",
111 "pgmajfault",
112};
113
Sha Zhengju58cf1882013-02-22 16:32:05 -0800114static const char * const mem_cgroup_lru_names[] = {
115 "inactive_anon",
116 "active_anon",
117 "inactive_file",
118 "active_file",
119 "unevictable",
120};
121
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700122/*
123 * Per memcg event counter is incremented at every pagein/pageout. With THP,
124 * it will be incremated by the number of pages. This counter is used for
125 * for trigger some periodic events. This is straightforward and better
126 * than using jiffies etc. to handle periodic memcg event.
127 */
128enum mem_cgroup_events_target {
129 MEM_CGROUP_TARGET_THRESH,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700131 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700132 MEM_CGROUP_NTARGETS,
133};
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700134#define THRESHOLDS_EVENTS_TARGET 128
135#define SOFTLIMIT_EVENTS_TARGET 1024
136#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700137
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800138struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700139 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700140 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner13114712012-05-29 15:07:07 -0700141 unsigned long nr_page_events;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700142 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800143};
144
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800145struct mem_cgroup_reclaim_iter {
Michal Hocko5f578162013-04-29 15:07:17 -0700146 /*
147 * last scanned hierarchy member. Valid only if last_dead_count
148 * matches memcg->dead_count of the hierarchy root group.
149 */
Michal Hocko542f85f2013-04-29 15:07:15 -0700150 struct mem_cgroup *last_visited;
Michal Hocko5f578162013-04-29 15:07:17 -0700151 unsigned long last_dead_count;
152
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800153 /* scan generation, increased every round-trip */
154 unsigned int generation;
155};
156
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800157/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800158 * per-zone information in memory controller.
159 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800160struct mem_cgroup_per_zone {
Johannes Weiner6290df52012-01-12 17:18:10 -0800161 struct lruvec lruvec;
Hugh Dickins1eb49272012-03-21 16:34:19 -0700162 unsigned long lru_size[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800163
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800164 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
165
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700166 struct rb_node tree_node; /* RB tree node */
167 unsigned long long usage_in_excess;/* Set to the value by which */
168 /* the soft limit is exceeded*/
169 bool on_tree;
Hugh Dickinsd79154b2012-03-21 16:34:18 -0700170 struct mem_cgroup *memcg; /* Back pointer, we cannot */
Balbir Singh4e416952009-09-23 15:56:39 -0700171 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800172};
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800173
174struct mem_cgroup_per_node {
175 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
176};
177
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700178/*
179 * Cgroups above their limits are maintained in a RB-Tree, independent of
180 * their hierarchy representation
181 */
182
183struct mem_cgroup_tree_per_zone {
184 struct rb_root rb_root;
185 spinlock_t lock;
186};
187
188struct mem_cgroup_tree_per_node {
189 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
190};
191
192struct mem_cgroup_tree {
193 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
194};
195
196static struct mem_cgroup_tree soft_limit_tree __read_mostly;
197
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800198struct mem_cgroup_threshold {
199 struct eventfd_ctx *eventfd;
200 u64 threshold;
201};
202
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700203/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800204struct mem_cgroup_threshold_ary {
Sha Zhengju748dad32012-05-29 15:06:57 -0700205 /* An array index points to threshold just below or equal to usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700206 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800207 /* Size of entries[] */
208 unsigned int size;
209 /* Array of thresholds */
210 struct mem_cgroup_threshold entries[0];
211};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700212
213struct mem_cgroup_thresholds {
214 /* Primary thresholds array */
215 struct mem_cgroup_threshold_ary *primary;
216 /*
217 * Spare threshold array.
218 * This is needed to make mem_cgroup_unregister_event() "never fail".
219 * It must be able to store at least primary->size - 1 entries.
220 */
221 struct mem_cgroup_threshold_ary *spare;
222};
223
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700224/* for OOM */
225struct mem_cgroup_eventfd_list {
226 struct list_head list;
227 struct eventfd_ctx *eventfd;
228};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800229
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700230static void mem_cgroup_threshold(struct mem_cgroup *memcg);
231static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800232
Balbir Singhf64c3f52009-09-23 15:56:37 -0700233/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800234 * The memory controller data structure. The memory controller controls both
235 * page cache and RSS per cgroup. We would eventually like to provide
236 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
237 * to help the administrator determine what knobs to tune.
238 *
239 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800240 * we hit the water mark. May be even add a low water mark, such that
241 * no reclaim occurs from a cgroup at it's low water mark, this is
242 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800243 */
244struct mem_cgroup {
245 struct cgroup_subsys_state css;
246 /*
247 * the counter to account for memory usage
248 */
249 struct res_counter res;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700250
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700251 /* vmpressure notifications */
252 struct vmpressure vmpressure;
253
Li Zefan465939a2013-07-08 16:00:38 -0700254 /*
255 * the counter to account for mem+swap usage.
256 */
257 struct res_counter memsw;
Hugh Dickins59927fb2012-03-15 15:17:07 -0700258
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800259 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -0800260 * the counter to account for kernel memory usage.
261 */
262 struct res_counter kmem;
263 /*
Balbir Singh18f59ea2009-01-07 18:08:07 -0800264 * Should the accounting and control be hierarchical, per subtree?
265 */
266 bool use_hierarchy;
Glauber Costa510fc4e2012-12-18 14:21:47 -0800267 unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */
Michal Hocko79dfdac2011-07-26 16:08:23 -0700268
269 bool oom_lock;
270 atomic_t under_oom;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700271 atomic_t oom_wakeups;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700272
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700273 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700274 /* OOM-Killer disable */
275 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800276
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700277 /* set when res.limit == memsw.limit */
278 bool memsw_is_minimum;
279
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800280 /* protect arrays of thresholds */
281 struct mutex thresholds_lock;
282
283 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700284 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700285
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800286 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700287 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700288
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700289 /* For oom notifier event fd */
290 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700291
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800292 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800293 * Should we move charges of a task when a task is moved into this
294 * mem_cgroup ? And what type of charges should we move ?
295 */
Andrew Mortonf894ffa2013-09-12 15:13:35 -0700296 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800297 /*
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700298 * set > 0 if pages under this cgroup are moving to other cgroup.
299 */
300 atomic_t moving_account;
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -0700301 /* taken only while moving_account > 0 */
302 spinlock_t move_lock;
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -0700303 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800304 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800305 */
Kirill A. Shutemov3a7951b2012-05-29 15:06:56 -0700306 struct mem_cgroup_stat_cpu __percpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700307 /*
308 * used when a cpu is offlined or other synchronizations
309 * See mem_cgroup_read_stat().
310 */
311 struct mem_cgroup_stat_cpu nocpu_base;
312 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000313
Michal Hocko5f578162013-04-29 15:07:17 -0700314 atomic_t dead_count;
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700315#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700316 struct cg_proto tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000317#endif
Glauber Costa2633d7a2012-12-18 14:22:34 -0800318#if defined(CONFIG_MEMCG_KMEM)
319 /* analogous to slab_common's slab_caches list. per-memcg */
320 struct list_head memcg_slab_caches;
321 /* Not a spinlock, we can take a lot of time walking the list */
322 struct mutex slab_caches_mutex;
323 /* Index in the kmem_cache->memcg_params->memcg_caches array */
324 int kmemcg_id;
325#endif
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800326
327 int last_scanned_node;
328#if MAX_NUMNODES > 1
329 nodemask_t scan_nodes;
330 atomic_t numainfo_events;
331 atomic_t numainfo_updating;
332#endif
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700333
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700334 struct mem_cgroup_per_node *nodeinfo[0];
335 /* WARNING: nodeinfo must be the last member here */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800336};
337
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800338static size_t memcg_size(void)
339{
340 return sizeof(struct mem_cgroup) +
Vladimir Davydov695c6082014-01-02 12:58:47 -0800341 nr_node_ids * sizeof(struct mem_cgroup_per_node *);
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800342}
343
Glauber Costa510fc4e2012-12-18 14:21:47 -0800344/* internal only representation about the status of kmem accounting. */
345enum {
346 KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */
Glauber Costaa8964b92012-12-18 14:22:09 -0800347 KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */
Glauber Costa7de37682012-12-18 14:22:07 -0800348 KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */
Glauber Costa510fc4e2012-12-18 14:21:47 -0800349};
350
Glauber Costaa8964b92012-12-18 14:22:09 -0800351/* We account when limit is on, but only after call sites are patched */
352#define KMEM_ACCOUNTED_MASK \
353 ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED))
Glauber Costa510fc4e2012-12-18 14:21:47 -0800354
355#ifdef CONFIG_MEMCG_KMEM
356static inline void memcg_kmem_set_active(struct mem_cgroup *memcg)
357{
358 set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
359}
Glauber Costa7de37682012-12-18 14:22:07 -0800360
361static bool memcg_kmem_is_active(struct mem_cgroup *memcg)
362{
363 return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags);
364}
365
Glauber Costaa8964b92012-12-18 14:22:09 -0800366static void memcg_kmem_set_activated(struct mem_cgroup *memcg)
367{
368 set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
369}
370
Glauber Costa55007d82012-12-18 14:22:38 -0800371static void memcg_kmem_clear_activated(struct mem_cgroup *memcg)
372{
373 clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags);
374}
375
Glauber Costa7de37682012-12-18 14:22:07 -0800376static void memcg_kmem_mark_dead(struct mem_cgroup *memcg)
377{
Li Zefan10d5ebf2013-07-08 16:00:33 -0700378 /*
379 * Our caller must use css_get() first, because memcg_uncharge_kmem()
380 * will call css_put() if it sees the memcg is dead.
381 */
382 smp_wmb();
Glauber Costa7de37682012-12-18 14:22:07 -0800383 if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags))
384 set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags);
385}
386
387static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg)
388{
389 return test_and_clear_bit(KMEM_ACCOUNTED_DEAD,
390 &memcg->kmem_account_flags);
391}
Glauber Costa510fc4e2012-12-18 14:21:47 -0800392#endif
393
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800394/* Stuffs for move charges at task migration. */
395/*
Glauber Costaee5e8472013-02-22 16:34:50 -0800396 * Types of charges to be moved. "move_charge_at_immitgrate" and
397 * "immigrate_flags" are treated as a left-shifted bitmap of these types.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800398 */
399enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800400 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700401 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800402 NR_MOVE_TYPE,
403};
404
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800405/* "mc" and its members are protected by cgroup_mutex */
406static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800407 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800408 struct mem_cgroup *from;
409 struct mem_cgroup *to;
Glauber Costaee5e8472013-02-22 16:34:50 -0800410 unsigned long immigrate_flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800411 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800412 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800413 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800414 struct task_struct *moving_task; /* a task moving charges */
415 wait_queue_head_t waitq; /* a waitq for other context */
416} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700417 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800418 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
419};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800420
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700421static bool move_anon(void)
422{
Glauber Costaee5e8472013-02-22 16:34:50 -0800423 return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags);
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700424}
425
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700426static bool move_file(void)
427{
Glauber Costaee5e8472013-02-22 16:34:50 -0800428 return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags);
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700429}
430
Balbir Singh4e416952009-09-23 15:56:39 -0700431/*
432 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
433 * limit reclaim to prevent infinite loops, if they ever occur.
434 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700435#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700436#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700437
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800438enum charge_type {
439 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700440 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800441 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700442 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700443 NR_CHARGE_TYPE,
444};
445
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800446/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800447enum res_type {
448 _MEM,
449 _MEMSWAP,
450 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800451 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800452};
453
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700454#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
455#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800456#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700457/* Used for OOM nofiier */
458#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800459
Balbir Singh75822b42009-09-23 15:56:38 -0700460/*
461 * Reclaim flags for mem_cgroup_hierarchical_reclaim
462 */
463#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
464#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
465#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
466#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
467
Glauber Costa09998212013-02-22 16:34:55 -0800468/*
469 * The memcg_create_mutex will be held whenever a new cgroup is created.
470 * As a consequence, any change that needs to protect against new child cgroups
471 * appearing has to hold it as well.
472 */
473static DEFINE_MUTEX(memcg_create_mutex);
474
Wanpeng Lib2145142012-07-31 16:46:01 -0700475struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
476{
Tejun Heoa7c6d552013-08-08 20:11:23 -0400477 return s ? container_of(s, struct mem_cgroup, css) : NULL;
Wanpeng Lib2145142012-07-31 16:46:01 -0700478}
479
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700480/* Some nice accessors for the vmpressure. */
481struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
482{
483 if (!memcg)
484 memcg = root_mem_cgroup;
485 return &memcg->vmpressure;
486}
487
488struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
489{
490 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
491}
492
493struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css)
494{
495 return &mem_cgroup_from_css(css)->vmpressure;
496}
497
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700498static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
499{
500 return (memcg == root_mem_cgroup);
501}
502
Li Zefan4219b2d2013-09-23 16:56:29 +0800503/*
504 * We restrict the id in the range of [1, 65535], so it can fit into
505 * an unsigned short.
506 */
507#define MEM_CGROUP_ID_MAX USHRT_MAX
508
Li Zefan34c00c32013-09-23 16:56:01 +0800509static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
510{
511 /*
512 * The ID of the root cgroup is 0, but memcg treat 0 as an
513 * invalid ID, so we return (cgroup_id + 1).
514 */
515 return memcg->css.cgroup->id + 1;
516}
517
518static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
519{
520 struct cgroup_subsys_state *css;
521
522 css = css_from_id(id - 1, &mem_cgroup_subsys);
523 return mem_cgroup_from_css(css);
524}
525
Glauber Costae1aab162011-12-11 21:47:03 +0000526/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700527#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000528
Glauber Costae1aab162011-12-11 21:47:03 +0000529void sock_update_memcg(struct sock *sk)
530{
Glauber Costa376be5f2012-01-20 04:57:14 +0000531 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000532 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700533 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000534
535 BUG_ON(!sk->sk_prot->proto_cgroup);
536
Glauber Costaf3f511e2012-01-05 20:16:39 +0000537 /* Socket cloning can throw us here with sk_cgrp already
538 * filled. It won't however, necessarily happen from
539 * process context. So the test for root memcg given
540 * the current task's memcg won't help us in this case.
541 *
542 * Respecting the original socket's memcg is a better
543 * decision in this case.
544 */
545 if (sk->sk_cgrp) {
546 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700547 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000548 return;
549 }
550
Glauber Costae1aab162011-12-11 21:47:03 +0000551 rcu_read_lock();
552 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700553 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Li Zefan5347e5a2013-07-08 16:00:30 -0700554 if (!mem_cgroup_is_root(memcg) &&
555 memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700556 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000557 }
558 rcu_read_unlock();
559 }
560}
561EXPORT_SYMBOL(sock_update_memcg);
562
563void sock_release_memcg(struct sock *sk)
564{
Glauber Costa376be5f2012-01-20 04:57:14 +0000565 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000566 struct mem_cgroup *memcg;
567 WARN_ON(!sk->sk_cgrp->memcg);
568 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700569 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000570 }
571}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000572
573struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
574{
575 if (!memcg || mem_cgroup_is_root(memcg))
576 return NULL;
577
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700578 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000579}
580EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000581
Glauber Costa3f134612012-05-29 15:07:11 -0700582static void disarm_sock_keys(struct mem_cgroup *memcg)
583{
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700584 if (!memcg_proto_activated(&memcg->tcp_mem))
Glauber Costa3f134612012-05-29 15:07:11 -0700585 return;
586 static_key_slow_dec(&memcg_socket_limit_enabled);
587}
588#else
589static void disarm_sock_keys(struct mem_cgroup *memcg)
590{
591}
592#endif
593
Glauber Costaa8964b92012-12-18 14:22:09 -0800594#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800595/*
596 * This will be the memcg's index in each cache's ->memcg_params->memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800597 * The main reason for not using cgroup id for this:
598 * this works better in sparse environments, where we have a lot of memcgs,
599 * but only a few kmem-limited. Or also, if we have, for instance, 200
600 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
601 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800602 *
603 * The current size of the caches array is stored in
604 * memcg_limited_groups_array_size. It will double each time we have to
605 * increase it.
606 */
607static DEFINE_IDA(kmem_limited_groups);
Glauber Costa749c5412012-12-18 14:23:01 -0800608int memcg_limited_groups_array_size;
609
Glauber Costa55007d82012-12-18 14:22:38 -0800610/*
611 * MIN_SIZE is different than 1, because we would like to avoid going through
612 * the alloc/free process all the time. In a small machine, 4 kmem-limited
613 * cgroups is a reasonable guess. In the future, it could be a parameter or
614 * tunable, but that is strictly not necessary.
615 *
Li Zefanb8627832013-09-23 16:56:47 +0800616 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800617 * this constant directly from cgroup, but it is understandable that this is
618 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800619 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800620 * increase ours as well if it increases.
621 */
622#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800623#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800624
Glauber Costad7f25f82012-12-18 14:22:40 -0800625/*
626 * A lot of the calls to the cache allocation functions are expected to be
627 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
628 * conditional to this static branch, we'll have to allow modules that does
629 * kmem_cache_alloc and the such to see this symbol as well
630 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800631struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800632EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800633
634static void disarm_kmem_keys(struct mem_cgroup *memcg)
635{
Glauber Costa55007d82012-12-18 14:22:38 -0800636 if (memcg_kmem_is_active(memcg)) {
Glauber Costaa8964b92012-12-18 14:22:09 -0800637 static_key_slow_dec(&memcg_kmem_enabled_key);
Glauber Costa55007d82012-12-18 14:22:38 -0800638 ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id);
639 }
Glauber Costabea207c2012-12-18 14:22:11 -0800640 /*
641 * This check can't live in kmem destruction function,
642 * since the charges will outlive the cgroup
643 */
644 WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0);
Glauber Costaa8964b92012-12-18 14:22:09 -0800645}
646#else
647static void disarm_kmem_keys(struct mem_cgroup *memcg)
648{
649}
650#endif /* CONFIG_MEMCG_KMEM */
651
652static void disarm_static_keys(struct mem_cgroup *memcg)
653{
654 disarm_sock_keys(memcg);
655 disarm_kmem_keys(memcg);
656}
657
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700658static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800659
Balbir Singhf64c3f52009-09-23 15:56:37 -0700660static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700661mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700662{
Glauber Costa45cf7eb2013-02-22 16:34:49 -0800663 VM_BUG_ON((unsigned)nid >= nr_node_ids);
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700664 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700665}
666
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700667struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100668{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700669 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100670}
671
Balbir Singhf64c3f52009-09-23 15:56:37 -0700672static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700673page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700674{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700675 int nid = page_to_nid(page);
676 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700677
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700678 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700679}
680
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700681static struct mem_cgroup_tree_per_zone *
682soft_limit_tree_node_zone(int nid, int zid)
683{
684 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
685}
686
687static struct mem_cgroup_tree_per_zone *
688soft_limit_tree_from_page(struct page *page)
689{
690 int nid = page_to_nid(page);
691 int zid = page_zonenum(page);
692
693 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
694}
695
696static void
697__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
698 struct mem_cgroup_per_zone *mz,
699 struct mem_cgroup_tree_per_zone *mctz,
700 unsigned long long new_usage_in_excess)
701{
702 struct rb_node **p = &mctz->rb_root.rb_node;
703 struct rb_node *parent = NULL;
704 struct mem_cgroup_per_zone *mz_node;
705
706 if (mz->on_tree)
707 return;
708
709 mz->usage_in_excess = new_usage_in_excess;
710 if (!mz->usage_in_excess)
711 return;
712 while (*p) {
713 parent = *p;
714 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
715 tree_node);
716 if (mz->usage_in_excess < mz_node->usage_in_excess)
717 p = &(*p)->rb_left;
718 /*
719 * We can't avoid mem cgroups that are over their soft
720 * limit by the same amount
721 */
722 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
723 p = &(*p)->rb_right;
724 }
725 rb_link_node(&mz->tree_node, parent, p);
726 rb_insert_color(&mz->tree_node, &mctz->rb_root);
727 mz->on_tree = true;
728}
729
730static void
731__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
732 struct mem_cgroup_per_zone *mz,
733 struct mem_cgroup_tree_per_zone *mctz)
734{
735 if (!mz->on_tree)
736 return;
737 rb_erase(&mz->tree_node, &mctz->rb_root);
738 mz->on_tree = false;
739}
740
741static void
742mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
743 struct mem_cgroup_per_zone *mz,
744 struct mem_cgroup_tree_per_zone *mctz)
745{
746 spin_lock(&mctz->lock);
747 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
748 spin_unlock(&mctz->lock);
749}
750
751
752static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
753{
754 unsigned long long excess;
755 struct mem_cgroup_per_zone *mz;
756 struct mem_cgroup_tree_per_zone *mctz;
757 int nid = page_to_nid(page);
758 int zid = page_zonenum(page);
759 mctz = soft_limit_tree_from_page(page);
760
761 /*
762 * Necessary to update all ancestors when hierarchy is used.
763 * because their event counter is not touched.
764 */
765 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
766 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
767 excess = res_counter_soft_limit_excess(&memcg->res);
768 /*
769 * We have to update the tree if mz is on RB-tree or
770 * mem is over its softlimit.
771 */
772 if (excess || mz->on_tree) {
773 spin_lock(&mctz->lock);
774 /* if on-tree, remove it */
775 if (mz->on_tree)
776 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
777 /*
778 * Insert again. mz->usage_in_excess will be updated.
779 * If excess is 0, no tree ops.
780 */
781 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
782 spin_unlock(&mctz->lock);
783 }
784 }
785}
786
787static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
788{
789 int node, zone;
790 struct mem_cgroup_per_zone *mz;
791 struct mem_cgroup_tree_per_zone *mctz;
792
793 for_each_node(node) {
794 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
795 mz = mem_cgroup_zoneinfo(memcg, node, zone);
796 mctz = soft_limit_tree_node_zone(node, zone);
797 mem_cgroup_remove_exceeded(memcg, mz, mctz);
798 }
799 }
800}
801
802static struct mem_cgroup_per_zone *
803__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
804{
805 struct rb_node *rightmost = NULL;
806 struct mem_cgroup_per_zone *mz;
807
808retry:
809 mz = NULL;
810 rightmost = rb_last(&mctz->rb_root);
811 if (!rightmost)
812 goto done; /* Nothing to reclaim from */
813
814 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
815 /*
816 * Remove the node now but someone else can add it back,
817 * we will to add it back at the end of reclaim to its correct
818 * position in the tree.
819 */
820 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
821 if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
822 !css_tryget(&mz->memcg->css))
823 goto retry;
824done:
825 return mz;
826}
827
828static struct mem_cgroup_per_zone *
829mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
830{
831 struct mem_cgroup_per_zone *mz;
832
833 spin_lock(&mctz->lock);
834 mz = __mem_cgroup_largest_soft_limit_node(mctz);
835 spin_unlock(&mctz->lock);
836 return mz;
837}
838
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700839/*
840 * Implementation Note: reading percpu statistics for memcg.
841 *
842 * Both of vmstat[] and percpu_counter has threshold and do periodic
843 * synchronization to implement "quick" read. There are trade-off between
844 * reading cost and precision of value. Then, we may have a chance to implement
845 * a periodic synchronizion of counter in memcg's counter.
846 *
847 * But this _read() function is used for user interface now. The user accounts
848 * memory usage by memory cgroup and he _always_ requires exact value because
849 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
850 * have to visit all online cpus and make sum. So, for now, unnecessary
851 * synchronization is not implemented. (just implemented for cpu hotplug)
852 *
853 * If there are kernel internal actions which can make use of some not-exact
854 * value, and reading all cpu value can be performance bottleneck in some
855 * common workload, threashold and synchonization as vmstat[] should be
856 * implemented.
857 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700858static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700859 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800860{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700861 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800862 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800863
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700864 get_online_cpus();
865 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700866 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700867#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700868 spin_lock(&memcg->pcp_counter_lock);
869 val += memcg->nocpu_base.count[idx];
870 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700871#endif
872 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800873 return val;
874}
875
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700876static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700877 bool charge)
878{
879 int val = (charge) ? 1 : -1;
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -0700880 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700881}
882
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700883static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700884 enum mem_cgroup_events_index idx)
885{
886 unsigned long val = 0;
887 int cpu;
888
David Rientjes9c567512013-10-16 13:46:43 -0700889 get_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700890 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700891 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700892#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700893 spin_lock(&memcg->pcp_counter_lock);
894 val += memcg->nocpu_base.events[idx];
895 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700896#endif
David Rientjes9c567512013-10-16 13:46:43 -0700897 put_online_cpus();
Johannes Weinere9f89742011-03-23 16:42:37 -0700898 return val;
899}
900
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700901static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700902 struct page *page,
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700903 bool anon, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800904{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800905 preempt_disable();
906
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700907 /*
908 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
909 * counted as CACHE even if it's on ANON LRU.
910 */
911 if (anon)
912 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700913 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800914 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700915 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700916 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700917
David Rientjesb070e652013-05-07 16:18:09 -0700918 if (PageTransHuge(page))
919 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
920 nr_pages);
921
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800922 /* pagein of a big page is an event. So, ignore page size */
923 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700924 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800925 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700926 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800927 nr_pages = -nr_pages; /* for event */
928 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800929
Johannes Weiner13114712012-05-29 15:07:07 -0700930 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800931
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800932 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800933}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800934
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700935unsigned long
Hugh Dickins4d7dcca2012-05-29 15:07:08 -0700936mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
Konstantin Khlebnikov074291f2012-05-29 15:07:00 -0700937{
938 struct mem_cgroup_per_zone *mz;
939
940 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
941 return mz->lru_size[lru];
942}
943
944static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700945mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700946 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700947{
948 struct mem_cgroup_per_zone *mz;
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700949 enum lru_list lru;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700950 unsigned long ret = 0;
951
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700952 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700953
Hugh Dickinsf156ab92012-03-21 16:34:19 -0700954 for_each_lru(lru) {
955 if (BIT(lru) & lru_mask)
956 ret += mz->lru_size[lru];
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700957 }
958 return ret;
959}
960
961static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700962mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700963 int nid, unsigned int lru_mask)
964{
Ying Han889976d2011-05-26 16:25:33 -0700965 u64 total = 0;
966 int zid;
967
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700968 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700969 total += mem_cgroup_zone_nr_lru_pages(memcg,
970 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700971
Ying Han889976d2011-05-26 16:25:33 -0700972 return total;
973}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700974
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700975static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700976 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800977{
Ying Han889976d2011-05-26 16:25:33 -0700978 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800979 u64 total = 0;
980
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800981 for_each_node_state(nid, N_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700982 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800983 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800984}
985
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800986static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
987 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800988{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700989 unsigned long val, next;
990
Johannes Weiner13114712012-05-29 15:07:07 -0700991 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700992 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700993 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800994 if ((long)next - (long)val < 0) {
995 switch (target) {
996 case MEM_CGROUP_TARGET_THRESH:
997 next = val + THRESHOLDS_EVENTS_TARGET;
998 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700999 case MEM_CGROUP_TARGET_SOFTLIMIT:
1000 next = val + SOFTLIMIT_EVENTS_TARGET;
1001 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001002 case MEM_CGROUP_TARGET_NUMAINFO:
1003 next = val + NUMAINFO_EVENTS_TARGET;
1004 break;
1005 default:
1006 break;
1007 }
1008 __this_cpu_write(memcg->stat->targets[target], next);
1009 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07001010 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001011 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001012}
1013
1014/*
1015 * Check events in order.
1016 *
1017 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001018static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001019{
Steven Rostedt47994012011-11-02 13:38:33 -07001020 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001021 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001022 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1023 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001024 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -08001025 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001026
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001027 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1028 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001029#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001030 do_numainfo = mem_cgroup_event_ratelimit(memcg,
1031 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001032#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001033 preempt_enable();
1034
1035 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001036 if (unlikely(do_softlimit))
1037 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -08001038#if MAX_NUMNODES > 1
1039 if (unlikely(do_numainfo))
1040 atomic_inc(&memcg->numainfo_events);
1041#endif
1042 } else
1043 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001044}
1045
Balbir Singhcf475ad2008-04-29 01:00:16 -07001046struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001047{
Balbir Singh31a78f22008-09-28 23:09:31 +01001048 /*
1049 * mm_update_next_owner() may clear mm->owner to NULL
1050 * if it races with swapoff, page migration, etc.
1051 * So this can be called with p == NULL.
1052 */
1053 if (unlikely(!p))
1054 return NULL;
1055
Tejun Heo8af01f52013-08-08 20:11:22 -04001056 return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001057}
1058
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07001059struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001060{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001061 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001062
1063 if (!mm)
1064 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001065 /*
1066 * Because we have no locks, mm->owner's may be being moved to other
1067 * cgroup. We use css_tryget() here even if this looks
1068 * pessimistic (rather than adding locks here).
1069 */
1070 rcu_read_lock();
1071 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001072 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1073 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001074 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001075 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001076 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001077 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001078}
1079
Michal Hocko16248d82013-04-29 15:07:19 -07001080/*
1081 * Returns a next (in a pre-order walk) alive memcg (with elevated css
1082 * ref. count) or NULL if the whole root's subtree has been visited.
1083 *
1084 * helper function to be used by mem_cgroup_iter
1085 */
1086static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root,
Andrew Morton694fbc02013-09-24 15:27:37 -07001087 struct mem_cgroup *last_visited)
Michal Hocko16248d82013-04-29 15:07:19 -07001088{
Tejun Heo492eb212013-08-08 20:11:25 -04001089 struct cgroup_subsys_state *prev_css, *next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001090
Tejun Heobd8815a2013-08-08 20:11:27 -04001091 prev_css = last_visited ? &last_visited->css : NULL;
Michal Hocko16248d82013-04-29 15:07:19 -07001092skip_node:
Tejun Heo492eb212013-08-08 20:11:25 -04001093 next_css = css_next_descendant_pre(prev_css, &root->css);
Michal Hocko16248d82013-04-29 15:07:19 -07001094
1095 /*
1096 * Even if we found a group we have to make sure it is
1097 * alive. css && !memcg means that the groups should be
1098 * skipped and we should continue the tree walk.
1099 * last_visited css is safe to use because it is
1100 * protected by css_get and the tree walk is rcu safe.
1101 */
Tejun Heo492eb212013-08-08 20:11:25 -04001102 if (next_css) {
1103 struct mem_cgroup *mem = mem_cgroup_from_css(next_css);
1104
Andrew Morton694fbc02013-09-24 15:27:37 -07001105 if (css_tryget(&mem->css))
1106 return mem;
1107 else {
Tejun Heo492eb212013-08-08 20:11:25 -04001108 prev_css = next_css;
Michal Hocko16248d82013-04-29 15:07:19 -07001109 goto skip_node;
1110 }
1111 }
1112
1113 return NULL;
1114}
1115
Johannes Weiner519ebea2013-07-03 15:04:51 -07001116static void mem_cgroup_iter_invalidate(struct mem_cgroup *root)
1117{
1118 /*
1119 * When a group in the hierarchy below root is destroyed, the
1120 * hierarchy iterator can no longer be trusted since it might
1121 * have pointed to the destroyed group. Invalidate it.
1122 */
1123 atomic_inc(&root->dead_count);
1124}
1125
1126static struct mem_cgroup *
1127mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter,
1128 struct mem_cgroup *root,
1129 int *sequence)
1130{
1131 struct mem_cgroup *position = NULL;
1132 /*
1133 * A cgroup destruction happens in two stages: offlining and
1134 * release. They are separated by a RCU grace period.
1135 *
1136 * If the iterator is valid, we may still race with an
1137 * offlining. The RCU lock ensures the object won't be
1138 * released, tryget will fail if we lost the race.
1139 */
1140 *sequence = atomic_read(&root->dead_count);
1141 if (iter->last_dead_count == *sequence) {
1142 smp_rmb();
1143 position = iter->last_visited;
1144 if (position && !css_tryget(&position->css))
1145 position = NULL;
1146 }
1147 return position;
1148}
1149
1150static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter,
1151 struct mem_cgroup *last_visited,
1152 struct mem_cgroup *new_position,
1153 int sequence)
1154{
1155 if (last_visited)
1156 css_put(&last_visited->css);
1157 /*
1158 * We store the sequence count from the time @last_visited was
1159 * loaded successfully instead of rereading it here so that we
1160 * don't lose destruction events in between. We could have
1161 * raced with the destruction of @new_position after all.
1162 */
1163 iter->last_visited = new_position;
1164 smp_wmb();
1165 iter->last_dead_count = sequence;
1166}
1167
Johannes Weiner56600482012-01-12 17:17:59 -08001168/**
1169 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1170 * @root: hierarchy root
1171 * @prev: previously returned memcg, NULL on first invocation
1172 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1173 *
1174 * Returns references to children of the hierarchy below @root, or
1175 * @root itself, or %NULL after a full round-trip.
1176 *
1177 * Caller must pass the return value in @prev on subsequent
1178 * invocations for reference counting, or use mem_cgroup_iter_break()
1179 * to cancel a hierarchy walk before the round-trip is complete.
1180 *
1181 * Reclaimers can specify a zone and a priority level in @reclaim to
1182 * divide up the memcgs in the hierarchy among all concurrent
1183 * reclaimers operating on the same zone and priority.
1184 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001185struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001186 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001187 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001188{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001189 struct mem_cgroup *memcg = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001190 struct mem_cgroup *last_visited = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001191
Andrew Morton694fbc02013-09-24 15:27:37 -07001192 if (mem_cgroup_disabled())
1193 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001194
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001195 if (!root)
1196 root = root_mem_cgroup;
1197
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001198 if (prev && !reclaim)
Michal Hocko542f85f2013-04-29 15:07:15 -07001199 last_visited = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001200
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001201 if (!root->use_hierarchy && root != root_mem_cgroup) {
1202 if (prev)
Michal Hockoc40046f2013-04-29 15:07:14 -07001203 goto out_css_put;
Andrew Morton694fbc02013-09-24 15:27:37 -07001204 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001205 }
1206
Michal Hocko542f85f2013-04-29 15:07:15 -07001207 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001208 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001209 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner519ebea2013-07-03 15:04:51 -07001210 int uninitialized_var(seq);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001211
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001212 if (reclaim) {
1213 int nid = zone_to_nid(reclaim->zone);
1214 int zid = zone_idx(reclaim->zone);
1215 struct mem_cgroup_per_zone *mz;
1216
1217 mz = mem_cgroup_zoneinfo(root, nid, zid);
1218 iter = &mz->reclaim_iter[reclaim->priority];
Michal Hocko542f85f2013-04-29 15:07:15 -07001219 if (prev && reclaim->generation != iter->generation) {
Michal Hocko5f578162013-04-29 15:07:17 -07001220 iter->last_visited = NULL;
Michal Hocko542f85f2013-04-29 15:07:15 -07001221 goto out_unlock;
1222 }
Michal Hocko5f578162013-04-29 15:07:17 -07001223
Johannes Weiner519ebea2013-07-03 15:04:51 -07001224 last_visited = mem_cgroup_iter_load(iter, root, &seq);
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001225 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001226
Andrew Morton694fbc02013-09-24 15:27:37 -07001227 memcg = __mem_cgroup_iter_next(root, last_visited);
Michal Hocko542f85f2013-04-29 15:07:15 -07001228
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001229 if (reclaim) {
Johannes Weiner519ebea2013-07-03 15:04:51 -07001230 mem_cgroup_iter_update(iter, last_visited, memcg, seq);
Michal Hocko542f85f2013-04-29 15:07:15 -07001231
Michal Hocko19f39402013-04-29 15:07:18 -07001232 if (!memcg)
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001233 iter->generation++;
1234 else if (!prev && memcg)
1235 reclaim->generation = iter->generation;
1236 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001237
Andrew Morton694fbc02013-09-24 15:27:37 -07001238 if (prev && !memcg)
Michal Hocko542f85f2013-04-29 15:07:15 -07001239 goto out_unlock;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001240 }
Michal Hocko542f85f2013-04-29 15:07:15 -07001241out_unlock:
1242 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001243out_css_put:
1244 if (prev && prev != root)
1245 css_put(&prev->css);
1246
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001247 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001248}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001249
Johannes Weiner56600482012-01-12 17:17:59 -08001250/**
1251 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1252 * @root: hierarchy root
1253 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1254 */
1255void mem_cgroup_iter_break(struct mem_cgroup *root,
1256 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001257{
1258 if (!root)
1259 root = root_mem_cgroup;
1260 if (prev && prev != root)
1261 css_put(&prev->css);
1262}
1263
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001264/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001265 * Iteration constructs for visiting all cgroups (under a tree). If
1266 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1267 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001268 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001269#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001270 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001271 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001272 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001273
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001274#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001275 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001276 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001277 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001278
David Rientjes68ae5642012-12-12 13:51:57 -08001279void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
Ying Han456f9982011-05-26 16:25:38 -07001280{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001281 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -07001282
Ying Han456f9982011-05-26 16:25:38 -07001283 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001284 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1285 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -07001286 goto out;
1287
1288 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -07001289 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -08001290 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1291 break;
1292 case PGMAJFAULT:
1293 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -07001294 break;
1295 default:
1296 BUG();
1297 }
1298out:
1299 rcu_read_unlock();
1300}
David Rientjes68ae5642012-12-12 13:51:57 -08001301EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
Ying Han456f9982011-05-26 16:25:38 -07001302
Johannes Weiner925b7672012-01-12 17:18:15 -08001303/**
1304 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1305 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001306 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001307 *
1308 * Returns the lru list vector holding pages for the given @zone and
1309 * @mem. This can be the global zone lruvec, if the memory controller
1310 * is disabled.
1311 */
1312struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1313 struct mem_cgroup *memcg)
1314{
1315 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001316 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001317
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001318 if (mem_cgroup_disabled()) {
1319 lruvec = &zone->lruvec;
1320 goto out;
1321 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001322
1323 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001324 lruvec = &mz->lruvec;
1325out:
1326 /*
1327 * Since a node can be onlined after the mem_cgroup was created,
1328 * we have to be prepared to initialize lruvec->zone here;
1329 * and if offlined then reonlined, we need to reinitialize it.
1330 */
1331 if (unlikely(lruvec->zone != zone))
1332 lruvec->zone = zone;
1333 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001334}
1335
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001336/*
1337 * Following LRU functions are allowed to be used without PCG_LOCK.
1338 * Operations are called by routine of global LRU independently from memcg.
1339 * What we have to take care of here is validness of pc->mem_cgroup.
1340 *
1341 * Changes to pc->mem_cgroup happens when
1342 * 1. charge
1343 * 2. moving account
1344 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1345 * It is added to LRU before charge.
1346 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1347 * When moving account, the page is not on LRU. It's isolated.
1348 */
1349
Johannes Weiner925b7672012-01-12 17:18:15 -08001350/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001351 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
Johannes Weiner925b7672012-01-12 17:18:15 -08001352 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001353 * @zone: zone of the page
Minchan Kim3f58a822011-03-22 16:32:53 -07001354 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001355struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001356{
1357 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001358 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001359 struct page_cgroup *pc;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001360 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001361
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001362 if (mem_cgroup_disabled()) {
1363 lruvec = &zone->lruvec;
1364 goto out;
1365 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001366
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001367 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001368 memcg = pc->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001369
1370 /*
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001371 * Surreptitiously switch any uncharged offlist page to root:
Hugh Dickins75121022012-03-05 14:59:18 -08001372 * an uncharged page off lru does nothing to secure
1373 * its former mem_cgroup from sudden removal.
1374 *
1375 * Our caller holds lru_lock, and PageCgroupUsed is updated
1376 * under page_cgroup lock: between them, they make all uses
1377 * of pc->mem_cgroup safe.
1378 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001379 if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
Hugh Dickins75121022012-03-05 14:59:18 -08001380 pc->mem_cgroup = memcg = root_mem_cgroup;
1381
Johannes Weiner925b7672012-01-12 17:18:15 -08001382 mz = page_cgroup_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001383 lruvec = &mz->lruvec;
1384out:
1385 /*
1386 * Since a node can be onlined after the mem_cgroup was created,
1387 * we have to be prepared to initialize lruvec->zone here;
1388 * and if offlined then reonlined, we need to reinitialize it.
1389 */
1390 if (unlikely(lruvec->zone != zone))
1391 lruvec->zone = zone;
1392 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001393}
1394
1395/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001396 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1397 * @lruvec: mem_cgroup per zone lru vector
1398 * @lru: index of lru list the page is sitting on
1399 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001400 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001401 * This function must be called when a page is added to or removed from an
1402 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001403 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001404void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1405 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001406{
1407 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001408 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001409
1410 if (mem_cgroup_disabled())
1411 return;
1412
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001413 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1414 lru_size = mz->lru_size + lru;
1415 *lru_size += nr_pages;
1416 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001417}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001418
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001419/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001420 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001421 * hierarchy subtree
1422 */
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001423bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1424 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001425{
Johannes Weiner91c637342012-05-29 15:06:24 -07001426 if (root_memcg == memcg)
1427 return true;
Hugh Dickins3a981f42012-06-20 12:52:58 -07001428 if (!root_memcg->use_hierarchy || !memcg)
Johannes Weiner91c637342012-05-29 15:06:24 -07001429 return false;
Li Zefanb47f77b2013-09-23 16:55:43 +08001430 return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup);
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001431}
1432
1433static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1434 struct mem_cgroup *memcg)
1435{
1436 bool ret;
1437
Johannes Weiner91c637342012-05-29 15:06:24 -07001438 rcu_read_lock();
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001439 ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
Johannes Weiner91c637342012-05-29 15:06:24 -07001440 rcu_read_unlock();
1441 return ret;
Michal Hocko3e920412011-07-26 16:08:29 -07001442}
1443
David Rientjesffbdccf2013-07-03 15:01:23 -07001444bool task_in_mem_cgroup(struct task_struct *task,
1445 const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001446{
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001447 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001448 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001449 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001450
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001451 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001452 if (p) {
1453 curr = try_get_mem_cgroup_from_mm(p->mm);
1454 task_unlock(p);
1455 } else {
1456 /*
1457 * All threads may have already detached their mm's, but the oom
1458 * killer still needs to detect if they have already been oom
1459 * killed to prevent needlessly killing additional tasks.
1460 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001461 rcu_read_lock();
David Rientjesde077d22012-01-12 17:18:52 -08001462 curr = mem_cgroup_from_task(task);
1463 if (curr)
1464 css_get(&curr->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001465 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001466 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001467 if (!curr)
David Rientjesffbdccf2013-07-03 15:01:23 -07001468 return false;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001469 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001470 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001471 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001472 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1473 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001474 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001475 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001476 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001477 return ret;
1478}
1479
Konstantin Khlebnikovc56d5c72012-05-29 15:07:00 -07001480int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001481{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001482 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001483 unsigned long inactive;
1484 unsigned long active;
1485 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001486
Hugh Dickins4d7dcca2012-05-29 15:07:08 -07001487 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1488 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001489
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001490 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1491 if (gb)
1492 inactive_ratio = int_sqrt(10 * gb);
1493 else
1494 inactive_ratio = 1;
1495
Johannes Weiner9b272972011-11-02 13:38:23 -07001496 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001497}
1498
Balbir Singh6d61ef42009-01-07 18:08:06 -08001499#define mem_cgroup_from_res_counter(counter, member) \
1500 container_of(counter, struct mem_cgroup, member)
1501
Johannes Weiner19942822011-02-01 15:52:43 -08001502/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001503 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001504 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001505 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001506 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001507 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001508 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001509static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001510{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001511 unsigned long long margin;
1512
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001513 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001514 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001515 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001516 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001517}
1518
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001519int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001520{
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001521 /* root ? */
Tejun Heo63876982013-08-08 20:11:23 -04001522 if (!css_parent(&memcg->css))
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001523 return vm_swappiness;
1524
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001525 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001526}
1527
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001528/*
1529 * memcg->moving_account is used for checking possibility that some thread is
1530 * calling move_account(). When a thread on CPU-A starts moving pages under
1531 * a memcg, other threads should check memcg->moving_account under
1532 * rcu_read_lock(), like this:
1533 *
1534 * CPU-A CPU-B
1535 * rcu_read_lock()
1536 * memcg->moving_account+1 if (memcg->mocing_account)
1537 * take heavy locks.
1538 * synchronize_rcu() update something.
1539 * rcu_read_unlock()
1540 * start move here.
1541 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001542
1543/* for quick checking without looking up memcg */
1544atomic_t memcg_moving __read_mostly;
1545
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001546static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001547{
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001548 atomic_inc(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001549 atomic_inc(&memcg->moving_account);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001550 synchronize_rcu();
1551}
1552
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001553static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001554{
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001555 /*
1556 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1557 * We check NULL in callee rather than caller.
1558 */
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001559 if (memcg) {
1560 atomic_dec(&memcg_moving);
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001561 atomic_dec(&memcg->moving_account);
KAMEZAWA Hiroyuki4331f7d2012-03-21 16:34:26 -07001562 }
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001563}
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001564
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001565/*
1566 * 2 routines for checking "mem" is under move_account() or not.
1567 *
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001568 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1569 * is used for avoiding races in accounting. If true,
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001570 * pc->mem_cgroup may be overwritten.
1571 *
1572 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1573 * under hierarchy of moving cgroups. This is for
1574 * waiting at hith-memory prressure caused by "move".
1575 */
1576
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001577static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001578{
1579 VM_BUG_ON(!rcu_read_lock_held());
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001580 return atomic_read(&memcg->moving_account) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001581}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001582
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001584{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001585 struct mem_cgroup *from;
1586 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001587 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001588 /*
1589 * Unlike task_move routines, we access mc.to, mc.from not under
1590 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1591 */
1592 spin_lock(&mc.lock);
1593 from = mc.from;
1594 to = mc.to;
1595 if (!from)
1596 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001598 ret = mem_cgroup_same_or_subtree(memcg, from)
1599 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001600unlock:
1601 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001602 return ret;
1603}
1604
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001605static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001606{
1607 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001608 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001609 DEFINE_WAIT(wait);
1610 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1611 /* moving charge context might have finished. */
1612 if (mc.moving_task)
1613 schedule();
1614 finish_wait(&mc.waitq, &wait);
1615 return true;
1616 }
1617 }
1618 return false;
1619}
1620
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001621/*
1622 * Take this lock when
1623 * - a code tries to modify page's memcg while it's USED.
1624 * - a code tries to modify page state accounting in a memcg.
Andrew Morton13fd1dd92012-03-21 16:34:26 -07001625 * see mem_cgroup_stolen(), too.
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07001626 */
1627static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1628 unsigned long *flags)
1629{
1630 spin_lock_irqsave(&memcg->move_lock, *flags);
1631}
1632
1633static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1634 unsigned long *flags)
1635{
1636 spin_unlock_irqrestore(&memcg->move_lock, *flags);
1637}
1638
Sha Zhengju58cf1882013-02-22 16:32:05 -08001639#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001640/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001641 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001642 * @memcg: The memory cgroup that went over limit
1643 * @p: Task that is going to be killed
1644 *
1645 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1646 * enabled
1647 */
1648void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1649{
1650 struct cgroup *task_cgrp;
1651 struct cgroup *mem_cgrp;
1652 /*
1653 * Need a buffer in BSS, can't rely on allocations. The code relies
1654 * on the assumption that OOM is serialized for memory controller.
1655 * If this assumption is broken, revisit this code.
1656 */
1657 static char memcg_name[PATH_MAX];
1658 int ret;
Sha Zhengju58cf1882013-02-22 16:32:05 -08001659 struct mem_cgroup *iter;
1660 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001661
Sha Zhengju58cf1882013-02-22 16:32:05 -08001662 if (!p)
Balbir Singhe2224322009-04-02 16:57:39 -07001663 return;
1664
Balbir Singhe2224322009-04-02 16:57:39 -07001665 rcu_read_lock();
1666
1667 mem_cgrp = memcg->css.cgroup;
1668 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1669
1670 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1671 if (ret < 0) {
1672 /*
1673 * Unfortunately, we are unable to convert to a useful name
1674 * But we'll still print out the usage information
1675 */
1676 rcu_read_unlock();
1677 goto done;
1678 }
1679 rcu_read_unlock();
1680
Andrew Mortond0451972013-02-22 16:32:06 -08001681 pr_info("Task in %s killed", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001682
1683 rcu_read_lock();
1684 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1685 if (ret < 0) {
1686 rcu_read_unlock();
1687 goto done;
1688 }
1689 rcu_read_unlock();
1690
1691 /*
1692 * Continues from above, so we don't need an KERN_ level
1693 */
Andrew Mortond0451972013-02-22 16:32:06 -08001694 pr_cont(" as a result of limit of %s\n", memcg_name);
Balbir Singhe2224322009-04-02 16:57:39 -07001695done:
1696
Andrew Mortond0451972013-02-22 16:32:06 -08001697 pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001698 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1699 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1700 res_counter_read_u64(&memcg->res, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001701 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n",
Balbir Singhe2224322009-04-02 16:57:39 -07001702 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1703 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1704 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
Andrew Mortond0451972013-02-22 16:32:06 -08001705 pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n",
Glauber Costa510fc4e2012-12-18 14:21:47 -08001706 res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10,
1707 res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10,
1708 res_counter_read_u64(&memcg->kmem, RES_FAILCNT));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001709
1710 for_each_mem_cgroup_tree(iter, memcg) {
1711 pr_info("Memory cgroup stats");
1712
1713 rcu_read_lock();
1714 ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX);
1715 if (!ret)
1716 pr_cont(" for %s", memcg_name);
1717 rcu_read_unlock();
1718 pr_cont(":");
1719
1720 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1721 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1722 continue;
1723 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1724 K(mem_cgroup_read_stat(iter, i)));
1725 }
1726
1727 for (i = 0; i < NR_LRU_LISTS; i++)
1728 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1729 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1730
1731 pr_cont("\n");
1732 }
Balbir Singhe2224322009-04-02 16:57:39 -07001733}
1734
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001735/*
1736 * This function returns the number of memcg under hierarchy tree. Returns
1737 * 1(self count) if no children.
1738 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001739static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001740{
1741 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001742 struct mem_cgroup *iter;
1743
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001744 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001745 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001746 return num;
1747}
1748
Balbir Singh6d61ef42009-01-07 18:08:06 -08001749/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001750 * Return the memory (and swap, if configured) limit for a memcg.
1751 */
David Rientjes9cbb78b2012-07-31 16:43:44 -07001752static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001753{
1754 u64 limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001755
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001756 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001757
David Rientjesa63d83f2010-08-09 17:19:46 -07001758 /*
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001759 * Do not consider swap space if we cannot swap due to swappiness
David Rientjesa63d83f2010-08-09 17:19:46 -07001760 */
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001761 if (mem_cgroup_swappiness(memcg)) {
1762 u64 memsw;
1763
1764 limit += total_swap_pages << PAGE_SHIFT;
1765 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1766
1767 /*
1768 * If memsw is finite and limits the amount of swap space
1769 * available to this memcg, return that limit.
1770 */
1771 limit = min(limit, memsw);
1772 }
1773
1774 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001775}
1776
David Rientjes19965462012-12-11 16:00:26 -08001777static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1778 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001779{
1780 struct mem_cgroup *iter;
1781 unsigned long chosen_points = 0;
1782 unsigned long totalpages;
1783 unsigned int points = 0;
1784 struct task_struct *chosen = NULL;
1785
David Rientjes876aafb2012-07-31 16:43:48 -07001786 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001787 * If current has a pending SIGKILL or is exiting, then automatically
1788 * select it. The goal is to allow it to allocate so that it may
1789 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001790 */
David Rientjes465adcf2013-04-29 15:08:45 -07001791 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes876aafb2012-07-31 16:43:48 -07001792 set_thread_flag(TIF_MEMDIE);
1793 return;
1794 }
1795
1796 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001797 totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1798 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001799 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001800 struct task_struct *task;
1801
Tejun Heo72ec7022013-08-08 20:11:26 -04001802 css_task_iter_start(&iter->css, &it);
1803 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001804 switch (oom_scan_process_thread(task, totalpages, NULL,
1805 false)) {
1806 case OOM_SCAN_SELECT:
1807 if (chosen)
1808 put_task_struct(chosen);
1809 chosen = task;
1810 chosen_points = ULONG_MAX;
1811 get_task_struct(chosen);
1812 /* fall through */
1813 case OOM_SCAN_CONTINUE:
1814 continue;
1815 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001816 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001817 mem_cgroup_iter_break(memcg, iter);
1818 if (chosen)
1819 put_task_struct(chosen);
1820 return;
1821 case OOM_SCAN_OK:
1822 break;
1823 };
1824 points = oom_badness(task, memcg, NULL, totalpages);
1825 if (points > chosen_points) {
1826 if (chosen)
1827 put_task_struct(chosen);
1828 chosen = task;
1829 chosen_points = points;
1830 get_task_struct(chosen);
1831 }
1832 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001833 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001834 }
1835
1836 if (!chosen)
1837 return;
1838 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001839 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1840 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001841}
1842
Johannes Weiner56600482012-01-12 17:17:59 -08001843static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1844 gfp_t gfp_mask,
1845 unsigned long flags)
1846{
1847 unsigned long total = 0;
1848 bool noswap = false;
1849 int loop;
1850
1851 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1852 noswap = true;
1853 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1854 noswap = true;
1855
1856 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1857 if (loop)
1858 drain_all_stock_async(memcg);
1859 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1860 /*
1861 * Allow limit shrinkers, which are triggered directly
1862 * by userspace, to catch signals and stop reclaim
1863 * after minimal progress, regardless of the margin.
1864 */
1865 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1866 break;
1867 if (mem_cgroup_margin(memcg))
1868 break;
1869 /*
1870 * If nothing was reclaimed after two attempts, there
1871 * may be no reclaimable pages in this hierarchy.
1872 */
1873 if (loop && !total)
1874 break;
1875 }
1876 return total;
1877}
1878
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001879/**
1880 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001881 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001882 * @nid: the node ID to be checked.
1883 * @noswap : specify true here if the user wants flle only information.
1884 *
1885 * This function returns whether the specified memcg contains any
1886 * reclaimable pages on a node. Returns true if there are any reclaimable
1887 * pages in the node.
1888 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001889static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001890 int nid, bool noswap)
1891{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001893 return true;
1894 if (noswap || !total_swap_pages)
1895 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001896 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001897 return true;
1898 return false;
1899
1900}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001901#if MAX_NUMNODES > 1
Ying Han889976d2011-05-26 16:25:33 -07001902
1903/*
1904 * Always updating the nodemask is not very good - even if we have an empty
1905 * list or the wrong list here, we can start from some node and traverse all
1906 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1907 *
1908 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001909static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001910{
1911 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001912 /*
1913 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1914 * pagein/pageout changes since the last update.
1915 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001916 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001917 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001918 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001919 return;
1920
Ying Han889976d2011-05-26 16:25:33 -07001921 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001922 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001923
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001924 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001925
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1927 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001928 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001929
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001930 atomic_set(&memcg->numainfo_events, 0);
1931 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001932}
1933
1934/*
1935 * Selecting a node where we start reclaim from. Because what we need is just
1936 * reducing usage counter, start from anywhere is O,K. Considering
1937 * memory reclaim from current node, there are pros. and cons.
1938 *
1939 * Freeing memory from current node means freeing memory from a node which
1940 * we'll use or we've used. So, it may make LRU bad. And if several threads
1941 * hit limits, it will see a contention on a node. But freeing from remote
1942 * node means more costs for memory reclaim because of memory latency.
1943 *
1944 * Now, we use round-robin. Better algorithm is welcomed.
1945 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001946int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001947{
1948 int node;
1949
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001950 mem_cgroup_may_update_nodemask(memcg);
1951 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001952
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001953 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001954 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001955 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001956 /*
1957 * We call this when we hit limit, not when pages are added to LRU.
1958 * No LRU may hold pages because all pages are UNEVICTABLE or
1959 * memcg is too small and all pages are not on LRU. In that case,
1960 * we use curret node.
1961 */
1962 if (unlikely(node == MAX_NUMNODES))
1963 node = numa_node_id();
1964
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001965 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001966 return node;
1967}
1968
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07001969/*
1970 * Check all nodes whether it contains reclaimable pages or not.
1971 * For quick scan, we make use of scan_nodes. This will allow us to skip
1972 * unused nodes. But scan_nodes is lazily updated and may not cotain
1973 * enough new information. We need to do double check.
1974 */
1975static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1976{
1977 int nid;
1978
1979 /*
1980 * quick check...making use of scan_node.
1981 * We can skip unused nodes.
1982 */
1983 if (!nodes_empty(memcg->scan_nodes)) {
1984 for (nid = first_node(memcg->scan_nodes);
1985 nid < MAX_NUMNODES;
1986 nid = next_node(nid, memcg->scan_nodes)) {
1987
1988 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1989 return true;
1990 }
1991 }
1992 /*
1993 * Check rest of nodes.
1994 */
1995 for_each_node_state(nid, N_MEMORY) {
1996 if (node_isset(nid, memcg->scan_nodes))
1997 continue;
1998 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1999 return true;
2000 }
2001 return false;
2002}
2003
Ying Han889976d2011-05-26 16:25:33 -07002004#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002005int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07002006{
2007 return 0;
2008}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07002009
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002010static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
2011{
2012 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
2013}
Ying Han889976d2011-05-26 16:25:33 -07002014#endif
2015
Andrew Morton0608f432013-09-24 15:27:41 -07002016static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
2017 struct zone *zone,
2018 gfp_t gfp_mask,
2019 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08002020{
Andrew Morton0608f432013-09-24 15:27:41 -07002021 struct mem_cgroup *victim = NULL;
2022 int total = 0;
2023 int loop = 0;
2024 unsigned long excess;
2025 unsigned long nr_scanned;
2026 struct mem_cgroup_reclaim_cookie reclaim = {
2027 .zone = zone,
2028 .priority = 0,
2029 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07002030
Andrew Morton0608f432013-09-24 15:27:41 -07002031 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002032
Andrew Morton0608f432013-09-24 15:27:41 -07002033 while (1) {
2034 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
2035 if (!victim) {
2036 loop++;
2037 if (loop >= 2) {
2038 /*
2039 * If we have not been able to reclaim
2040 * anything, it might because there are
2041 * no reclaimable pages under this hierarchy
2042 */
2043 if (!total)
2044 break;
2045 /*
2046 * We want to do more targeted reclaim.
2047 * excess >> 2 is not to excessive so as to
2048 * reclaim too much, nor too less that we keep
2049 * coming back to reclaim from this cgroup
2050 */
2051 if (total >= (excess >> 2) ||
2052 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
2053 break;
2054 }
2055 continue;
2056 }
2057 if (!mem_cgroup_reclaimable(victim, false))
2058 continue;
2059 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
2060 zone, &nr_scanned);
2061 *total_scanned += nr_scanned;
2062 if (!res_counter_soft_limit_excess(&root_memcg->res))
2063 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002064 }
Andrew Morton0608f432013-09-24 15:27:41 -07002065 mem_cgroup_iter_break(root_memcg, victim);
2066 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08002067}
2068
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002069#ifdef CONFIG_LOCKDEP
2070static struct lockdep_map memcg_oom_lock_dep_map = {
2071 .name = "memcg_oom_lock",
2072};
2073#endif
2074
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002075static DEFINE_SPINLOCK(memcg_oom_lock);
2076
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002077/*
2078 * Check OOM-Killer is already running under our hierarchy.
2079 * If someone is running, return false.
2080 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002081static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002082{
Michal Hocko79dfdac2011-07-26 16:08:23 -07002083 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002084
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002085 spin_lock(&memcg_oom_lock);
2086
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002087 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07002088 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07002089 /*
2090 * this subtree of our hierarchy is already locked
2091 * so we cannot give a lock.
2092 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07002093 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002094 mem_cgroup_iter_break(memcg, iter);
2095 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07002096 } else
2097 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002098 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002099
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002100 if (failed) {
2101 /*
2102 * OK, we failed to lock the whole subtree so we have
2103 * to clean up what we set up to the failing subtree
2104 */
2105 for_each_mem_cgroup_tree(iter, memcg) {
2106 if (iter == failed) {
2107 mem_cgroup_iter_break(memcg, iter);
2108 break;
2109 }
2110 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07002111 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002112 } else
2113 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002114
2115 spin_unlock(&memcg_oom_lock);
2116
2117 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002118}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002119
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002120static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002121{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07002122 struct mem_cgroup *iter;
2123
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002124 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07002125 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002126 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002127 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002128 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07002129}
2130
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002131static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002132{
2133 struct mem_cgroup *iter;
2134
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002136 atomic_inc(&iter->under_oom);
2137}
2138
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002139static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002140{
2141 struct mem_cgroup *iter;
2142
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002143 /*
2144 * When a new child is created while the hierarchy is under oom,
2145 * mem_cgroup_oom_lock() may not be called. We have to use
2146 * atomic_add_unless() here.
2147 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002148 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07002149 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002150}
2151
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002152static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
2153
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002154struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002155 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002156 wait_queue_t wait;
2157};
2158
2159static int memcg_oom_wake_function(wait_queue_t *wait,
2160 unsigned mode, int sync, void *arg)
2161{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002162 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
2163 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002164 struct oom_wait_info *oom_wait_info;
2165
2166 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002167 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002168
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002169 /*
Hugh Dickinsd79154b2012-03-21 16:34:18 -07002170 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002171 * Then we can use css_is_ancestor without taking care of RCU.
2172 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002173 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
2174 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002175 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002176 return autoremove_wake_function(wait, mode, sync, arg);
2177}
2178
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002179static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002180{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002181 atomic_inc(&memcg->oom_wakeups);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002182 /* for filtering, pass "memcg" as argument. */
2183 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07002184}
2185
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002186static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002187{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002188 if (memcg && atomic_read(&memcg->under_oom))
2189 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002190}
2191
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002192static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002193{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002194 if (!current->memcg_oom.may_oom)
2195 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002196 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002197 * We are in the middle of the charge context here, so we
2198 * don't want to block when potentially sitting on a callstack
2199 * that holds all kinds of filesystem and mm locks.
2200 *
2201 * Also, the caller may handle a failed allocation gracefully
2202 * (like optional page cache readahead) and so an OOM killer
2203 * invocation might not even be necessary.
2204 *
2205 * That's why we don't do anything here except remember the
2206 * OOM context and then deal with it at the end of the page
2207 * fault when the stack is unwound, the locks are released,
2208 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002209 */
Johannes Weiner49426422013-10-16 13:46:59 -07002210 css_get(&memcg->css);
2211 current->memcg_oom.memcg = memcg;
2212 current->memcg_oom.gfp_mask = mask;
2213 current->memcg_oom.order = order;
2214}
2215
2216/**
2217 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2218 * @handle: actually kill/wait or just clean up the OOM state
2219 *
2220 * This has to be called at the end of a page fault if the memcg OOM
2221 * handler was enabled.
2222 *
2223 * Memcg supports userspace OOM handling where failed allocations must
2224 * sleep on a waitqueue until the userspace task resolves the
2225 * situation. Sleeping directly in the charge context with all kinds
2226 * of locks held is not a good idea, instead we remember an OOM state
2227 * in the task and mem_cgroup_oom_synchronize() has to be called at
2228 * the end of the page fault to complete the OOM handling.
2229 *
2230 * Returns %true if an ongoing memcg OOM situation was detected and
2231 * completed, %false otherwise.
2232 */
2233bool mem_cgroup_oom_synchronize(bool handle)
2234{
2235 struct mem_cgroup *memcg = current->memcg_oom.memcg;
2236 struct oom_wait_info owait;
2237 bool locked;
2238
2239 /* OOM is global, do not handle */
2240 if (!memcg)
2241 return false;
2242
2243 if (!handle)
2244 goto cleanup;
2245
2246 owait.memcg = memcg;
2247 owait.wait.flags = 0;
2248 owait.wait.func = memcg_oom_wake_function;
2249 owait.wait.private = current;
2250 INIT_LIST_HEAD(&owait.wait.task_list);
2251
2252 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002253 mem_cgroup_mark_under_oom(memcg);
2254
2255 locked = mem_cgroup_oom_trylock(memcg);
2256
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002257 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002258 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002259
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002260 if (locked && !memcg->oom_kill_disable) {
2261 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002262 finish_wait(&memcg_oom_waitq, &owait.wait);
2263 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
2264 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002265 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002266 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002267 mem_cgroup_unmark_under_oom(memcg);
2268 finish_wait(&memcg_oom_waitq, &owait.wait);
2269 }
2270
2271 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002272 mem_cgroup_oom_unlock(memcg);
2273 /*
2274 * There is no guarantee that an OOM-lock contender
2275 * sees the wakeups triggered by the OOM kill
2276 * uncharges. Wake any sleepers explicitely.
2277 */
2278 memcg_oom_recover(memcg);
2279 }
Johannes Weiner49426422013-10-16 13:46:59 -07002280cleanup:
2281 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002282 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002283 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002284}
2285
Balbir Singhd69b0422009-06-17 16:26:34 -07002286/*
2287 * Currently used to update mapped file statistics, but the routine can be
2288 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002289 *
2290 * Notes: Race condition
2291 *
2292 * We usually use page_cgroup_lock() for accessing page_cgroup member but
2293 * it tends to be costly. But considering some conditions, we doesn't need
2294 * to do so _always_.
2295 *
2296 * Considering "charge", lock_page_cgroup() is not required because all
2297 * file-stat operations happen after a page is attached to radix-tree. There
2298 * are no race with "charge".
2299 *
2300 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
2301 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
2302 * if there are race with "uncharge". Statistics itself is properly handled
2303 * by flags.
2304 *
2305 * Considering "move", this is an only case we see a race. To make the race
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002306 * small, we check mm->moving_account and detect there are possibility of race
2307 * If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07002308 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002309
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002310void __mem_cgroup_begin_update_page_stat(struct page *page,
2311 bool *locked, unsigned long *flags)
2312{
2313 struct mem_cgroup *memcg;
2314 struct page_cgroup *pc;
2315
2316 pc = lookup_page_cgroup(page);
2317again:
2318 memcg = pc->mem_cgroup;
2319 if (unlikely(!memcg || !PageCgroupUsed(pc)))
2320 return;
2321 /*
2322 * If this memory cgroup is not under account moving, we don't
Wanpeng Lida92c472012-07-31 16:43:26 -07002323 * need to take move_lock_mem_cgroup(). Because we already hold
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002324 * rcu_read_lock(), any calls to move_account will be delayed until
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002325 * rcu_read_unlock() if mem_cgroup_stolen() == true.
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002326 */
Andrew Morton13fd1dd92012-03-21 16:34:26 -07002327 if (!mem_cgroup_stolen(memcg))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002328 return;
2329
2330 move_lock_mem_cgroup(memcg, flags);
2331 if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2332 move_unlock_mem_cgroup(memcg, flags);
2333 goto again;
2334 }
2335 *locked = true;
2336}
2337
2338void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2339{
2340 struct page_cgroup *pc = lookup_page_cgroup(page);
2341
2342 /*
2343 * It's guaranteed that pc->mem_cgroup never changes while
2344 * lock is held because a routine modifies pc->mem_cgroup
Wanpeng Lida92c472012-07-31 16:43:26 -07002345 * should take move_lock_mem_cgroup().
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002346 */
2347 move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2348}
2349
Greg Thelen2a7106f2011-01-13 15:47:37 -08002350void mem_cgroup_update_page_stat(struct page *page,
Sha Zhengju68b48762013-09-12 15:13:50 -07002351 enum mem_cgroup_stat_index idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07002352{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002353 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002354 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002355 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07002356
Johannes Weinercfa44942012-01-12 17:18:38 -08002357 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07002358 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002359
Sha Zhengju658b72c2013-09-12 15:13:52 -07002360 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002361 memcg = pc->mem_cgroup;
2362 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002363 return;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002364
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002365 this_cpu_add(memcg->stat->count[idx], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07002366}
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002367
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002368/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369 * size of first charge trial. "32" comes from vmscan.c's magic value.
2370 * TODO: maybe necessary to use big numbers in big irons.
2371 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002372#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002373struct memcg_stock_pcp {
2374 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002375 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002376 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002377 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002378#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002379};
2380static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002381static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002382
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002383/**
2384 * consume_stock: Try to consume stocked charge on this cpu.
2385 * @memcg: memcg to consume from.
2386 * @nr_pages: how many pages to charge.
2387 *
2388 * The charges will only happen if @memcg matches the current cpu's memcg
2389 * stock, and at least @nr_pages are available in that stock. Failure to
2390 * service an allocation will refill the stock.
2391 *
2392 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002393 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002394static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002395{
2396 struct memcg_stock_pcp *stock;
2397 bool ret = true;
2398
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002399 if (nr_pages > CHARGE_BATCH)
2400 return false;
2401
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002402 stock = &get_cpu_var(memcg_stock);
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002403 if (memcg == stock->cached && stock->nr_pages >= nr_pages)
2404 stock->nr_pages -= nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002405 else /* need to call res_counter_charge */
2406 ret = false;
2407 put_cpu_var(memcg_stock);
2408 return ret;
2409}
2410
2411/*
2412 * Returns stocks cached in percpu to res_counter and reset cached information.
2413 */
2414static void drain_stock(struct memcg_stock_pcp *stock)
2415{
2416 struct mem_cgroup *old = stock->cached;
2417
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002418 if (stock->nr_pages) {
2419 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2420
2421 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002422 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002423 res_counter_uncharge(&old->memsw, bytes);
2424 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002425 }
2426 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002427}
2428
2429/*
2430 * This must be called under preempt disabled or must be called by
2431 * a thread which is pinned to local cpu.
2432 */
2433static void drain_local_stock(struct work_struct *dummy)
2434{
2435 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2436 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002437 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002438}
2439
Michal Hockoe4777492013-02-22 16:35:40 -08002440static void __init memcg_stock_init(void)
2441{
2442 int cpu;
2443
2444 for_each_possible_cpu(cpu) {
2445 struct memcg_stock_pcp *stock =
2446 &per_cpu(memcg_stock, cpu);
2447 INIT_WORK(&stock->work, drain_local_stock);
2448 }
2449}
2450
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002451/*
2452 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002453 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002454 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002455static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002456{
2457 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2458
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002459 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002460 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002461 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002462 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002463 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002464 put_cpu_var(memcg_stock);
2465}
2466
2467/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002468 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002469 * of the hierarchy under it. sync flag says whether we should block
2470 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002471 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002472static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002473{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002474 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002475
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002476 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002477 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002478 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002479 for_each_online_cpu(cpu) {
2480 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002481 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002482
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002483 memcg = stock->cached;
2484 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002485 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002486 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002487 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002488 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2489 if (cpu == curcpu)
2490 drain_local_stock(&stock->work);
2491 else
2492 schedule_work_on(cpu, &stock->work);
2493 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002494 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002495 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002496
2497 if (!sync)
2498 goto out;
2499
2500 for_each_online_cpu(cpu) {
2501 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002502 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002503 flush_work(&stock->work);
2504 }
2505out:
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002506 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002507}
2508
2509/*
2510 * Tries to drain stocked charges in other cpus. This function is asynchronous
2511 * and just put a work per cpu for draining localy on each cpu. Caller can
2512 * expects some charges will be back to res_counter later but cannot wait for
2513 * it.
2514 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002515static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002516{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002517 /*
2518 * If someone calls draining, avoid adding more kworker runs.
2519 */
2520 if (!mutex_trylock(&percpu_charge_mutex))
2521 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002522 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002523 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002524}
2525
2526/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002527static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002528{
2529 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002530 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002531 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002532 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002533}
2534
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002535/*
2536 * This function drains percpu counter value from DEAD cpu and
2537 * move it to local cpu. Note that this function can be preempted.
2538 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002539static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002540{
2541 int i;
2542
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002543 spin_lock(&memcg->pcp_counter_lock);
Johannes Weiner61046212012-05-29 15:07:05 -07002544 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002545 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002546
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002547 per_cpu(memcg->stat->count[i], cpu) = 0;
2548 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002549 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002550 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002551 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002552
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002553 per_cpu(memcg->stat->events[i], cpu) = 0;
2554 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002555 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002556 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002557}
2558
Paul Gortmaker0db06282013-06-19 14:53:51 -04002559static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002560 unsigned long action,
2561 void *hcpu)
2562{
2563 int cpu = (unsigned long)hcpu;
2564 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002565 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002566
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07002567 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002568 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002569
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07002570 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002571 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002572
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002573 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002574 mem_cgroup_drain_pcp_counter(iter, cpu);
2575
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002576 stock = &per_cpu(memcg_stock, cpu);
2577 drain_stock(stock);
2578 return NOTIFY_OK;
2579}
2580
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002581
2582/* See __mem_cgroup_try_charge() for details */
2583enum {
2584 CHARGE_OK, /* success */
2585 CHARGE_RETRY, /* need to retry but retry is not bad */
2586 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2587 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002588};
2589
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002590static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002591 unsigned int nr_pages, unsigned int min_pages,
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002592 bool invoke_oom)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002593{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002594 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002595 struct mem_cgroup *mem_over_limit;
2596 struct res_counter *fail_res;
2597 unsigned long flags = 0;
2598 int ret;
2599
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002600 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002601
2602 if (likely(!ret)) {
2603 if (!do_swap_account)
2604 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002605 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002606 if (likely(!ret))
2607 return CHARGE_OK;
2608
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002609 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002610 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2611 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2612 } else
2613 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002614 /*
Johannes Weiner9221edb2011-02-01 15:52:42 -08002615 * Never reclaim on behalf of optional batching, retry with a
2616 * single page instead.
2617 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002618 if (nr_pages > min_pages)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002619 return CHARGE_RETRY;
2620
2621 if (!(gfp_mask & __GFP_WAIT))
2622 return CHARGE_WOULDBLOCK;
2623
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002624 if (gfp_mask & __GFP_NORETRY)
2625 return CHARGE_NOMEM;
2626
Johannes Weiner56600482012-01-12 17:17:59 -08002627 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002628 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002629 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002630 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002631 * Even though the limit is exceeded at this point, reclaim
2632 * may have been able to free some pages. Retry the charge
2633 * before killing the task.
2634 *
2635 * Only for regular pages, though: huge pages are rather
2636 * unlikely to succeed so close to the limit, and we fall back
2637 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002638 */
Suleiman Souhlal4c9c5352012-12-18 14:21:41 -08002639 if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002640 return CHARGE_RETRY;
2641
2642 /*
2643 * At task move, charge accounts can be doubly counted. So, it's
2644 * better to wait until the end of task_move if something is going on.
2645 */
2646 if (mem_cgroup_wait_acct_move(mem_over_limit))
2647 return CHARGE_RETRY;
2648
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002649 if (invoke_oom)
2650 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize));
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002651
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002652 return CHARGE_NOMEM;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002653}
2654
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002655/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002656 * __mem_cgroup_try_charge() does
2657 * 1. detect memcg to be charged against from passed *mm and *ptr,
2658 * 2. update res_counter
2659 * 3. call memory reclaim if necessary.
2660 *
2661 * In some special case, if the task is fatal, fatal_signal_pending() or
2662 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2663 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2664 * as possible without any hazards. 2: all pages should have a valid
2665 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2666 * pointer, that is treated as a charge to root_mem_cgroup.
2667 *
2668 * So __mem_cgroup_try_charge() will return
2669 * 0 ... on success, filling *ptr with a valid memcg pointer.
2670 * -ENOMEM ... charge failure because of resource limits.
2671 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2672 *
2673 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2674 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002675 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002676static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002677 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002678 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002679 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002680 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002681{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002682 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002683 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002684 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002685 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002686
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002687 /*
2688 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2689 * in system level. So, allow to go ahead dying process in addition to
2690 * MEMDIE process.
2691 */
2692 if (unlikely(test_thread_flag(TIF_MEMDIE)
2693 || fatal_signal_pending(current)))
2694 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002695
Johannes Weiner49426422013-10-16 13:46:59 -07002696 if (unlikely(task_in_memcg_oom(current)))
Johannes Weiner1f14c1a2013-12-12 17:12:35 -08002697 goto nomem;
Johannes Weiner49426422013-10-16 13:46:59 -07002698
Johannes Weinera0d8b002013-12-12 17:12:20 -08002699 if (gfp_mask & __GFP_NOFAIL)
2700 oom = false;
2701
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002702 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002703 * We always charge the cgroup the mm_struct belongs to.
2704 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002705 * thread group leader migrates. It's possible that mm is not
Johannes Weiner24467ca2012-07-31 16:45:40 -07002706 * set, if so charge the root memcg (happens for pagecache usage).
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002707 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002708 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002709 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002710again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002711 if (*ptr) { /* css should be a valid one */
2712 memcg = *ptr;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002713 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002714 goto done;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002715 if (consume_stock(memcg, nr_pages))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002716 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002717 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002718 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002719 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002720
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002721 rcu_read_lock();
2722 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002723 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002724 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002725 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002726 * race with swapoff. Then, we have small risk of mis-accouning.
2727 * But such kind of mis-account by race always happens because
2728 * we don't have cgroup_mutex(). It's overkill and we allo that
2729 * small race, here.
2730 * (*) swapoff at el will charge against mm-struct not against
2731 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002732 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002733 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002734 if (!memcg)
2735 memcg = root_mem_cgroup;
2736 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002737 rcu_read_unlock();
2738 goto done;
2739 }
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002740 if (consume_stock(memcg, nr_pages)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002741 /*
2742 * It seems dagerous to access memcg without css_get().
2743 * But considering how consume_stok works, it's not
2744 * necessary. If consume_stock success, some charges
2745 * from this memcg are cached on this cpu. So, we
2746 * don't need to call css_get()/css_tryget() before
2747 * calling consume_stock().
2748 */
2749 rcu_read_unlock();
2750 goto done;
2751 }
2752 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002753 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002754 rcu_read_unlock();
2755 goto again;
2756 }
2757 rcu_read_unlock();
2758 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002759
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002760 do {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002761 bool invoke_oom = oom && !nr_oom_retries;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002762
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002763 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002764 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002765 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002766 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002767 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002768
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002769 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch,
2770 nr_pages, invoke_oom);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002771 switch (ret) {
2772 case CHARGE_OK:
2773 break;
2774 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002775 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002776 css_put(&memcg->css);
2777 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002778 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002779 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002780 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002781 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002782 case CHARGE_NOMEM: /* OOM routine works */
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002783 if (!oom || invoke_oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002784 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002785 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002786 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002787 nr_oom_retries--;
2788 break;
Balbir Singh66e17072008-02-07 00:13:56 -08002789 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002790 } while (ret != CHARGE_OK);
2791
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002792 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002793 refill_stock(memcg, batch - nr_pages);
2794 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002795done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002796 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002797 return 0;
2798nomem:
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002799 if (!(gfp_mask & __GFP_NOFAIL)) {
2800 *ptr = NULL;
2801 return -ENOMEM;
2802 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002803bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002804 *ptr = root_mem_cgroup;
2805 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002806}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002807
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002808/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002809 * Somemtimes we have to undo a charge we got by try_charge().
2810 * This function is for that and do uncharge, put css's refcnt.
2811 * gotten by try_charge().
2812 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002813static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002814 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002815{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002816 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002817 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002818
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002819 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002820 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002821 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002822 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002823}
2824
2825/*
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002826 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2827 * This is useful when moving usage to parent cgroup.
2828 */
2829static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2830 unsigned int nr_pages)
2831{
2832 unsigned long bytes = nr_pages * PAGE_SIZE;
2833
2834 if (mem_cgroup_is_root(memcg))
2835 return;
2836
2837 res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2838 if (do_swap_account)
2839 res_counter_uncharge_until(&memcg->memsw,
2840 memcg->memsw.parent, bytes);
2841}
2842
2843/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002844 * A helper function to get mem_cgroup from ID. must be called under
Tejun Heoe9316082012-11-05 09:16:58 -08002845 * rcu_read_lock(). The caller is responsible for calling css_tryget if
2846 * the mem_cgroup is used for charging. (dropping refcnt from swap can be
2847 * called against removed memcg.)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002848 */
2849static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2850{
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002851 /* ID 0 is unused ID */
2852 if (!id)
2853 return NULL;
Li Zefan34c00c32013-09-23 16:56:01 +08002854 return mem_cgroup_from_id(id);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002855}
2856
Wu Fengguange42d9d52009-12-16 12:19:59 +01002857struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002858{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002859 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002860 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002861 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002862 swp_entry_t ent;
2863
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002864 VM_BUG_ON(!PageLocked(page));
2865
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002866 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002867 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002868 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002869 memcg = pc->mem_cgroup;
2870 if (memcg && !css_tryget(&memcg->css))
2871 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002872 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002873 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002874 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002875 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002876 memcg = mem_cgroup_lookup(id);
2877 if (memcg && !css_tryget(&memcg->css))
2878 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002879 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002880 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002881 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002882 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002883}
2884
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002885static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002886 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002887 unsigned int nr_pages,
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002888 enum charge_type ctype,
2889 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002890{
Johannes Weinerce587e62012-04-24 20:22:33 +02002891 struct page_cgroup *pc = lookup_page_cgroup(page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002892 struct zone *uninitialized_var(zone);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002893 struct lruvec *lruvec;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002894 bool was_on_lru = false;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002895 bool anon;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002896
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002897 lock_page_cgroup(pc);
Johannes Weiner90deb782012-07-31 16:45:47 -07002898 VM_BUG_ON(PageCgroupUsed(pc));
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002899 /*
2900 * we don't need page_cgroup_lock about tail pages, becase they are not
2901 * accessed by any other context at this point.
2902 */
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002903
2904 /*
2905 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2906 * may already be on some other mem_cgroup's LRU. Take care of it.
2907 */
2908 if (lrucare) {
2909 zone = page_zone(page);
2910 spin_lock_irq(&zone->lru_lock);
2911 if (PageLRU(page)) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002912 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002913 ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002914 del_page_from_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002915 was_on_lru = true;
2916 }
2917 }
2918
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002919 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002920 /*
2921 * We access a page_cgroup asynchronously without lock_page_cgroup().
2922 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2923 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2924 * before USED bit, we need memory barrier here.
2925 * See mem_cgroup_add_lru_list(), etc.
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002926 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002927 smp_wmb();
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002928 SetPageCgroupUsed(pc);
Hugh Dickins3be91272008-02-07 00:14:19 -08002929
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002930 if (lrucare) {
2931 if (was_on_lru) {
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002932 lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002933 VM_BUG_ON(PageLRU(page));
2934 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07002935 add_page_to_lru_list(page, lruvec, page_lru(page));
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002936 }
2937 spin_unlock_irq(&zone->lru_lock);
2938 }
2939
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07002940 if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07002941 anon = true;
2942 else
2943 anon = false;
2944
David Rientjesb070e652013-05-07 16:18:09 -07002945 mem_cgroup_charge_statistics(memcg, page, anon, nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002946 unlock_page_cgroup(pc);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002947
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002948 /*
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07002949 * "charge_statistics" updated event counter. Then, check it.
2950 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2951 * if they exceeds softlimit.
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002952 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002953 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002954}
2955
Glauber Costa7cf27982012-12-18 14:22:55 -08002956static DEFINE_MUTEX(set_limit_mutex);
2957
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002958#ifdef CONFIG_MEMCG_KMEM
2959static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg)
2960{
2961 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) &&
Vladimir Davydov1c98dd92014-01-21 15:49:41 -08002962 (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK) ==
2963 KMEM_ACCOUNTED_MASK;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002964}
2965
Glauber Costa1f458cb2012-12-18 14:22:50 -08002966/*
2967 * This is a bit cumbersome, but it is rarely used and avoids a backpointer
2968 * in the memcg_cache_params struct.
2969 */
2970static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p)
2971{
2972 struct kmem_cache *cachep;
2973
2974 VM_BUG_ON(p->is_root_cache);
2975 cachep = p->root_cache;
Qiang Huang7a67d7a2013-11-12 15:08:24 -08002976 return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg));
Glauber Costa1f458cb2012-12-18 14:22:50 -08002977}
2978
Glauber Costa749c5412012-12-18 14:23:01 -08002979#ifdef CONFIG_SLABINFO
Tejun Heo182446d2013-08-08 20:11:24 -04002980static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css,
2981 struct cftype *cft, struct seq_file *m)
Glauber Costa749c5412012-12-18 14:23:01 -08002982{
Tejun Heo182446d2013-08-08 20:11:24 -04002983 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa749c5412012-12-18 14:23:01 -08002984 struct memcg_cache_params *params;
2985
2986 if (!memcg_can_account_kmem(memcg))
2987 return -EIO;
2988
2989 print_slabinfo_header(m);
2990
2991 mutex_lock(&memcg->slab_caches_mutex);
2992 list_for_each_entry(params, &memcg->memcg_slab_caches, list)
2993 cache_show(memcg_params_to_cache(params), m);
2994 mutex_unlock(&memcg->slab_caches_mutex);
2995
2996 return 0;
2997}
2998#endif
2999
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003000static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size)
3001{
3002 struct res_counter *fail_res;
3003 struct mem_cgroup *_memcg;
3004 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003005
3006 ret = res_counter_charge(&memcg->kmem, size, &fail_res);
3007 if (ret)
3008 return ret;
3009
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003010 _memcg = memcg;
3011 ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT,
Qiang Huangb9921ec2013-11-12 15:07:22 -08003012 &_memcg, oom_gfp_allowed(gfp));
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003013
3014 if (ret == -EINTR) {
3015 /*
3016 * __mem_cgroup_try_charge() chosed to bypass to root due to
3017 * OOM kill or fatal signal. Since our only options are to
3018 * either fail the allocation or charge it to this cgroup, do
3019 * it as a temporary condition. But we can't fail. From a
3020 * kmem/slab perspective, the cache has already been selected,
3021 * by mem_cgroup_kmem_get_cache(), so it is too late to change
3022 * our minds.
3023 *
3024 * This condition will only trigger if the task entered
3025 * memcg_charge_kmem in a sane state, but was OOM-killed during
3026 * __mem_cgroup_try_charge() above. Tasks that were already
3027 * dying when the allocation triggers should have been already
3028 * directed to the root cgroup in memcontrol.h
3029 */
3030 res_counter_charge_nofail(&memcg->res, size, &fail_res);
3031 if (do_swap_account)
3032 res_counter_charge_nofail(&memcg->memsw, size,
3033 &fail_res);
3034 ret = 0;
3035 } else if (ret)
3036 res_counter_uncharge(&memcg->kmem, size);
3037
3038 return ret;
3039}
3040
3041static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size)
3042{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003043 res_counter_uncharge(&memcg->res, size);
3044 if (do_swap_account)
3045 res_counter_uncharge(&memcg->memsw, size);
Glauber Costa7de37682012-12-18 14:22:07 -08003046
3047 /* Not down to 0 */
3048 if (res_counter_uncharge(&memcg->kmem, size))
3049 return;
3050
Li Zefan10d5ebf2013-07-08 16:00:33 -07003051 /*
3052 * Releases a reference taken in kmem_cgroup_css_offline in case
3053 * this last uncharge is racing with the offlining code or it is
3054 * outliving the memcg existence.
3055 *
3056 * The memory barrier imposed by test&clear is paired with the
3057 * explicit one in memcg_kmem_mark_dead().
3058 */
Glauber Costa7de37682012-12-18 14:22:07 -08003059 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07003060 css_put(&memcg->css);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003061}
3062
Glauber Costa2633d7a2012-12-18 14:22:34 -08003063void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep)
3064{
3065 if (!memcg)
3066 return;
3067
3068 mutex_lock(&memcg->slab_caches_mutex);
3069 list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches);
3070 mutex_unlock(&memcg->slab_caches_mutex);
3071}
3072
3073/*
3074 * helper for acessing a memcg's index. It will be used as an index in the
3075 * child cache array in kmem_cache, and also to derive its name. This function
3076 * will return -1 when this is not a kmem-limited memcg.
3077 */
3078int memcg_cache_id(struct mem_cgroup *memcg)
3079{
3080 return memcg ? memcg->kmemcg_id : -1;
3081}
3082
Glauber Costa55007d82012-12-18 14:22:38 -08003083/*
3084 * This ends up being protected by the set_limit mutex, during normal
3085 * operation, because that is its main call site.
3086 *
3087 * But when we create a new cache, we can call this as well if its parent
3088 * is kmem-limited. That will have to hold set_limit_mutex as well.
3089 */
3090int memcg_update_cache_sizes(struct mem_cgroup *memcg)
3091{
3092 int num, ret;
3093
3094 num = ida_simple_get(&kmem_limited_groups,
3095 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
3096 if (num < 0)
3097 return num;
3098 /*
3099 * After this point, kmem_accounted (that we test atomically in
3100 * the beginning of this conditional), is no longer 0. This
3101 * guarantees only one process will set the following boolean
3102 * to true. We don't need test_and_set because we're protected
3103 * by the set_limit_mutex anyway.
3104 */
3105 memcg_kmem_set_activated(memcg);
3106
3107 ret = memcg_update_all_caches(num+1);
3108 if (ret) {
3109 ida_simple_remove(&kmem_limited_groups, num);
3110 memcg_kmem_clear_activated(memcg);
3111 return ret;
3112 }
3113
3114 memcg->kmemcg_id = num;
3115 INIT_LIST_HEAD(&memcg->memcg_slab_caches);
3116 mutex_init(&memcg->slab_caches_mutex);
3117 return 0;
3118}
3119
3120static size_t memcg_caches_array_size(int num_groups)
3121{
3122 ssize_t size;
3123 if (num_groups <= 0)
3124 return 0;
3125
3126 size = 2 * num_groups;
3127 if (size < MEMCG_CACHES_MIN_SIZE)
3128 size = MEMCG_CACHES_MIN_SIZE;
3129 else if (size > MEMCG_CACHES_MAX_SIZE)
3130 size = MEMCG_CACHES_MAX_SIZE;
3131
3132 return size;
3133}
3134
3135/*
3136 * We should update the current array size iff all caches updates succeed. This
3137 * can only be done from the slab side. The slab mutex needs to be held when
3138 * calling this.
3139 */
3140void memcg_update_array_size(int num)
3141{
3142 if (num > memcg_limited_groups_array_size)
3143 memcg_limited_groups_array_size = memcg_caches_array_size(num);
3144}
3145
Konstantin Khlebnikov15cf17d2013-03-08 12:43:36 -08003146static void kmem_cache_destroy_work_func(struct work_struct *w);
3147
Glauber Costa55007d82012-12-18 14:22:38 -08003148int memcg_update_cache_size(struct kmem_cache *s, int num_groups)
3149{
3150 struct memcg_cache_params *cur_params = s->memcg_params;
3151
Qiang Huangf35c3a82013-11-12 15:08:22 -08003152 VM_BUG_ON(!is_root_cache(s));
Glauber Costa55007d82012-12-18 14:22:38 -08003153
3154 if (num_groups > memcg_limited_groups_array_size) {
3155 int i;
3156 ssize_t size = memcg_caches_array_size(num_groups);
3157
3158 size *= sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003159 size += offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003160
3161 s->memcg_params = kzalloc(size, GFP_KERNEL);
3162 if (!s->memcg_params) {
3163 s->memcg_params = cur_params;
3164 return -ENOMEM;
3165 }
3166
3167 s->memcg_params->is_root_cache = true;
3168
3169 /*
3170 * There is the chance it will be bigger than
3171 * memcg_limited_groups_array_size, if we failed an allocation
3172 * in a cache, in which case all caches updated before it, will
3173 * have a bigger array.
3174 *
3175 * But if that is the case, the data after
3176 * memcg_limited_groups_array_size is certainly unused
3177 */
3178 for (i = 0; i < memcg_limited_groups_array_size; i++) {
3179 if (!cur_params->memcg_caches[i])
3180 continue;
3181 s->memcg_params->memcg_caches[i] =
3182 cur_params->memcg_caches[i];
3183 }
3184
3185 /*
3186 * Ideally, we would wait until all caches succeed, and only
3187 * then free the old one. But this is not worth the extra
3188 * pointer per-cache we'd have to have for this.
3189 *
3190 * It is not a big deal if some caches are left with a size
3191 * bigger than the others. And all updates will reset this
3192 * anyway.
3193 */
3194 kfree(cur_params);
3195 }
3196 return 0;
3197}
3198
Glauber Costa943a4512012-12-18 14:23:03 -08003199int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
3200 struct kmem_cache *root_cache)
Glauber Costa2633d7a2012-12-18 14:22:34 -08003201{
Andrey Vagin90c7a792013-09-11 14:22:18 -07003202 size_t size;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003203
3204 if (!memcg_kmem_enabled())
3205 return 0;
3206
Andrey Vagin90c7a792013-09-11 14:22:18 -07003207 if (!memcg) {
3208 size = offsetof(struct memcg_cache_params, memcg_caches);
Glauber Costa55007d82012-12-18 14:22:38 -08003209 size += memcg_limited_groups_array_size * sizeof(void *);
Andrey Vagin90c7a792013-09-11 14:22:18 -07003210 } else
3211 size = sizeof(struct memcg_cache_params);
Glauber Costa55007d82012-12-18 14:22:38 -08003212
Glauber Costa2633d7a2012-12-18 14:22:34 -08003213 s->memcg_params = kzalloc(size, GFP_KERNEL);
3214 if (!s->memcg_params)
3215 return -ENOMEM;
3216
Glauber Costa943a4512012-12-18 14:23:03 -08003217 if (memcg) {
Glauber Costa2633d7a2012-12-18 14:22:34 -08003218 s->memcg_params->memcg = memcg;
Glauber Costa943a4512012-12-18 14:23:03 -08003219 s->memcg_params->root_cache = root_cache;
Andrey Vagin3e6b11d2013-08-13 16:00:47 -07003220 INIT_WORK(&s->memcg_params->destroy,
3221 kmem_cache_destroy_work_func);
Glauber Costa4ba902b2013-02-12 13:46:22 -08003222 } else
3223 s->memcg_params->is_root_cache = true;
3224
Glauber Costa2633d7a2012-12-18 14:22:34 -08003225 return 0;
3226}
3227
3228void memcg_release_cache(struct kmem_cache *s)
3229{
Glauber Costad7f25f82012-12-18 14:22:40 -08003230 struct kmem_cache *root;
3231 struct mem_cgroup *memcg;
3232 int id;
3233
3234 /*
3235 * This happens, for instance, when a root cache goes away before we
3236 * add any memcg.
3237 */
3238 if (!s->memcg_params)
3239 return;
3240
3241 if (s->memcg_params->is_root_cache)
3242 goto out;
3243
3244 memcg = s->memcg_params->memcg;
3245 id = memcg_cache_id(memcg);
3246
3247 root = s->memcg_params->root_cache;
3248 root->memcg_params->memcg_caches[id] = NULL;
Glauber Costad7f25f82012-12-18 14:22:40 -08003249
3250 mutex_lock(&memcg->slab_caches_mutex);
3251 list_del(&s->memcg_params->list);
3252 mutex_unlock(&memcg->slab_caches_mutex);
3253
Li Zefan20f05312013-07-08 16:00:31 -07003254 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003255out:
Glauber Costa2633d7a2012-12-18 14:22:34 -08003256 kfree(s->memcg_params);
3257}
3258
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003259/*
3260 * During the creation a new cache, we need to disable our accounting mechanism
3261 * altogether. This is true even if we are not creating, but rather just
3262 * enqueing new caches to be created.
3263 *
3264 * This is because that process will trigger allocations; some visible, like
3265 * explicit kmallocs to auxiliary data structures, name strings and internal
3266 * cache structures; some well concealed, like INIT_WORK() that can allocate
3267 * objects during debug.
3268 *
3269 * If any allocation happens during memcg_kmem_get_cache, we will recurse back
3270 * to it. This may not be a bounded recursion: since the first cache creation
3271 * failed to complete (waiting on the allocation), we'll just try to create the
3272 * cache again, failing at the same point.
3273 *
3274 * memcg_kmem_get_cache is prepared to abort after seeing a positive count of
3275 * memcg_kmem_skip_account. So we enclose anything that might allocate memory
3276 * inside the following two functions.
3277 */
3278static inline void memcg_stop_kmem_account(void)
3279{
3280 VM_BUG_ON(!current->mm);
3281 current->memcg_kmem_skip_account++;
3282}
3283
3284static inline void memcg_resume_kmem_account(void)
3285{
3286 VM_BUG_ON(!current->mm);
3287 current->memcg_kmem_skip_account--;
3288}
3289
Glauber Costa1f458cb2012-12-18 14:22:50 -08003290static void kmem_cache_destroy_work_func(struct work_struct *w)
3291{
3292 struct kmem_cache *cachep;
3293 struct memcg_cache_params *p;
3294
3295 p = container_of(w, struct memcg_cache_params, destroy);
3296
3297 cachep = memcg_params_to_cache(p);
3298
Glauber Costa22933152012-12-18 14:22:59 -08003299 /*
3300 * If we get down to 0 after shrink, we could delete right away.
3301 * However, memcg_release_pages() already puts us back in the workqueue
3302 * in that case. If we proceed deleting, we'll get a dangling
3303 * reference, and removing the object from the workqueue in that case
3304 * is unnecessary complication. We are not a fast path.
3305 *
3306 * Note that this case is fundamentally different from racing with
3307 * shrink_slab(): if memcg_cgroup_destroy_cache() is called in
3308 * kmem_cache_shrink, not only we would be reinserting a dead cache
3309 * into the queue, but doing so from inside the worker racing to
3310 * destroy it.
3311 *
3312 * So if we aren't down to zero, we'll just schedule a worker and try
3313 * again
3314 */
3315 if (atomic_read(&cachep->memcg_params->nr_pages) != 0) {
3316 kmem_cache_shrink(cachep);
3317 if (atomic_read(&cachep->memcg_params->nr_pages) == 0)
3318 return;
3319 } else
Glauber Costa1f458cb2012-12-18 14:22:50 -08003320 kmem_cache_destroy(cachep);
3321}
3322
3323void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
3324{
3325 if (!cachep->memcg_params->dead)
3326 return;
3327
3328 /*
Glauber Costa22933152012-12-18 14:22:59 -08003329 * There are many ways in which we can get here.
3330 *
3331 * We can get to a memory-pressure situation while the delayed work is
3332 * still pending to run. The vmscan shrinkers can then release all
3333 * cache memory and get us to destruction. If this is the case, we'll
3334 * be executed twice, which is a bug (the second time will execute over
3335 * bogus data). In this case, cancelling the work should be fine.
3336 *
3337 * But we can also get here from the worker itself, if
3338 * kmem_cache_shrink is enough to shake all the remaining objects and
3339 * get the page count to 0. In this case, we'll deadlock if we try to
3340 * cancel the work (the worker runs with an internal lock held, which
3341 * is the same lock we would hold for cancel_work_sync().)
3342 *
3343 * Since we can't possibly know who got us here, just refrain from
3344 * running if there is already work pending
3345 */
3346 if (work_pending(&cachep->memcg_params->destroy))
3347 return;
3348 /*
Glauber Costa1f458cb2012-12-18 14:22:50 -08003349 * We have to defer the actual destroying to a workqueue, because
3350 * we might currently be in a context that cannot sleep.
3351 */
3352 schedule_work(&cachep->memcg_params->destroy);
3353}
3354
Glauber Costad7f25f82012-12-18 14:22:40 -08003355/*
3356 * This lock protects updaters, not readers. We want readers to be as fast as
3357 * they can, and they will either see NULL or a valid cache value. Our model
3358 * allow them to see NULL, in which case the root memcg will be selected.
3359 *
3360 * We need this lock because multiple allocations to the same cache from a non
3361 * will span more than one worker. Only one of them can create the cache.
3362 */
3363static DEFINE_MUTEX(memcg_cache_mutex);
Michal Hockod9c10dd2013-03-28 08:48:14 +01003364
3365/*
3366 * Called with memcg_cache_mutex held
3367 */
3368static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg,
3369 struct kmem_cache *s)
3370{
3371 struct kmem_cache *new;
3372 static char *tmp_name = NULL;
3373
3374 lockdep_assert_held(&memcg_cache_mutex);
3375
3376 /*
3377 * kmem_cache_create_memcg duplicates the given name and
3378 * cgroup_name for this name requires RCU context.
3379 * This static temporary buffer is used to prevent from
3380 * pointless shortliving allocation.
3381 */
3382 if (!tmp_name) {
3383 tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);
3384 if (!tmp_name)
3385 return NULL;
3386 }
3387
3388 rcu_read_lock();
3389 snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
3390 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
3391 rcu_read_unlock();
3392
3393 new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align,
3394 (s->flags & ~SLAB_PANIC), s->ctor, s);
3395
3396 if (new)
3397 new->allocflags |= __GFP_KMEMCG;
3398
3399 return new;
3400}
3401
Glauber Costad7f25f82012-12-18 14:22:40 -08003402static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
3403 struct kmem_cache *cachep)
3404{
3405 struct kmem_cache *new_cachep;
3406 int idx;
3407
3408 BUG_ON(!memcg_can_account_kmem(memcg));
3409
3410 idx = memcg_cache_id(memcg);
3411
3412 mutex_lock(&memcg_cache_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003413 new_cachep = cache_from_memcg_idx(cachep, idx);
Li Zefan20f05312013-07-08 16:00:31 -07003414 if (new_cachep) {
3415 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003416 goto out;
Li Zefan20f05312013-07-08 16:00:31 -07003417 }
Glauber Costad7f25f82012-12-18 14:22:40 -08003418
3419 new_cachep = kmem_cache_dup(memcg, cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003420 if (new_cachep == NULL) {
3421 new_cachep = cachep;
Li Zefan20f05312013-07-08 16:00:31 -07003422 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003423 goto out;
3424 }
3425
Glauber Costa1f458cb2012-12-18 14:22:50 -08003426 atomic_set(&new_cachep->memcg_params->nr_pages , 0);
Glauber Costad7f25f82012-12-18 14:22:40 -08003427
3428 cachep->memcg_params->memcg_caches[idx] = new_cachep;
3429 /*
3430 * the readers won't lock, make sure everybody sees the updated value,
3431 * so they won't put stuff in the queue again for no reason
3432 */
3433 wmb();
3434out:
3435 mutex_unlock(&memcg_cache_mutex);
3436 return new_cachep;
3437}
3438
Glauber Costa7cf27982012-12-18 14:22:55 -08003439void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
3440{
3441 struct kmem_cache *c;
3442 int i;
3443
3444 if (!s->memcg_params)
3445 return;
3446 if (!s->memcg_params->is_root_cache)
3447 return;
3448
3449 /*
3450 * If the cache is being destroyed, we trust that there is no one else
3451 * requesting objects from it. Even if there are, the sanity checks in
3452 * kmem_cache_destroy should caught this ill-case.
3453 *
3454 * Still, we don't want anyone else freeing memcg_caches under our
3455 * noses, which can happen if a new memcg comes to life. As usual,
3456 * we'll take the set_limit_mutex to protect ourselves against this.
3457 */
3458 mutex_lock(&set_limit_mutex);
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003459 for_each_memcg_cache_index(i) {
3460 c = cache_from_memcg_idx(s, i);
Glauber Costa7cf27982012-12-18 14:22:55 -08003461 if (!c)
3462 continue;
3463
3464 /*
3465 * We will now manually delete the caches, so to avoid races
3466 * we need to cancel all pending destruction workers and
3467 * proceed with destruction ourselves.
3468 *
3469 * kmem_cache_destroy() will call kmem_cache_shrink internally,
3470 * and that could spawn the workers again: it is likely that
3471 * the cache still have active pages until this very moment.
3472 * This would lead us back to mem_cgroup_destroy_cache.
3473 *
3474 * But that will not execute at all if the "dead" flag is not
3475 * set, so flip it down to guarantee we are in control.
3476 */
3477 c->memcg_params->dead = false;
Glauber Costa22933152012-12-18 14:22:59 -08003478 cancel_work_sync(&c->memcg_params->destroy);
Glauber Costa7cf27982012-12-18 14:22:55 -08003479 kmem_cache_destroy(c);
3480 }
3481 mutex_unlock(&set_limit_mutex);
3482}
3483
Glauber Costad7f25f82012-12-18 14:22:40 -08003484struct create_work {
3485 struct mem_cgroup *memcg;
3486 struct kmem_cache *cachep;
3487 struct work_struct work;
3488};
3489
Glauber Costa1f458cb2012-12-18 14:22:50 -08003490static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3491{
3492 struct kmem_cache *cachep;
3493 struct memcg_cache_params *params;
3494
3495 if (!memcg_kmem_is_active(memcg))
3496 return;
3497
3498 mutex_lock(&memcg->slab_caches_mutex);
3499 list_for_each_entry(params, &memcg->memcg_slab_caches, list) {
3500 cachep = memcg_params_to_cache(params);
3501 cachep->memcg_params->dead = true;
Glauber Costa1f458cb2012-12-18 14:22:50 -08003502 schedule_work(&cachep->memcg_params->destroy);
3503 }
3504 mutex_unlock(&memcg->slab_caches_mutex);
3505}
3506
Glauber Costad7f25f82012-12-18 14:22:40 -08003507static void memcg_create_cache_work_func(struct work_struct *w)
3508{
3509 struct create_work *cw;
3510
3511 cw = container_of(w, struct create_work, work);
3512 memcg_create_kmem_cache(cw->memcg, cw->cachep);
Glauber Costad7f25f82012-12-18 14:22:40 -08003513 kfree(cw);
3514}
3515
3516/*
3517 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08003518 */
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003519static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3520 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08003521{
3522 struct create_work *cw;
3523
3524 cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT);
Li Zefanca0dde92013-04-29 15:08:57 -07003525 if (cw == NULL) {
3526 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08003527 return;
3528 }
3529
3530 cw->memcg = memcg;
3531 cw->cachep = cachep;
3532
3533 INIT_WORK(&cw->work, memcg_create_cache_work_func);
3534 schedule_work(&cw->work);
3535}
3536
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003537static void memcg_create_cache_enqueue(struct mem_cgroup *memcg,
3538 struct kmem_cache *cachep)
3539{
3540 /*
3541 * We need to stop accounting when we kmalloc, because if the
3542 * corresponding kmalloc cache is not yet created, the first allocation
3543 * in __memcg_create_cache_enqueue will recurse.
3544 *
3545 * However, it is better to enclose the whole function. Depending on
3546 * the debugging options enabled, INIT_WORK(), for instance, can
3547 * trigger an allocation. This too, will make us recurse. Because at
3548 * this point we can't allow ourselves back into memcg_kmem_get_cache,
3549 * the safest choice is to do it like this, wrapping the whole function.
3550 */
3551 memcg_stop_kmem_account();
3552 __memcg_create_cache_enqueue(memcg, cachep);
3553 memcg_resume_kmem_account();
3554}
Glauber Costad7f25f82012-12-18 14:22:40 -08003555/*
3556 * Return the kmem_cache we're supposed to use for a slab allocation.
3557 * We try to use the current memcg's version of the cache.
3558 *
3559 * If the cache does not exist yet, if we are the first user of it,
3560 * we either create it immediately, if possible, or create it asynchronously
3561 * in a workqueue.
3562 * In the latter case, we will let the current allocation go through with
3563 * the original cache.
3564 *
3565 * Can't be called in interrupt context or from kernel threads.
3566 * This function needs to be called with rcu_read_lock() held.
3567 */
3568struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
3569 gfp_t gfp)
3570{
3571 struct mem_cgroup *memcg;
3572 int idx;
3573
3574 VM_BUG_ON(!cachep->memcg_params);
3575 VM_BUG_ON(!cachep->memcg_params->is_root_cache);
3576
Glauber Costa0e9d92f2012-12-18 14:22:42 -08003577 if (!current->mm || current->memcg_kmem_skip_account)
3578 return cachep;
3579
Glauber Costad7f25f82012-12-18 14:22:40 -08003580 rcu_read_lock();
3581 memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner));
Glauber Costad7f25f82012-12-18 14:22:40 -08003582
3583 if (!memcg_can_account_kmem(memcg))
Li Zefanca0dde92013-04-29 15:08:57 -07003584 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003585
3586 idx = memcg_cache_id(memcg);
3587
3588 /*
3589 * barrier to mare sure we're always seeing the up to date value. The
3590 * code updating memcg_caches will issue a write barrier to match this.
3591 */
3592 read_barrier_depends();
Qiang Huang7a67d7a2013-11-12 15:08:24 -08003593 if (likely(cache_from_memcg_idx(cachep, idx))) {
3594 cachep = cache_from_memcg_idx(cachep, idx);
Li Zefanca0dde92013-04-29 15:08:57 -07003595 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08003596 }
3597
Li Zefanca0dde92013-04-29 15:08:57 -07003598 /* The corresponding put will be done in the workqueue. */
3599 if (!css_tryget(&memcg->css))
3600 goto out;
3601 rcu_read_unlock();
3602
3603 /*
3604 * If we are in a safe context (can wait, and not in interrupt
3605 * context), we could be be predictable and return right away.
3606 * This would guarantee that the allocation being performed
3607 * already belongs in the new cache.
3608 *
3609 * However, there are some clashes that can arrive from locking.
3610 * For instance, because we acquire the slab_mutex while doing
3611 * kmem_cache_dup, this means no further allocation could happen
3612 * with the slab_mutex held.
3613 *
3614 * Also, because cache creation issue get_online_cpus(), this
3615 * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex,
3616 * that ends up reversed during cpu hotplug. (cpuset allocates
3617 * a bunch of GFP_KERNEL memory during cpuup). Due to all that,
3618 * better to defer everything.
3619 */
3620 memcg_create_cache_enqueue(memcg, cachep);
3621 return cachep;
3622out:
3623 rcu_read_unlock();
3624 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08003625}
3626EXPORT_SYMBOL(__memcg_kmem_get_cache);
3627
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003628/*
3629 * We need to verify if the allocation against current->mm->owner's memcg is
3630 * possible for the given order. But the page is not allocated yet, so we'll
3631 * need a further commit step to do the final arrangements.
3632 *
3633 * It is possible for the task to switch cgroups in this mean time, so at
3634 * commit time, we can't rely on task conversion any longer. We'll then use
3635 * the handle argument to return to the caller which cgroup we should commit
3636 * against. We could also return the memcg directly and avoid the pointer
3637 * passing, but a boolean return value gives better semantics considering
3638 * the compiled-out case as well.
3639 *
3640 * Returning true means the allocation is possible.
3641 */
3642bool
3643__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
3644{
3645 struct mem_cgroup *memcg;
3646 int ret;
3647
3648 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07003649
3650 /*
3651 * Disabling accounting is only relevant for some specific memcg
3652 * internal allocations. Therefore we would initially not have such
3653 * check here, since direct calls to the page allocator that are marked
3654 * with GFP_KMEMCG only happen outside memcg core. We are mostly
3655 * concerned with cache allocations, and by having this test at
3656 * memcg_kmem_get_cache, we are already able to relay the allocation to
3657 * the root cache and bypass the memcg cache altogether.
3658 *
3659 * There is one exception, though: the SLUB allocator does not create
3660 * large order caches, but rather service large kmallocs directly from
3661 * the page allocator. Therefore, the following sequence when backed by
3662 * the SLUB allocator:
3663 *
Andrew Mortonf894ffa2013-09-12 15:13:35 -07003664 * memcg_stop_kmem_account();
3665 * kmalloc(<large_number>)
3666 * memcg_resume_kmem_account();
Glauber Costa6d42c232013-07-08 16:00:00 -07003667 *
3668 * would effectively ignore the fact that we should skip accounting,
3669 * since it will drive us directly to this function without passing
3670 * through the cache selector memcg_kmem_get_cache. Such large
3671 * allocations are extremely rare but can happen, for instance, for the
3672 * cache arrays. We bring this test here.
3673 */
3674 if (!current->mm || current->memcg_kmem_skip_account)
3675 return true;
3676
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003677 memcg = try_get_mem_cgroup_from_mm(current->mm);
3678
3679 /*
3680 * very rare case described in mem_cgroup_from_task. Unfortunately there
3681 * isn't much we can do without complicating this too much, and it would
3682 * be gfp-dependent anyway. Just let it go
3683 */
3684 if (unlikely(!memcg))
3685 return true;
3686
3687 if (!memcg_can_account_kmem(memcg)) {
3688 css_put(&memcg->css);
3689 return true;
3690 }
3691
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003692 ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order);
3693 if (!ret)
3694 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003695
3696 css_put(&memcg->css);
3697 return (ret == 0);
3698}
3699
3700void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
3701 int order)
3702{
3703 struct page_cgroup *pc;
3704
3705 VM_BUG_ON(mem_cgroup_is_root(memcg));
3706
3707 /* The page allocation failed. Revert */
3708 if (!page) {
3709 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003710 return;
3711 }
3712
3713 pc = lookup_page_cgroup(page);
3714 lock_page_cgroup(pc);
3715 pc->mem_cgroup = memcg;
3716 SetPageCgroupUsed(pc);
3717 unlock_page_cgroup(pc);
3718}
3719
3720void __memcg_kmem_uncharge_pages(struct page *page, int order)
3721{
3722 struct mem_cgroup *memcg = NULL;
3723 struct page_cgroup *pc;
3724
3725
3726 pc = lookup_page_cgroup(page);
3727 /*
3728 * Fast unlocked return. Theoretically might have changed, have to
3729 * check again after locking.
3730 */
3731 if (!PageCgroupUsed(pc))
3732 return;
3733
3734 lock_page_cgroup(pc);
3735 if (PageCgroupUsed(pc)) {
3736 memcg = pc->mem_cgroup;
3737 ClearPageCgroupUsed(pc);
3738 }
3739 unlock_page_cgroup(pc);
3740
3741 /*
3742 * We trust that only if there is a memcg associated with the page, it
3743 * is a valid allocation
3744 */
3745 if (!memcg)
3746 return;
3747
3748 VM_BUG_ON(mem_cgroup_is_root(memcg));
3749 memcg_uncharge_kmem(memcg, PAGE_SIZE << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003750}
Glauber Costa1f458cb2012-12-18 14:22:50 -08003751#else
3752static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg)
3753{
3754}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003755#endif /* CONFIG_MEMCG_KMEM */
3756
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003757#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3758
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07003759#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003760/*
3761 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003762 * zone->lru_lock, 'splitting on pmd' and compound_lock.
3763 * charge/uncharge will be never happen and move_account() is done under
3764 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003765 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003766void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003767{
3768 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003769 struct page_cgroup *pc;
David Rientjesb070e652013-05-07 16:18:09 -07003770 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003771 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003772
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003773 if (mem_cgroup_disabled())
3774 return;
David Rientjesb070e652013-05-07 16:18:09 -07003775
3776 memcg = head_pc->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003777 for (i = 1; i < HPAGE_PMD_NR; i++) {
3778 pc = head_pc + i;
David Rientjesb070e652013-05-07 16:18:09 -07003779 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003780 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003781 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
3782 }
David Rientjesb070e652013-05-07 16:18:09 -07003783 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
3784 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003785}
Hugh Dickins12d27102012-01-12 17:19:52 -08003786#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003787
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003788static inline
3789void mem_cgroup_move_account_page_stat(struct mem_cgroup *from,
3790 struct mem_cgroup *to,
3791 unsigned int nr_pages,
3792 enum mem_cgroup_stat_index idx)
3793{
3794 /* Update stat data for mem_cgroup */
3795 preempt_disable();
Greg Thelen5e8cfc32013-10-30 13:56:21 -07003796 __this_cpu_sub(from->stat->count[idx], nr_pages);
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003797 __this_cpu_add(to->stat->count[idx], nr_pages);
3798 preempt_enable();
3799}
3800
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003801/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07003802 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07003803 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003804 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003805 * @pc: page_cgroup of the page.
3806 * @from: mem_cgroup which the page is moved from.
3807 * @to: mem_cgroup which the page is moved to. @from != @to.
3808 *
3809 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003810 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003811 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003812 *
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003813 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
3814 * from old cgroup.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003815 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003816static int mem_cgroup_move_account(struct page *page,
3817 unsigned int nr_pages,
3818 struct page_cgroup *pc,
3819 struct mem_cgroup *from,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003820 struct mem_cgroup *to)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003821{
Johannes Weinerde3638d2011-03-23 16:42:28 -07003822 unsigned long flags;
3823 int ret;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07003824 bool anon = PageAnon(page);
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003825
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003826 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07003827 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07003828 /*
3829 * The page is isolated from LRU. So, collapse function
3830 * will not handle this page. But page splitting can happen.
3831 * Do this check under compound_page_lock(). The caller should
3832 * hold it.
3833 */
3834 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003835 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07003836 goto out;
3837
3838 lock_page_cgroup(pc);
3839
3840 ret = -EINVAL;
3841 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
3842 goto unlock;
3843
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003844 move_lock_mem_cgroup(from, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003845
Sha Zhengju3ea67d02013-09-12 15:13:53 -07003846 if (!anon && page_mapped(page))
3847 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3848 MEM_CGROUP_STAT_FILE_MAPPED);
3849
3850 if (PageWriteback(page))
3851 mem_cgroup_move_account_page_stat(from, to, nr_pages,
3852 MEM_CGROUP_STAT_WRITEBACK);
3853
David Rientjesb070e652013-05-07 16:18:09 -07003854 mem_cgroup_charge_statistics(from, page, anon, -nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07003855
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003856 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003857 pc->mem_cgroup = to;
David Rientjesb070e652013-05-07 16:18:09 -07003858 mem_cgroup_charge_statistics(to, page, anon, nr_pages);
KAMEZAWA Hiroyuki312734c02012-03-21 16:34:24 -07003859 move_unlock_mem_cgroup(from, &flags);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003860 ret = 0;
3861unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003862 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08003863 /*
3864 * check events
3865 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003866 memcg_check_events(to, page);
3867 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07003868out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003869 return ret;
3870}
3871
Michal Hocko2ef37d32012-10-26 13:37:30 +02003872/**
3873 * mem_cgroup_move_parent - moves page to the parent group
3874 * @page: the page to move
3875 * @pc: page_cgroup of the page
3876 * @child: page's cgroup
3877 *
3878 * move charges to its parent or the root cgroup if the group has no
3879 * parent (aka use_hierarchy==0).
3880 * Although this might fail (get_page_unless_zero, isolate_lru_page or
3881 * mem_cgroup_move_account fails) the failure is always temporary and
3882 * it signals a race with a page removal/uncharge or migration. In the
3883 * first case the page is on the way out and it will vanish from the LRU
3884 * on the next attempt and the call should be retried later.
3885 * Isolation from the LRU fails only if page has been isolated from
3886 * the LRU since we looked at it and that usually means either global
3887 * reclaim or migration going on. The page will either get back to the
3888 * LRU or vanish.
3889 * Finaly mem_cgroup_move_account fails only if the page got uncharged
3890 * (!PageCgroupUsed) or moved to a different group. The page will
3891 * disappear in the next attempt.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003892 */
Johannes Weiner5564e882011-03-23 16:42:29 -07003893static int mem_cgroup_move_parent(struct page *page,
3894 struct page_cgroup *pc,
KAMEZAWA Hiroyuki6068bf02012-07-31 16:42:45 -07003895 struct mem_cgroup *child)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003896{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003897 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003898 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07003899 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003900 int ret;
3901
Michal Hockod8423012012-10-26 13:37:29 +02003902 VM_BUG_ON(mem_cgroup_is_root(child));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003903
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003904 ret = -EBUSY;
3905 if (!get_page_unless_zero(page))
3906 goto out;
3907 if (isolate_lru_page(page))
3908 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08003909
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003910 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003911
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003912 parent = parent_mem_cgroup(child);
3913 /*
3914 * If no parent, move charges to root cgroup.
3915 */
3916 if (!parent)
3917 parent = root_mem_cgroup;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003918
Michal Hocko2ef37d32012-10-26 13:37:30 +02003919 if (nr_pages > 1) {
3920 VM_BUG_ON(!PageTransHuge(page));
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003921 flags = compound_lock_irqsave(page);
Michal Hocko2ef37d32012-10-26 13:37:30 +02003922 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003923
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003924 ret = mem_cgroup_move_account(page, nr_pages,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07003925 pc, child, parent);
KAMEZAWA Hiroyukicc926f72012-05-29 15:07:04 -07003926 if (!ret)
3927 __mem_cgroup_cancel_local_charge(child, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08003928
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003929 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08003930 compound_unlock_irqrestore(page, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003931 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003932put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08003933 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08003934out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003935 return ret;
3936}
3937
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003938/*
3939 * Charge the memory controller for page usage.
3940 * Return
3941 * 0 if the charge was successful
3942 * < 0 if the cgroup is over its limit
3943 */
3944static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07003945 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003946{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003947 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003948 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08003949 bool oom = true;
3950 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08003951
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003952 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003953 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003954 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08003955 /*
3956 * Never OOM-kill a process for a huge page. The
3957 * fault handler will fall back to regular pages.
3958 */
3959 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003960 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003961
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003962 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003963 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003964 return ret;
Johannes Weinerce587e62012-04-24 20:22:33 +02003965 __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003966 return 0;
3967}
3968
3969int mem_cgroup_newpage_charge(struct page *page,
3970 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003971{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003972 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07003973 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08003974 VM_BUG_ON(page_mapped(page));
3975 VM_BUG_ON(page->mapping && !PageAnon(page));
3976 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003977 return mem_cgroup_charge_common(page, mm, gfp_mask,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07003978 MEM_CGROUP_CHARGE_TYPE_ANON);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08003979}
3980
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003981/*
3982 * While swap-in, try_charge -> commit or cancel, the page is locked.
3983 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02003984 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003985 * "commit()" or removed by "cancel()"
3986 */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07003987static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
3988 struct page *page,
3989 gfp_t mask,
3990 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003991{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003992 struct mem_cgroup *memcg;
Johannes Weiner90deb782012-07-31 16:45:47 -07003993 struct page_cgroup *pc;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08003994 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003995
Johannes Weiner90deb782012-07-31 16:45:47 -07003996 pc = lookup_page_cgroup(page);
3997 /*
3998 * Every swap fault against a single page tries to charge the
3999 * page, bail as early as possible. shmem_unuse() encounters
4000 * already charged pages, too. The USED bit is protected by
4001 * the page lock, which serializes swap cache removal, which
4002 * in turn serializes uncharging.
4003 */
4004 if (PageCgroupUsed(pc))
4005 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004006 if (!do_swap_account)
4007 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004008 memcg = try_get_mem_cgroup_from_page(page);
4009 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08004010 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08004011 *memcgp = memcg;
4012 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004013 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004014 if (ret == -EINTR)
4015 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08004016 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004017charge_cur_mm:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004018 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
4019 if (ret == -EINTR)
4020 ret = 0;
4021 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004022}
4023
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004024int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
4025 gfp_t gfp_mask, struct mem_cgroup **memcgp)
4026{
4027 *memcgp = NULL;
4028 if (mem_cgroup_disabled())
4029 return 0;
Johannes Weinerbdf4f4d2012-07-31 16:45:50 -07004030 /*
4031 * A racing thread's fault, or swapoff, may have already
4032 * updated the pte, and even removed page from swap cache: in
4033 * those cases unuse_pte()'s pte_same() test will fail; but
4034 * there's also a KSM case which does need to charge the page.
4035 */
4036 if (!PageSwapCache(page)) {
4037 int ret;
4038
4039 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
4040 if (ret == -EINTR)
4041 ret = 0;
4042 return ret;
4043 }
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004044 return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
4045}
4046
Johannes Weiner827a03d2012-07-31 16:45:36 -07004047void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
4048{
4049 if (mem_cgroup_disabled())
4050 return;
4051 if (!memcg)
4052 return;
4053 __mem_cgroup_cancel_charge(memcg, 1);
4054}
4055
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004056static void
Johannes Weiner72835c82012-01-12 17:18:32 -08004057__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004058 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004059{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004060 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004061 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08004062 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004063 return;
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07004064
Johannes Weinerce587e62012-04-24 20:22:33 +02004065 __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004066 /*
4067 * Now swap is on-memory. This means this page may be
4068 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004069 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
4070 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
4071 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004072 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08004073 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004074 swp_entry_t ent = {.val = page_private(page)};
Hugh Dickins86493002012-05-29 15:06:52 -07004075 mem_cgroup_uncharge_swap(ent);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004076 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004077}
4078
Johannes Weiner72835c82012-01-12 17:18:32 -08004079void mem_cgroup_commit_charge_swapin(struct page *page,
4080 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004081{
Johannes Weiner72835c82012-01-12 17:18:32 -08004082 __mem_cgroup_commit_charge_swapin(page, memcg,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004083 MEM_CGROUP_CHARGE_TYPE_ANON);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07004084}
4085
Johannes Weiner827a03d2012-07-31 16:45:36 -07004086int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
4087 gfp_t gfp_mask)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004088{
Johannes Weiner827a03d2012-07-31 16:45:36 -07004089 struct mem_cgroup *memcg = NULL;
4090 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
4091 int ret;
4092
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004093 if (mem_cgroup_disabled())
Johannes Weiner827a03d2012-07-31 16:45:36 -07004094 return 0;
4095 if (PageCompound(page))
4096 return 0;
4097
Johannes Weiner827a03d2012-07-31 16:45:36 -07004098 if (!PageSwapCache(page))
4099 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
4100 else { /* page is swapcache/shmem */
Johannes Weiner0435a2f2012-07-31 16:45:43 -07004101 ret = __mem_cgroup_try_charge_swapin(mm, page,
4102 gfp_mask, &memcg);
Johannes Weiner827a03d2012-07-31 16:45:36 -07004103 if (!ret)
4104 __mem_cgroup_commit_charge_swapin(page, memcg, type);
4105 }
4106 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004107}
4108
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004109static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004110 unsigned int nr_pages,
4111 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004112{
4113 struct memcg_batch_info *batch = NULL;
4114 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004115
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004116 /* If swapout, usage of swap doesn't decrease */
4117 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
4118 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004119
4120 batch = &current->memcg_batch;
4121 /*
4122 * In usual, we do css_get() when we remember memcg pointer.
4123 * But in this case, we keep res->usage until end of a series of
4124 * uncharges. Then, it's ok to ignore memcg's refcnt.
4125 */
4126 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004127 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004128 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004129 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004130 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004131 * the same cgroup and we have chance to coalesce uncharges.
4132 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
4133 * because we want to do uncharge as soon as possible.
4134 */
4135
4136 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
4137 goto direct_uncharge;
4138
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004139 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08004140 goto direct_uncharge;
4141
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004142 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004143 * In typical case, batch->memcg == mem. This means we can
4144 * merge a series of uncharges to an uncharge of res_counter.
4145 * If not, we uncharge res_counter ony by one.
4146 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004147 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004148 goto direct_uncharge;
4149 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004150 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004151 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004152 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004153 return;
4154direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004155 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004156 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004157 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
4158 if (unlikely(batch->memcg != memcg))
4159 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004160}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08004161
Balbir Singh8697d332008-02-07 00:13:59 -08004162/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004163 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004164 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004165static struct mem_cgroup *
Johannes Weiner0030f532012-07-31 16:45:25 -07004166__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
4167 bool end_migration)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004168{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004169 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004170 unsigned int nr_pages = 1;
4171 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004172 bool anon;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004173
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004174 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004175 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07004176
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004177 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004178 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08004179 VM_BUG_ON(!PageTransHuge(page));
4180 }
Balbir Singh8697d332008-02-07 00:13:59 -08004181 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08004182 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08004183 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004184 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004185 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004186 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08004187
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004188 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004189
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004190 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004191
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004192 if (!PageCgroupUsed(pc))
4193 goto unlock_out;
4194
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004195 anon = PageAnon(page);
4196
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004197 switch (ctype) {
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004198 case MEM_CGROUP_CHARGE_TYPE_ANON:
KAMEZAWA Hiroyuki2ff76f12012-03-21 16:34:25 -07004199 /*
4200 * Generally PageAnon tells if it's the anon statistics to be
4201 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
4202 * used before page reached the stage of being marked PageAnon.
4203 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004204 anon = true;
4205 /* fallthrough */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004206 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004207 /* See mem_cgroup_prepare_migration() */
Johannes Weiner0030f532012-07-31 16:45:25 -07004208 if (page_mapped(page))
4209 goto unlock_out;
4210 /*
4211 * Pages under migration may not be uncharged. But
4212 * end_migration() /must/ be the one uncharging the
4213 * unused post-migration page and so it has to call
4214 * here with the migration bit still set. See the
4215 * res_counter handling below.
4216 */
4217 if (!end_migration && PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004218 goto unlock_out;
4219 break;
4220 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
4221 if (!PageAnon(page)) { /* Shared memory */
4222 if (page->mapping && !page_is_file_cache(page))
4223 goto unlock_out;
4224 } else if (page_mapped(page)) /* Anon */
4225 goto unlock_out;
4226 break;
4227 default:
4228 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004229 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004230
David Rientjesb070e652013-05-07 16:18:09 -07004231 mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004232
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004233 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08004234 /*
4235 * pc->mem_cgroup is not cleared here. It will be accessed when it's
4236 * freed from LRU. This is safe because uncharged page is expected not
4237 * to be reused (freed soon). Exception is SwapCache, it's handled by
4238 * special functions.
4239 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004240
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004241 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004242 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004243 * even after unlock, we have memcg->res.usage here and this memcg
Li Zefan40503772013-07-08 16:00:34 -07004244 * will never be freed, so it's safe to call css_get().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004245 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004246 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004247 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004248 mem_cgroup_swap_statistics(memcg, true);
Li Zefan40503772013-07-08 16:00:34 -07004249 css_get(&memcg->css);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004250 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004251 /*
4252 * Migration does not charge the res_counter for the
4253 * replacement page, so leave it alone when phasing out the
4254 * page that is unused after the migration.
4255 */
4256 if (!end_migration && !mem_cgroup_is_root(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004257 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004258
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004259 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004260
4261unlock_out:
4262 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004263 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08004264}
4265
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004266void mem_cgroup_uncharge_page(struct page *page)
4267{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004268 /* early check. */
4269 if (page_mapped(page))
4270 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08004271 VM_BUG_ON(page->mapping && !PageAnon(page));
Johannes Weiner28ccddf2013-05-24 15:55:15 -07004272 /*
4273 * If the page is in swap cache, uncharge should be deferred
4274 * to the swap path, which also properly accounts swap usage
4275 * and handles memcg lifetime.
4276 *
4277 * Note that this check is not stable and reclaim may add the
4278 * page to swap cache at any time after this. However, if the
4279 * page is not in swap cache by the time page->mapcount hits
4280 * 0, there won't be any page table references to the swap
4281 * slot, and reclaim will free it and not actually write the
4282 * page to disk.
4283 */
Johannes Weiner0c59b892012-07-31 16:45:31 -07004284 if (PageSwapCache(page))
4285 return;
Johannes Weiner0030f532012-07-31 16:45:25 -07004286 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004287}
4288
4289void mem_cgroup_uncharge_cache_page(struct page *page)
4290{
4291 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07004292 VM_BUG_ON(page->mapping);
Johannes Weiner0030f532012-07-31 16:45:25 -07004293 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004294}
4295
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004296/*
4297 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
4298 * In that cases, pages are freed continuously and we can expect pages
4299 * are in the same memcg. All these calls itself limits the number of
4300 * pages freed at once, then uncharge_start/end() is called properly.
4301 * This may be called prural(2) times in a context,
4302 */
4303
4304void mem_cgroup_uncharge_start(void)
4305{
4306 current->memcg_batch.do_batch++;
4307 /* We can do nest. */
4308 if (current->memcg_batch.do_batch == 1) {
4309 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004310 current->memcg_batch.nr_pages = 0;
4311 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004312 }
4313}
4314
4315void mem_cgroup_uncharge_end(void)
4316{
4317 struct memcg_batch_info *batch = &current->memcg_batch;
4318
4319 if (!batch->do_batch)
4320 return;
4321
4322 batch->do_batch--;
4323 if (batch->do_batch) /* If stacked, do nothing. */
4324 return;
4325
4326 if (!batch->memcg)
4327 return;
4328 /*
4329 * This "batch->memcg" is valid without any css_get/put etc...
4330 * bacause we hide charges behind us.
4331 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07004332 if (batch->nr_pages)
4333 res_counter_uncharge(&batch->memcg->res,
4334 batch->nr_pages * PAGE_SIZE);
4335 if (batch->memsw_nr_pages)
4336 res_counter_uncharge(&batch->memcg->memsw,
4337 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004338 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08004339 /* forget this pointer (for sanity check) */
4340 batch->memcg = NULL;
4341}
4342
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004343#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004344/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004345 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004346 * memcg information is recorded to swap_cgroup of "ent"
4347 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004348void
4349mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004350{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004351 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004352 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004353
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004354 if (!swapout) /* this was a swap cache but the swap is unused ! */
4355 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
4356
Johannes Weiner0030f532012-07-31 16:45:25 -07004357 memcg = __mem_cgroup_uncharge_common(page, ctype, false);
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07004358
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004359 /*
4360 * record memcg information, if swapout && memcg != NULL,
Li Zefan40503772013-07-08 16:00:34 -07004361 * css_get() was called in uncharge().
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07004362 */
4363 if (do_swap_account && swapout && memcg)
Li Zefan34c00c32013-09-23 16:56:01 +08004364 swap_cgroup_record(ent, mem_cgroup_id(memcg));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004365}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07004366#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08004367
Andrew Mortonc255a452012-07-31 16:43:02 -07004368#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004369/*
4370 * called from swap_entry_free(). remove record in swap_cgroup and
4371 * uncharge "memsw" account.
4372 */
4373void mem_cgroup_uncharge_swap(swp_entry_t ent)
4374{
4375 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004376 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004377
4378 if (!do_swap_account)
4379 return;
4380
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004381 id = swap_cgroup_record(ent, 0);
4382 rcu_read_lock();
4383 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004384 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004385 /*
4386 * We uncharge this because swap is freed.
4387 * This memcg can be obsolete one. We avoid calling css_tryget
4388 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004389 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07004390 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07004391 mem_cgroup_swap_statistics(memcg, false);
Li Zefan40503772013-07-08 16:00:34 -07004392 css_put(&memcg->css);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004393 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07004394 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004395}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004396
4397/**
4398 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
4399 * @entry: swap entry to be moved
4400 * @from: mem_cgroup which the entry is moved from
4401 * @to: mem_cgroup which the entry is moved to
4402 *
4403 * It succeeds only when the swap_cgroup's record for this entry is the same
4404 * as the mem_cgroup's id of @from.
4405 *
4406 * Returns 0 on success, -EINVAL on failure.
4407 *
4408 * The caller must have charged to @to, IOW, called res_counter_charge() about
4409 * both res and memsw, and called css_get().
4410 */
4411static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004412 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004413{
4414 unsigned short old_id, new_id;
4415
Li Zefan34c00c32013-09-23 16:56:01 +08004416 old_id = mem_cgroup_id(from);
4417 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004418
4419 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004420 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004421 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004422 /*
4423 * This function is only called from task migration context now.
4424 * It postpones res_counter and refcount handling till the end
4425 * of task migration(mem_cgroup_clear_mc()) for performance
Li Zefan40503772013-07-08 16:00:34 -07004426 * improvement. But we cannot postpone css_get(to) because if
4427 * the process that has been moved to @to does swap-in, the
4428 * refcount of @to might be decreased to 0.
4429 *
4430 * We are in attach() phase, so the cgroup is guaranteed to be
4431 * alive, so we can just call css_get().
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004432 */
Li Zefan40503772013-07-08 16:00:34 -07004433 css_get(&to->css);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004434 return 0;
4435 }
4436 return -EINVAL;
4437}
4438#else
4439static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004440 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08004441{
4442 return -EINVAL;
4443}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004444#endif
4445
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004446/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004447 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
4448 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004449 */
Johannes Weiner0030f532012-07-31 16:45:25 -07004450void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
4451 struct mem_cgroup **memcgp)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004452{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004453 struct mem_cgroup *memcg = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00004454 unsigned int nr_pages = 1;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07004455 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004456 enum charge_type ctype;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004457
Johannes Weiner72835c82012-01-12 17:18:32 -08004458 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07004459
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004460 if (mem_cgroup_disabled())
Johannes Weiner0030f532012-07-31 16:45:25 -07004461 return;
Balbir Singh40779602008-04-04 14:29:59 -07004462
Mel Gormanb32967f2012-11-19 12:35:47 +00004463 if (PageTransHuge(page))
4464 nr_pages <<= compound_order(page);
4465
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004466 pc = lookup_page_cgroup(page);
4467 lock_page_cgroup(pc);
4468 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004469 memcg = pc->mem_cgroup;
4470 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004471 /*
4472 * At migrating an anonymous page, its mapcount goes down
4473 * to 0 and uncharge() will be called. But, even if it's fully
4474 * unmapped, migration may fail and this page has to be
4475 * charged again. We set MIGRATION flag here and delay uncharge
4476 * until end_migration() is called
4477 *
4478 * Corner Case Thinking
4479 * A)
4480 * When the old page was mapped as Anon and it's unmap-and-freed
4481 * while migration was ongoing.
4482 * If unmap finds the old page, uncharge() of it will be delayed
4483 * until end_migration(). If unmap finds a new page, it's
4484 * uncharged when it make mapcount to be 1->0. If unmap code
4485 * finds swap_migration_entry, the new page will not be mapped
4486 * and end_migration() will find it(mapcount==0).
4487 *
4488 * B)
4489 * When the old page was mapped but migraion fails, the kernel
4490 * remaps it. A charge for it is kept by MIGRATION flag even
4491 * if mapcount goes down to 0. We can do remap successfully
4492 * without charging it again.
4493 *
4494 * C)
4495 * The "old" page is under lock_page() until the end of
4496 * migration, so, the old page itself will not be swapped-out.
4497 * If the new page is swapped out before end_migraton, our
4498 * hook to usual swap-out path will catch the event.
4499 */
4500 if (PageAnon(page))
4501 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08004502 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004503 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004504 /*
4505 * If the page is not charged at this point,
4506 * we return here.
4507 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004508 if (!memcg)
Johannes Weiner0030f532012-07-31 16:45:25 -07004509 return;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004510
Johannes Weiner72835c82012-01-12 17:18:32 -08004511 *memcgp = memcg;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004512 /*
4513 * We charge new page before it's used/mapped. So, even if unlock_page()
4514 * is called before end_migration, we can catch all events on this new
4515 * page. In the case new page is migrated but not remapped, new page's
4516 * mapcount will be finally 0 and we call uncharge in end_migration().
4517 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004518 if (PageAnon(page))
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -07004519 ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004520 else
Johannes Weiner62ba7442012-07-31 16:45:39 -07004521 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
Johannes Weiner0030f532012-07-31 16:45:25 -07004522 /*
4523 * The page is committed to the memcg, but it's not actually
4524 * charged to the res_counter since we plan on replacing the
4525 * old one and only one page is going to be left afterwards.
4526 */
Mel Gormanb32967f2012-11-19 12:35:47 +00004527 __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004528}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08004529
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004530/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004531void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004532 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07004533{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004534 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004535 struct page_cgroup *pc;
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004536 bool anon;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004537
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004538 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004539 return;
Tejun Heob25ed602012-11-05 09:16:59 -08004540
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08004541 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004542 used = oldpage;
4543 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004544 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004545 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004546 unused = oldpage;
4547 }
Johannes Weiner0030f532012-07-31 16:45:25 -07004548 anon = PageAnon(used);
Johannes Weiner7d188952012-07-31 16:45:34 -07004549 __mem_cgroup_uncharge_common(unused,
4550 anon ? MEM_CGROUP_CHARGE_TYPE_ANON
4551 : MEM_CGROUP_CHARGE_TYPE_CACHE,
4552 true);
Johannes Weiner0030f532012-07-31 16:45:25 -07004553 css_put(&memcg->css);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004554 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004555 * We disallowed uncharge of pages under migration because mapcount
4556 * of the page goes down to zero, temporarly.
4557 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07004558 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004559 pc = lookup_page_cgroup(oldpage);
4560 lock_page_cgroup(pc);
4561 ClearPageCgroupMigration(pc);
4562 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004563
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004564 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004565 * If a page is a file cache, radix-tree replacement is very atomic
4566 * and we can skip this check. When it was an Anon page, its mapcount
4567 * goes down to 0. But because we added MIGRATION flage, it's not
4568 * uncharged yet. There are several case but page->mapcount check
4569 * and USED bit check in mem_cgroup_uncharge_page() will do enough
4570 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08004571 */
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -07004572 if (anon)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07004573 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08004574}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004575
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004576/*
4577 * At replace page cache, newpage is not under any memcg but it's on
4578 * LRU. So, this function doesn't touch res_counter but handles LRU
4579 * in correct way. Both pages are locked so we cannot race with uncharge.
4580 */
4581void mem_cgroup_replace_page_cache(struct page *oldpage,
4582 struct page *newpage)
4583{
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004584 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004585 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004586 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004587
4588 if (mem_cgroup_disabled())
4589 return;
4590
4591 pc = lookup_page_cgroup(oldpage);
4592 /* fix accounting on old pages */
4593 lock_page_cgroup(pc);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004594 if (PageCgroupUsed(pc)) {
4595 memcg = pc->mem_cgroup;
David Rientjesb070e652013-05-07 16:18:09 -07004596 mem_cgroup_charge_statistics(memcg, oldpage, false, -1);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004597 ClearPageCgroupUsed(pc);
4598 }
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004599 unlock_page_cgroup(pc);
4600
Hugh Dickinsbde05d12012-05-29 15:06:38 -07004601 /*
4602 * When called from shmem_replace_page(), in some cases the
4603 * oldpage has already been charged, and in some cases not.
4604 */
4605 if (!memcg)
4606 return;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004607 /*
4608 * Even if newpage->mapping was NULL before starting replacement,
4609 * the newpage may be on LRU(or pagevec for LRU) already. We lock
4610 * LRU while we overwrite pc->mem_cgroup.
4611 */
Johannes Weinerce587e62012-04-24 20:22:33 +02004612 __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08004613}
4614
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004615#ifdef CONFIG_DEBUG_VM
4616static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
4617{
4618 struct page_cgroup *pc;
4619
4620 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08004621 /*
4622 * Can be NULL while feeding pages into the page allocator for
4623 * the first time, i.e. during boot or memory hotplug;
4624 * or when mem_cgroup_disabled().
4625 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004626 if (likely(pc) && PageCgroupUsed(pc))
4627 return pc;
4628 return NULL;
4629}
4630
4631bool mem_cgroup_bad_page_check(struct page *page)
4632{
4633 if (mem_cgroup_disabled())
4634 return false;
4635
4636 return lookup_page_cgroup_used(page) != NULL;
4637}
4638
4639void mem_cgroup_print_bad_page(struct page *page)
4640{
4641 struct page_cgroup *pc;
4642
4643 pc = lookup_page_cgroup_used(page);
4644 if (pc) {
Andrew Mortond0451972013-02-22 16:32:06 -08004645 pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
4646 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07004647 }
4648}
4649#endif
4650
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08004651static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004652 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004653{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004654 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004655 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004656 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004657 int children = mem_cgroup_count_children(memcg);
4658 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004659 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004660
4661 /*
4662 * For keeping hierarchical_reclaim simple, how long we should retry
4663 * is depends on callers. We set our retry-count to be function
4664 * of # of children which we should visit in this loop.
4665 */
4666 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
4667
4668 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004669
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004670 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004671 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004672 if (signal_pending(current)) {
4673 ret = -EINTR;
4674 break;
4675 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004676 /*
4677 * Rather than hide all in some function, I do this in
4678 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004679 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004680 */
4681 mutex_lock(&set_limit_mutex);
4682 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4683 if (memswlimit < val) {
4684 ret = -EINVAL;
4685 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004686 break;
4687 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004688
4689 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4690 if (memlimit < val)
4691 enlarge = 1;
4692
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004693 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004694 if (!ret) {
4695 if (memswlimit == val)
4696 memcg->memsw_is_minimum = true;
4697 else
4698 memcg->memsw_is_minimum = false;
4699 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004700 mutex_unlock(&set_limit_mutex);
4701
4702 if (!ret)
4703 break;
4704
Johannes Weiner56600482012-01-12 17:17:59 -08004705 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4706 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004707 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
4708 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004709 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004710 retry_count--;
4711 else
4712 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004713 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004714 if (!ret && enlarge)
4715 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08004716
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004717 return ret;
4718}
4719
Li Zefan338c8432009-06-17 16:27:15 -07004720static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
4721 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004722{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004723 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004724 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004725 int children = mem_cgroup_count_children(memcg);
4726 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004727 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004728
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004729 /* see mem_cgroup_resize_res_limit */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07004730 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004731 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004732 while (retry_count) {
4733 if (signal_pending(current)) {
4734 ret = -EINTR;
4735 break;
4736 }
4737 /*
4738 * Rather than hide all in some function, I do this in
4739 * open coded manner. You see what this really does.
Wanpeng Liaaad1532012-07-31 16:43:23 -07004740 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004741 */
4742 mutex_lock(&set_limit_mutex);
4743 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4744 if (memlimit > val) {
4745 ret = -EINVAL;
4746 mutex_unlock(&set_limit_mutex);
4747 break;
4748 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004749 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4750 if (memswlimit < val)
4751 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004752 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07004753 if (!ret) {
4754 if (memlimit == val)
4755 memcg->memsw_is_minimum = true;
4756 else
4757 memcg->memsw_is_minimum = false;
4758 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004759 mutex_unlock(&set_limit_mutex);
4760
4761 if (!ret)
4762 break;
4763
Johannes Weiner56600482012-01-12 17:17:59 -08004764 mem_cgroup_reclaim(memcg, GFP_KERNEL,
4765 MEM_CGROUP_RECLAIM_NOSWAP |
4766 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004767 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004768 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004769 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004770 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07004771 else
4772 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004773 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004774 if (!ret && enlarge)
4775 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004776 return ret;
4777}
4778
Andrew Morton0608f432013-09-24 15:27:41 -07004779unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
4780 gfp_t gfp_mask,
4781 unsigned long *total_scanned)
4782{
4783 unsigned long nr_reclaimed = 0;
4784 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
4785 unsigned long reclaimed;
4786 int loop = 0;
4787 struct mem_cgroup_tree_per_zone *mctz;
4788 unsigned long long excess;
4789 unsigned long nr_scanned;
4790
4791 if (order > 0)
4792 return 0;
4793
4794 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
4795 /*
4796 * This loop can run a while, specially if mem_cgroup's continuously
4797 * keep exceeding their soft limit and putting the system under
4798 * pressure
4799 */
4800 do {
4801 if (next_mz)
4802 mz = next_mz;
4803 else
4804 mz = mem_cgroup_largest_soft_limit_node(mctz);
4805 if (!mz)
4806 break;
4807
4808 nr_scanned = 0;
4809 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
4810 gfp_mask, &nr_scanned);
4811 nr_reclaimed += reclaimed;
4812 *total_scanned += nr_scanned;
4813 spin_lock(&mctz->lock);
4814
4815 /*
4816 * If we failed to reclaim anything from this memory cgroup
4817 * it is time to move on to the next cgroup
4818 */
4819 next_mz = NULL;
4820 if (!reclaimed) {
4821 do {
4822 /*
4823 * Loop until we find yet another one.
4824 *
4825 * By the time we get the soft_limit lock
4826 * again, someone might have aded the
4827 * group back on the RB tree. Iterate to
4828 * make sure we get a different mem.
4829 * mem_cgroup_largest_soft_limit_node returns
4830 * NULL if no other cgroup is present on
4831 * the tree
4832 */
4833 next_mz =
4834 __mem_cgroup_largest_soft_limit_node(mctz);
4835 if (next_mz == mz)
4836 css_put(&next_mz->memcg->css);
4837 else /* next_mz == NULL or other memcg */
4838 break;
4839 } while (1);
4840 }
4841 __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
4842 excess = res_counter_soft_limit_excess(&mz->memcg->res);
4843 /*
4844 * One school of thought says that we should not add
4845 * back the node to the tree if reclaim returns 0.
4846 * But our reclaim could return 0, simply because due
4847 * to priority we are exposing a smaller subset of
4848 * memory to reclaim from. Consider this as a longer
4849 * term TODO.
4850 */
4851 /* If excess == 0, no tree ops */
4852 __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
4853 spin_unlock(&mctz->lock);
4854 css_put(&mz->memcg->css);
4855 loop++;
4856 /*
4857 * Could not reclaim anything and there are no more
4858 * mem cgroups to try or we seem to be looping without
4859 * reclaiming anything.
4860 */
4861 if (!nr_reclaimed &&
4862 (next_mz == NULL ||
4863 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
4864 break;
4865 } while (!nr_reclaimed);
4866 if (next_mz)
4867 css_put(&next_mz->memcg->css);
4868 return nr_reclaimed;
4869}
4870
Michal Hocko2ef37d32012-10-26 13:37:30 +02004871/**
4872 * mem_cgroup_force_empty_list - clears LRU of a group
4873 * @memcg: group to clear
4874 * @node: NUMA node
4875 * @zid: zone id
4876 * @lru: lru to to clear
4877 *
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004878 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
Michal Hocko2ef37d32012-10-26 13:37:30 +02004879 * reclaim the pages page themselves - pages are moved to the parent (or root)
4880 * group.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004881 */
Michal Hocko2ef37d32012-10-26 13:37:30 +02004882static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004883 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004884{
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004885 struct lruvec *lruvec;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004886 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004887 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08004888 struct page *busy;
4889 struct zone *zone;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08004890
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004891 zone = &NODE_DATA(node)->node_zones[zid];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004892 lruvec = mem_cgroup_zone_lruvec(zone, memcg);
4893 list = &lruvec->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004894
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004895 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004896 do {
Johannes Weiner925b7672012-01-12 17:18:15 -08004897 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07004898 struct page *page;
4899
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004900 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004901 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004902 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004903 break;
4904 }
Johannes Weiner925b7672012-01-12 17:18:15 -08004905 page = list_entry(list->prev, struct page, lru);
4906 if (busy == page) {
4907 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08004908 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004909 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004910 continue;
4911 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004912 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004913
Johannes Weiner925b7672012-01-12 17:18:15 -08004914 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07004915
KAMEZAWA Hiroyuki3c935d12012-07-31 16:42:46 -07004916 if (mem_cgroup_move_parent(page, pc, memcg)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004917 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08004918 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004919 cond_resched();
4920 } else
4921 busy = NULL;
Michal Hocko2ef37d32012-10-26 13:37:30 +02004922 } while (!list_empty(list));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004923}
4924
4925/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004926 * make mem_cgroup's charge to be 0 if there is no task by moving
4927 * all the charges and pages to the parent.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004928 * This enables deleting this mem_cgroup.
Michal Hockoc26251f2012-10-26 13:37:28 +02004929 *
4930 * Caller is responsible for holding css reference on the memcg.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004931 */
Michal Hockoab5196c2012-10-26 13:37:32 +02004932static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08004933{
Michal Hockoc26251f2012-10-26 13:37:28 +02004934 int node, zid;
Glauber Costabea207c2012-12-18 14:22:11 -08004935 u64 usage;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08004936
Daisuke Nishimurafce66472010-01-15 17:01:30 -08004937 do {
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004938 /* This is for making all *used* pages to be on LRU. */
4939 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004940 drain_all_stock_sync(memcg);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004941 mem_cgroup_start_move(memcg);
Lai Jiangshan31aaea42012-12-12 13:51:27 -08004942 for_each_node_state(node, N_MEMORY) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004943 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004944 enum lru_list lru;
4945 for_each_lru(lru) {
Michal Hocko2ef37d32012-10-26 13:37:30 +02004946 mem_cgroup_force_empty_list(memcg,
Hugh Dickinsf156ab92012-03-21 16:34:19 -07004947 node, zid, lru);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004948 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004949 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004950 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004951 mem_cgroup_end_move(memcg);
4952 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07004953 cond_resched();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08004954
Michal Hocko2ef37d32012-10-26 13:37:30 +02004955 /*
Glauber Costabea207c2012-12-18 14:22:11 -08004956 * Kernel memory may not necessarily be trackable to a specific
4957 * process. So they are not migrated, and therefore we can't
4958 * expect their value to drop to 0 here.
4959 * Having res filled up with kmem only is enough.
4960 *
Michal Hocko2ef37d32012-10-26 13:37:30 +02004961 * This is a safety check because mem_cgroup_force_empty_list
4962 * could have raced with mem_cgroup_replace_page_cache callers
4963 * so the lru seemed empty but the page could have been added
4964 * right after the check. RES_USAGE should be safe as we always
4965 * charge before adding to the LRU.
4966 */
Glauber Costabea207c2012-12-18 14:22:11 -08004967 usage = res_counter_read_u64(&memcg->res, RES_USAGE) -
4968 res_counter_read_u64(&memcg->kmem, RES_USAGE);
4969 } while (usage > 0);
Michal Hockoc26251f2012-10-26 13:37:28 +02004970}
4971
Glauber Costab5f99b52013-02-22 16:34:53 -08004972static inline bool memcg_has_children(struct mem_cgroup *memcg)
4973{
Johannes Weiner696ac172013-10-31 16:34:15 -07004974 lockdep_assert_held(&memcg_create_mutex);
4975 /*
4976 * The lock does not prevent addition or deletion to the list
4977 * of children, but it prevents a new child from being
4978 * initialized based on this parent in css_online(), so it's
4979 * enough to decide whether hierarchically inherited
4980 * attributes can still be changed or not.
4981 */
4982 return memcg->use_hierarchy &&
4983 !list_empty(&memcg->css.cgroup->children);
Glauber Costab5f99b52013-02-22 16:34:53 -08004984}
4985
4986/*
Michal Hockoc26251f2012-10-26 13:37:28 +02004987 * Reclaims as many pages from the given memcg as possible and moves
4988 * the rest to the parent.
4989 *
4990 * Caller is responsible for holding css reference for memcg.
4991 */
4992static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
4993{
4994 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
4995 struct cgroup *cgrp = memcg->css.cgroup;
4996
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004997 /* returns EBUSY if there is a task or if we come here twice. */
Michal Hockoc26251f2012-10-26 13:37:28 +02004998 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
4999 return -EBUSY;
5000
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005001 /* we call try-to-free pages for make this cgroup empty */
5002 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005003 /* try to free all pages in this cgroup */
Glauber Costa569530f2012-04-12 12:49:13 -07005004 while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005005 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005006
Michal Hockoc26251f2012-10-26 13:37:28 +02005007 if (signal_pending(current))
5008 return -EINTR;
5009
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005010 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07005011 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005012 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005013 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005014 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02005015 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005016 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08005017
5018 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08005019 lru_add_drain();
Michal Hockoab5196c2012-10-26 13:37:32 +02005020 mem_cgroup_reparent_charges(memcg);
5021
5022 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08005023}
5024
Tejun Heo182446d2013-08-08 20:11:24 -04005025static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css,
5026 unsigned int event)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005027{
Tejun Heo182446d2013-08-08 20:11:24 -04005028 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Michal Hockoc26251f2012-10-26 13:37:28 +02005029
Michal Hockod8423012012-10-26 13:37:29 +02005030 if (mem_cgroup_is_root(memcg))
5031 return -EINVAL;
Li Zefanc33bd832013-09-12 15:13:19 -07005032 return mem_cgroup_force_empty(memcg);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005033}
5034
Tejun Heo182446d2013-08-08 20:11:24 -04005035static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
5036 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005037{
Tejun Heo182446d2013-08-08 20:11:24 -04005038 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005039}
5040
Tejun Heo182446d2013-08-08 20:11:24 -04005041static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
5042 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08005043{
5044 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04005045 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005046 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css));
Balbir Singh18f59ea2009-01-07 18:08:07 -08005047
Glauber Costa09998212013-02-22 16:34:55 -08005048 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07005049
5050 if (memcg->use_hierarchy == val)
5051 goto out;
5052
Balbir Singh18f59ea2009-01-07 18:08:07 -08005053 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02005054 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08005055 * in the child subtrees. If it is unset, then the change can
5056 * occur, provided the current cgroup has no children.
5057 *
5058 * For the root cgroup, parent_mem is NULL, we allow value to be
5059 * set if there are no children.
5060 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005061 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08005062 (val == 1 || val == 0)) {
Johannes Weiner696ac172013-10-31 16:34:15 -07005063 if (list_empty(&memcg->css.cgroup->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005064 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005065 else
5066 retval = -EBUSY;
5067 } else
5068 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07005069
5070out:
Glauber Costa09998212013-02-22 16:34:55 -08005071 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005072
5073 return retval;
5074}
5075
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005076
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005077static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005078 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005079{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005080 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005081 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005082
Johannes Weiner7a159cc2011-03-23 16:42:38 -07005083 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005084 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005085 val += mem_cgroup_read_stat(iter, idx);
5086
5087 if (val < 0) /* race ? */
5088 val = 0;
5089 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07005090}
5091
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005092static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005093{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005094 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005095
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005096 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005097 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00005098 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005099 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00005100 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005101 }
5102
David Rientjesb070e652013-05-07 16:18:09 -07005103 /*
5104 * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS
5105 * as well as in MEM_CGROUP_STAT_RSS_HUGE.
5106 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005107 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
5108 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005109
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005110 if (swap)
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005111 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005112
5113 return val << PAGE_SHIFT;
5114}
5115
Tejun Heo182446d2013-08-08 20:11:24 -04005116static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css,
5117 struct cftype *cft, struct file *file,
5118 char __user *buf, size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005119{
Tejun Heo182446d2013-08-08 20:11:24 -04005120 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heoaf36f902012-04-01 12:09:55 -07005121 char str[64];
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005122 u64 val;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005123 int name, len;
5124 enum res_type type;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005125
5126 type = MEMFILE_TYPE(cft->private);
5127 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005128
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005129 switch (type) {
5130 case _MEM:
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, false);
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->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005135 break;
5136 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005137 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005138 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08005139 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005140 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005141 break;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005142 case _KMEM:
5143 val = res_counter_read_u64(&memcg->kmem, name);
5144 break;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005145 default:
5146 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005147 }
Tejun Heoaf36f902012-04-01 12:09:55 -07005148
5149 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
5150 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005151}
Glauber Costa510fc4e2012-12-18 14:21:47 -08005152
Tejun Heo182446d2013-08-08 20:11:24 -04005153static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005154{
5155 int ret = -EINVAL;
5156#ifdef CONFIG_MEMCG_KMEM
Tejun Heo182446d2013-08-08 20:11:24 -04005157 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005158 /*
5159 * For simplicity, we won't allow this to be disabled. It also can't
5160 * be changed if the cgroup has children already, or if tasks had
5161 * already joined.
5162 *
5163 * If tasks join before we set the limit, a person looking at
5164 * kmem.usage_in_bytes will have no way to determine when it took
5165 * place, which makes the value quite meaningless.
5166 *
5167 * After it first became limited, changes in the value of the limit are
5168 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08005169 */
Glauber Costa09998212013-02-22 16:34:55 -08005170 mutex_lock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005171 mutex_lock(&set_limit_mutex);
Sha Zhengju6de5a8b2013-09-12 15:13:47 -07005172 if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) {
Tejun Heo182446d2013-08-08 20:11:24 -04005173 if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005174 ret = -EBUSY;
5175 goto out;
5176 }
5177 ret = res_counter_set_limit(&memcg->kmem, val);
5178 VM_BUG_ON(ret);
5179
Glauber Costa55007d82012-12-18 14:22:38 -08005180 ret = memcg_update_cache_sizes(memcg);
5181 if (ret) {
Sha Zhengju6de5a8b2013-09-12 15:13:47 -07005182 res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX);
Glauber Costa55007d82012-12-18 14:22:38 -08005183 goto out;
5184 }
Glauber Costa692e89a2013-02-22 16:34:56 -08005185 static_key_slow_inc(&memcg_kmem_enabled_key);
5186 /*
5187 * setting the active bit after the inc will guarantee no one
5188 * starts accounting before all call sites are patched
5189 */
5190 memcg_kmem_set_active(memcg);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005191 } else
5192 ret = res_counter_set_limit(&memcg->kmem, val);
5193out:
5194 mutex_unlock(&set_limit_mutex);
Glauber Costa09998212013-02-22 16:34:55 -08005195 mutex_unlock(&memcg_create_mutex);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005196#endif
5197 return ret;
5198}
5199
Hugh Dickins6d0439902013-02-22 16:35:50 -08005200#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -08005201static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08005202{
Glauber Costa55007d82012-12-18 14:22:38 -08005203 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005204 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5205 if (!parent)
Glauber Costa55007d82012-12-18 14:22:38 -08005206 goto out;
5207
Glauber Costa510fc4e2012-12-18 14:21:47 -08005208 memcg->kmem_account_flags = parent->kmem_account_flags;
Glauber Costaa8964b92012-12-18 14:22:09 -08005209 /*
5210 * When that happen, we need to disable the static branch only on those
5211 * memcgs that enabled it. To achieve this, we would be forced to
5212 * complicate the code by keeping track of which memcgs were the ones
5213 * that actually enabled limits, and which ones got it from its
5214 * parents.
5215 *
5216 * It is a lot simpler just to do static_key_slow_inc() on every child
5217 * that is accounted.
5218 */
Glauber Costa55007d82012-12-18 14:22:38 -08005219 if (!memcg_kmem_is_active(memcg))
5220 goto out;
5221
5222 /*
Li Zefan10d5ebf2013-07-08 16:00:33 -07005223 * __mem_cgroup_free() will issue static_key_slow_dec() because this
5224 * memcg is active already. If the later initialization fails then the
5225 * cgroup core triggers the cleanup so we do not have to do it here.
Glauber Costa55007d82012-12-18 14:22:38 -08005226 */
Glauber Costa55007d82012-12-18 14:22:38 -08005227 static_key_slow_inc(&memcg_kmem_enabled_key);
5228
5229 mutex_lock(&set_limit_mutex);
Glauber Costa425c5982013-07-08 16:00:01 -07005230 memcg_stop_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005231 ret = memcg_update_cache_sizes(memcg);
Glauber Costa425c5982013-07-08 16:00:01 -07005232 memcg_resume_kmem_account();
Glauber Costa55007d82012-12-18 14:22:38 -08005233 mutex_unlock(&set_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08005234out:
5235 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08005236}
Hugh Dickins6d0439902013-02-22 16:35:50 -08005237#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08005238
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005239/*
5240 * The user of this function is...
5241 * RES_LIMIT.
5242 */
Tejun Heo182446d2013-08-08 20:11:24 -04005243static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Paul Menage856c13a2008-07-25 01:47:04 -07005244 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005245{
Tejun Heo182446d2013-08-08 20:11:24 -04005246 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005247 enum res_type type;
5248 int name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005249 unsigned long long val;
5250 int ret;
5251
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005252 type = MEMFILE_TYPE(cft->private);
5253 name = MEMFILE_ATTR(cft->private);
Tejun Heoaf36f902012-04-01 12:09:55 -07005254
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005255 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005256 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07005257 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
5258 ret = -EINVAL;
5259 break;
5260 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005261 /* This function does all necessary parse...reuse it */
5262 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005263 if (ret)
5264 break;
5265 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005266 ret = mem_cgroup_resize_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005267 else if (type == _MEMSWAP)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005268 ret = mem_cgroup_resize_memsw_limit(memcg, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005269 else if (type == _KMEM)
Tejun Heo182446d2013-08-08 20:11:24 -04005270 ret = memcg_update_kmem_limit(css, val);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005271 else
5272 return -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005273 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07005274 case RES_SOFT_LIMIT:
5275 ret = res_counter_memparse_write_strategy(buffer, &val);
5276 if (ret)
5277 break;
5278 /*
5279 * For memsw, soft limits are hard to implement in terms
5280 * of semantics, for now, we support soft limits for
5281 * control without swap
5282 */
5283 if (type == _MEM)
5284 ret = res_counter_set_soft_limit(&memcg->res, val);
5285 else
5286 ret = -EINVAL;
5287 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07005288 default:
5289 ret = -EINVAL; /* should be BUG() ? */
5290 break;
5291 }
5292 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005293}
5294
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005295static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
5296 unsigned long long *mem_limit, unsigned long long *memsw_limit)
5297{
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005298 unsigned long long min_limit, min_memsw_limit, tmp;
5299
5300 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
5301 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005302 if (!memcg->use_hierarchy)
5303 goto out;
5304
Tejun Heo63876982013-08-08 20:11:23 -04005305 while (css_parent(&memcg->css)) {
5306 memcg = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005307 if (!memcg->use_hierarchy)
5308 break;
5309 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
5310 min_limit = min(min_limit, tmp);
5311 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
5312 min_memsw_limit = min(min_memsw_limit, tmp);
5313 }
5314out:
5315 *mem_limit = min_limit;
5316 *memsw_limit = min_memsw_limit;
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005317}
5318
Tejun Heo182446d2013-08-08 20:11:24 -04005319static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005320{
Tejun Heo182446d2013-08-08 20:11:24 -04005321 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa86ae53e2012-12-18 14:21:45 -08005322 int name;
5323 enum res_type type;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005324
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005325 type = MEMFILE_TYPE(event);
5326 name = MEMFILE_ATTR(event);
Tejun Heoaf36f902012-04-01 12:09:55 -07005327
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005328 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005329 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005330 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005331 res_counter_reset_max(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005332 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005333 res_counter_reset_max(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005334 else if (type == _KMEM)
5335 res_counter_reset_max(&memcg->kmem);
5336 else
5337 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005338 break;
5339 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005340 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005341 res_counter_reset_failcnt(&memcg->res);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005342 else if (type == _MEMSWAP)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005343 res_counter_reset_failcnt(&memcg->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08005344 else if (type == _KMEM)
5345 res_counter_reset_failcnt(&memcg->kmem);
5346 else
5347 return -EINVAL;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005348 break;
5349 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07005350
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07005351 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005352}
5353
Tejun Heo182446d2013-08-08 20:11:24 -04005354static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005355 struct cftype *cft)
5356{
Tejun Heo182446d2013-08-08 20:11:24 -04005357 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005358}
5359
Daisuke Nishimura02491442010-03-10 15:22:17 -08005360#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04005361static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005362 struct cftype *cft, u64 val)
5363{
Tejun Heo182446d2013-08-08 20:11:24 -04005364 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005365
5366 if (val >= (1 << NR_MOVE_TYPE))
5367 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005368
Glauber Costaee5e8472013-02-22 16:34:50 -08005369 /*
5370 * No kind of locking is needed in here, because ->can_attach() will
5371 * check this value once in the beginning of the process, and then carry
5372 * on with stale data. This means that changes to this value will only
5373 * affect task migrations starting after the change.
5374 */
5375 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005376 return 0;
5377}
Daisuke Nishimura02491442010-03-10 15:22:17 -08005378#else
Tejun Heo182446d2013-08-08 20:11:24 -04005379static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005380 struct cftype *cft, u64 val)
5381{
5382 return -ENOSYS;
5383}
5384#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005385
Ying Han406eb0c2011-05-26 16:25:37 -07005386#ifdef CONFIG_NUMA
Tejun Heo182446d2013-08-08 20:11:24 -04005387static int memcg_numa_stat_show(struct cgroup_subsys_state *css,
5388 struct cftype *cft, struct seq_file *m)
Ying Han406eb0c2011-05-26 16:25:37 -07005389{
Greg Thelen25485de2013-11-12 15:07:40 -08005390 struct numa_stat {
5391 const char *name;
5392 unsigned int lru_mask;
5393 };
5394
5395 static const struct numa_stat stats[] = {
5396 { "total", LRU_ALL },
5397 { "file", LRU_ALL_FILE },
5398 { "anon", LRU_ALL_ANON },
5399 { "unevictable", BIT(LRU_UNEVICTABLE) },
5400 };
5401 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07005402 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08005403 unsigned long nr;
Tejun Heo182446d2013-08-08 20:11:24 -04005404 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Ying Han406eb0c2011-05-26 16:25:37 -07005405
Greg Thelen25485de2013-11-12 15:07:40 -08005406 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5407 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
5408 seq_printf(m, "%s=%lu", stat->name, nr);
5409 for_each_node_state(nid, N_MEMORY) {
5410 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
5411 stat->lru_mask);
5412 seq_printf(m, " N%d=%lu", nid, nr);
5413 }
5414 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005415 }
Ying Han406eb0c2011-05-26 16:25:37 -07005416
Ying Han071aee12013-11-12 15:07:41 -08005417 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
5418 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07005419
Ying Han071aee12013-11-12 15:07:41 -08005420 nr = 0;
5421 for_each_mem_cgroup_tree(iter, memcg)
5422 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
5423 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
5424 for_each_node_state(nid, N_MEMORY) {
5425 nr = 0;
5426 for_each_mem_cgroup_tree(iter, memcg)
5427 nr += mem_cgroup_node_nr_lru_pages(
5428 iter, nid, stat->lru_mask);
5429 seq_printf(m, " N%d=%lu", nid, nr);
5430 }
5431 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07005432 }
Ying Han406eb0c2011-05-26 16:25:37 -07005433
Ying Han406eb0c2011-05-26 16:25:37 -07005434 return 0;
5435}
5436#endif /* CONFIG_NUMA */
5437
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005438static inline void mem_cgroup_lru_names_not_uptodate(void)
5439{
5440 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
5441}
5442
Tejun Heo182446d2013-08-08 20:11:24 -04005443static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft,
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005444 struct seq_file *m)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005445{
Tejun Heo182446d2013-08-08 20:11:24 -04005446 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005447 struct mem_cgroup *mi;
5448 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005449
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005450 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005451 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005452 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005453 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
5454 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005455 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005456
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005457 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
5458 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
5459 mem_cgroup_read_events(memcg, i));
5460
5461 for (i = 0; i < NR_LRU_LISTS; i++)
5462 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
5463 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
5464
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005465 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005466 {
5467 unsigned long long limit, memsw_limit;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005468 memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005469 seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005470 if (do_swap_account)
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005471 seq_printf(m, "hierarchical_memsw_limit %llu\n",
5472 memsw_limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08005473 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005474
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005475 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
5476 long long val = 0;
5477
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07005478 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005479 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07005480 for_each_mem_cgroup_tree(mi, memcg)
5481 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
5482 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
5483 }
5484
5485 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
5486 unsigned long long val = 0;
5487
5488 for_each_mem_cgroup_tree(mi, memcg)
5489 val += mem_cgroup_read_events(mi, i);
5490 seq_printf(m, "total_%s %llu\n",
5491 mem_cgroup_events_names[i], val);
5492 }
5493
5494 for (i = 0; i < NR_LRU_LISTS; i++) {
5495 unsigned long long val = 0;
5496
5497 for_each_mem_cgroup_tree(mi, memcg)
5498 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
5499 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07005500 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07005501
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005502#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005503 {
5504 int nid, zid;
5505 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07005506 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005507 unsigned long recent_rotated[2] = {0, 0};
5508 unsigned long recent_scanned[2] = {0, 0};
5509
5510 for_each_online_node(nid)
5511 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005512 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
Hugh Dickins89abfab2012-05-29 15:06:53 -07005513 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005514
Hugh Dickins89abfab2012-05-29 15:06:53 -07005515 recent_rotated[0] += rstat->recent_rotated[0];
5516 recent_rotated[1] += rstat->recent_rotated[1];
5517 recent_scanned[0] += rstat->recent_scanned[0];
5518 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005519 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07005520 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
5521 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
5522 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
5523 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08005524 }
5525#endif
5526
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005527 return 0;
5528}
5529
Tejun Heo182446d2013-08-08 20:11:24 -04005530static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
5531 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005532{
Tejun Heo182446d2013-08-08 20:11:24 -04005533 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005534
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07005535 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005536}
5537
Tejun Heo182446d2013-08-08 20:11:24 -04005538static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
5539 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005540{
Tejun Heo182446d2013-08-08 20:11:24 -04005541 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005542 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
Li Zefan068b38c2009-01-15 13:51:26 -08005543
Tejun Heo63876982013-08-08 20:11:23 -04005544 if (val > 100 || !parent)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005545 return -EINVAL;
5546
Glauber Costa09998212013-02-22 16:34:55 -08005547 mutex_lock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005548
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005549 /* If under hierarchy, only empty-root can set this value */
Glauber Costab5f99b52013-02-22 16:34:53 -08005550 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005551 mutex_unlock(&memcg_create_mutex);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005552 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08005553 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005554
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005555 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005556
Glauber Costa09998212013-02-22 16:34:55 -08005557 mutex_unlock(&memcg_create_mutex);
Li Zefan068b38c2009-01-15 13:51:26 -08005558
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005559 return 0;
5560}
5561
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005562static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
5563{
5564 struct mem_cgroup_threshold_ary *t;
5565 u64 usage;
5566 int i;
5567
5568 rcu_read_lock();
5569 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005570 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005571 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005572 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005573
5574 if (!t)
5575 goto unlock;
5576
5577 usage = mem_cgroup_usage(memcg, swap);
5578
5579 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07005580 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005581 * If it's not true, a threshold was crossed after last
5582 * call of __mem_cgroup_threshold().
5583 */
Phil Carmody5407a562010-05-26 14:42:42 -07005584 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005585
5586 /*
5587 * Iterate backward over array of thresholds starting from
5588 * current_threshold and check if a threshold is crossed.
5589 * If none of thresholds below usage is crossed, we read
5590 * only one element of the array here.
5591 */
5592 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
5593 eventfd_signal(t->entries[i].eventfd, 1);
5594
5595 /* i = current_threshold + 1 */
5596 i++;
5597
5598 /*
5599 * Iterate forward over array of thresholds starting from
5600 * current_threshold+1 and check if a threshold is crossed.
5601 * If none of thresholds above usage is crossed, we read
5602 * only one element of the array here.
5603 */
5604 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
5605 eventfd_signal(t->entries[i].eventfd, 1);
5606
5607 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07005608 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005609unlock:
5610 rcu_read_unlock();
5611}
5612
5613static void mem_cgroup_threshold(struct mem_cgroup *memcg)
5614{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07005615 while (memcg) {
5616 __mem_cgroup_threshold(memcg, false);
5617 if (do_swap_account)
5618 __mem_cgroup_threshold(memcg, true);
5619
5620 memcg = parent_mem_cgroup(memcg);
5621 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005622}
5623
5624static int compare_thresholds(const void *a, const void *b)
5625{
5626 const struct mem_cgroup_threshold *_a = a;
5627 const struct mem_cgroup_threshold *_b = b;
5628
Greg Thelen2bff24a2013-09-11 14:23:08 -07005629 if (_a->threshold > _b->threshold)
5630 return 1;
5631
5632 if (_a->threshold < _b->threshold)
5633 return -1;
5634
5635 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005636}
5637
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005638static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005639{
5640 struct mem_cgroup_eventfd_list *ev;
5641
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005642 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005643 eventfd_signal(ev->eventfd, 1);
5644 return 0;
5645}
5646
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005647static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005648{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005649 struct mem_cgroup *iter;
5650
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005651 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07005652 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005653}
5654
Tejun Heo81eeaf02013-08-08 20:11:26 -04005655static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005656 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005657{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005658 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005659 struct mem_cgroup_thresholds *thresholds;
5660 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005661 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005662 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005663 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005664
5665 ret = res_counter_memparse_write_strategy(args, &threshold);
5666 if (ret)
5667 return ret;
5668
5669 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005670
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005671 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005672 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005673 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005674 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005675 else
5676 BUG();
5677
5678 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5679
5680 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005681 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005682 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5683
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005684 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005685
5686 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005687 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005688 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005689 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005690 ret = -ENOMEM;
5691 goto unlock;
5692 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005693 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005694
5695 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005696 if (thresholds->primary) {
5697 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005698 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005699 }
5700
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005701 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005702 new->entries[size - 1].eventfd = eventfd;
5703 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005704
5705 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005706 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005707 compare_thresholds, NULL);
5708
5709 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005710 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005711 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07005712 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005713 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005714 * new->current_threshold will not be used until
5715 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005716 * it here.
5717 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005718 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07005719 } else
5720 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005721 }
5722
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005723 /* Free old spare buffer and save old primary buffer as spare */
5724 kfree(thresholds->spare);
5725 thresholds->spare = thresholds->primary;
5726
5727 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005728
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005729 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005730 synchronize_rcu();
5731
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005732unlock:
5733 mutex_unlock(&memcg->thresholds_lock);
5734
5735 return ret;
5736}
5737
Tejun Heo81eeaf02013-08-08 20:11:26 -04005738static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005739 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005740{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005741 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005742 struct mem_cgroup_thresholds *thresholds;
5743 struct mem_cgroup_threshold_ary *new;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005744 enum res_type type = MEMFILE_TYPE(cft->private);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005745 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005746 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005747
5748 mutex_lock(&memcg->thresholds_lock);
5749 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005750 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005751 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005752 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005753 else
5754 BUG();
5755
Anton Vorontsov371528c2012-02-24 05:14:46 +04005756 if (!thresholds->primary)
5757 goto unlock;
5758
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005759 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
5760
5761 /* Check if a threshold crossed before removing */
5762 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
5763
5764 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005765 size = 0;
5766 for (i = 0; i < thresholds->primary->size; i++) {
5767 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005768 size++;
5769 }
5770
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005771 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005772
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005773 /* Set thresholds array to NULL if we don't have thresholds */
5774 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005775 kfree(new);
5776 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005777 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005778 }
5779
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005780 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005781
5782 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005783 new->current_threshold = -1;
5784 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
5785 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005786 continue;
5787
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005788 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07005789 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005790 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005791 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005792 * until rcu_assign_pointer(), so it's safe to increment
5793 * it here.
5794 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005795 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005796 }
5797 j++;
5798 }
5799
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005800swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005801 /* Swap primary and spare array */
5802 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07005803 /* If all events are unregistered, free the spare array */
5804 if (!new) {
5805 kfree(thresholds->spare);
5806 thresholds->spare = NULL;
5807 }
5808
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07005809 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005810
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07005811 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005812 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04005813unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005814 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08005815}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005816
Tejun Heo81eeaf02013-08-08 20:11:26 -04005817static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005818 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
5819{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005820 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005821 struct mem_cgroup_eventfd_list *event;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005822 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005823
5824 BUG_ON(type != _OOM_TYPE);
5825 event = kmalloc(sizeof(*event), GFP_KERNEL);
5826 if (!event)
5827 return -ENOMEM;
5828
Michal Hocko1af8efe2011-07-26 16:08:24 -07005829 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005830
5831 event->eventfd = eventfd;
5832 list_add(&event->list, &memcg->oom_notify);
5833
5834 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07005835 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005836 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07005837 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005838
5839 return 0;
5840}
5841
Tejun Heo81eeaf02013-08-08 20:11:26 -04005842static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005843 struct cftype *cft, struct eventfd_ctx *eventfd)
5844{
Tejun Heo81eeaf02013-08-08 20:11:26 -04005845 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005846 struct mem_cgroup_eventfd_list *ev, *tmp;
Glauber Costa86ae53e2012-12-18 14:21:45 -08005847 enum res_type type = MEMFILE_TYPE(cft->private);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005848
5849 BUG_ON(type != _OOM_TYPE);
5850
Michal Hocko1af8efe2011-07-26 16:08:24 -07005851 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005852
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005853 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005854 if (ev->eventfd == eventfd) {
5855 list_del(&ev->list);
5856 kfree(ev);
5857 }
5858 }
5859
Michal Hocko1af8efe2011-07-26 16:08:24 -07005860 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005861}
5862
Tejun Heo182446d2013-08-08 20:11:24 -04005863static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005864 struct cftype *cft, struct cgroup_map_cb *cb)
5865{
Tejun Heo182446d2013-08-08 20:11:24 -04005866 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005867
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005868 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005869
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005870 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005871 cb->fill(cb, "under_oom", 1);
5872 else
5873 cb->fill(cb, "under_oom", 0);
5874 return 0;
5875}
5876
Tejun Heo182446d2013-08-08 20:11:24 -04005877static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005878 struct cftype *cft, u64 val)
5879{
Tejun Heo182446d2013-08-08 20:11:24 -04005880 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo63876982013-08-08 20:11:23 -04005881 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005882
5883 /* cannot set to root cgroup and only 0 and 1 are allowed */
Tejun Heo63876982013-08-08 20:11:23 -04005884 if (!parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005885 return -EINVAL;
5886
Glauber Costa09998212013-02-22 16:34:55 -08005887 mutex_lock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005888 /* oom-kill-disable is a flag for subhierarchy. */
Glauber Costab5f99b52013-02-22 16:34:53 -08005889 if ((parent->use_hierarchy) || memcg_has_children(memcg)) {
Glauber Costa09998212013-02-22 16:34:55 -08005890 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005891 return -EINVAL;
5892 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005893 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07005894 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005895 memcg_oom_recover(memcg);
Glauber Costa09998212013-02-22 16:34:55 -08005896 mutex_unlock(&memcg_create_mutex);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005897 return 0;
5898}
5899
Andrew Mortonc255a452012-07-31 16:43:02 -07005900#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03005901static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005902{
Glauber Costa55007d82012-12-18 14:22:38 -08005903 int ret;
5904
Glauber Costa2633d7a2012-12-18 14:22:34 -08005905 memcg->kmemcg_id = -1;
Glauber Costa55007d82012-12-18 14:22:38 -08005906 ret = memcg_propagate_kmem(memcg);
5907 if (ret)
5908 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08005909
Glauber Costa1d62e432012-04-09 19:36:33 -03005910 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07005911}
Glauber Costae5671df2011-12-11 21:47:01 +00005912
Li Zefan10d5ebf2013-07-08 16:00:33 -07005913static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005914{
Glauber Costa1d62e432012-04-09 19:36:33 -03005915 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07005916}
5917
5918static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
5919{
5920 if (!memcg_kmem_is_active(memcg))
5921 return;
5922
5923 /*
5924 * kmem charges can outlive the cgroup. In the case of slab
5925 * pages, for instance, a page contain objects from various
5926 * processes. As we prevent from taking a reference for every
5927 * such allocation we have to be careful when doing uncharge
5928 * (see memcg_uncharge_kmem) and here during offlining.
5929 *
5930 * The idea is that that only the _last_ uncharge which sees
5931 * the dead memcg will drop the last reference. An additional
5932 * reference is taken here before the group is marked dead
5933 * which is then paired with css_put during uncharge resp. here.
5934 *
5935 * Although this might sound strange as this path is called from
5936 * css_offline() when the referencemight have dropped down to 0
5937 * and shouldn't be incremented anymore (css_tryget would fail)
5938 * we do not have other options because of the kmem allocations
5939 * lifetime.
5940 */
5941 css_get(&memcg->css);
Glauber Costa7de37682012-12-18 14:22:07 -08005942
5943 memcg_kmem_mark_dead(memcg);
5944
5945 if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0)
5946 return;
5947
Glauber Costa7de37682012-12-18 14:22:07 -08005948 if (memcg_kmem_test_and_clear_dead(memcg))
Li Zefan10d5ebf2013-07-08 16:00:33 -07005949 css_put(&memcg->css);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005950}
Glauber Costae5671df2011-12-11 21:47:01 +00005951#else
Glauber Costacbe128e32012-04-09 19:36:34 -03005952static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00005953{
5954 return 0;
5955}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005956
Li Zefan10d5ebf2013-07-08 16:00:33 -07005957static void memcg_destroy_kmem(struct mem_cgroup *memcg)
5958{
5959}
5960
5961static void kmem_cgroup_css_offline(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005962{
5963}
Glauber Costae5671df2011-12-11 21:47:01 +00005964#endif
5965
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005966static struct cftype mem_cgroup_files[] = {
5967 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005968 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005969 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heoaf36f902012-04-01 12:09:55 -07005970 .read = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005971 .register_event = mem_cgroup_usage_register_event,
5972 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005973 },
5974 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005975 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005976 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005977 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005978 .read = mem_cgroup_read,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005979 },
5980 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005981 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005982 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07005983 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005984 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005985 },
5986 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005987 .name = "soft_limit_in_bytes",
5988 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
5989 .write_string = mem_cgroup_write,
Tejun Heoaf36f902012-04-01 12:09:55 -07005990 .read = mem_cgroup_read,
Balbir Singh296c81d2009-09-23 15:56:36 -07005991 },
5992 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005993 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005994 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07005995 .trigger = mem_cgroup_reset,
Tejun Heoaf36f902012-04-01 12:09:55 -07005996 .read = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005997 },
Balbir Singh8697d332008-02-07 00:13:59 -08005998 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005999 .name = "stat",
Wanpeng Liab215882012-07-31 16:43:09 -07006000 .read_seq_string = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08006001 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08006002 {
6003 .name = "force_empty",
6004 .trigger = mem_cgroup_force_empty_write,
6005 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08006006 {
6007 .name = "use_hierarchy",
Tejun Heof00baae2013-04-15 13:41:15 -07006008 .flags = CFTYPE_INSANE,
Balbir Singh18f59ea2009-01-07 18:08:07 -08006009 .write_u64 = mem_cgroup_hierarchy_write,
6010 .read_u64 = mem_cgroup_hierarchy_read,
6011 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08006012 {
6013 .name = "swappiness",
6014 .read_u64 = mem_cgroup_swappiness_read,
6015 .write_u64 = mem_cgroup_swappiness_write,
6016 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006017 {
6018 .name = "move_charge_at_immigrate",
6019 .read_u64 = mem_cgroup_move_charge_read,
6020 .write_u64 = mem_cgroup_move_charge_write,
6021 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006022 {
6023 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07006024 .read_map = mem_cgroup_oom_control_read,
6025 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07006026 .register_event = mem_cgroup_oom_register_event,
6027 .unregister_event = mem_cgroup_oom_unregister_event,
6028 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
6029 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006030 {
6031 .name = "pressure_level",
6032 .register_event = vmpressure_register_event,
6033 .unregister_event = vmpressure_unregister_event,
6034 },
Ying Han406eb0c2011-05-26 16:25:37 -07006035#ifdef CONFIG_NUMA
6036 {
6037 .name = "numa_stat",
Wanpeng Liab215882012-07-31 16:43:09 -07006038 .read_seq_string = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07006039 },
6040#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006041#ifdef CONFIG_MEMCG_KMEM
6042 {
6043 .name = "kmem.limit_in_bytes",
6044 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
6045 .write_string = mem_cgroup_write,
6046 .read = mem_cgroup_read,
6047 },
6048 {
6049 .name = "kmem.usage_in_bytes",
6050 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
6051 .read = mem_cgroup_read,
6052 },
6053 {
6054 .name = "kmem.failcnt",
6055 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
6056 .trigger = mem_cgroup_reset,
6057 .read = mem_cgroup_read,
6058 },
6059 {
6060 .name = "kmem.max_usage_in_bytes",
6061 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
6062 .trigger = mem_cgroup_reset,
6063 .read = mem_cgroup_read,
6064 },
Glauber Costa749c5412012-12-18 14:23:01 -08006065#ifdef CONFIG_SLABINFO
6066 {
6067 .name = "kmem.slabinfo",
6068 .read_seq_string = mem_cgroup_slabinfo_read,
6069 },
6070#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08006071#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07006072 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07006073};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006074
Michal Hocko2d110852013-02-22 16:34:43 -08006075#ifdef CONFIG_MEMCG_SWAP
6076static struct cftype memsw_cgroup_files[] = {
6077 {
6078 .name = "memsw.usage_in_bytes",
6079 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6080 .read = mem_cgroup_read,
6081 .register_event = mem_cgroup_usage_register_event,
6082 .unregister_event = mem_cgroup_usage_unregister_event,
6083 },
6084 {
6085 .name = "memsw.max_usage_in_bytes",
6086 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6087 .trigger = mem_cgroup_reset,
6088 .read = mem_cgroup_read,
6089 },
6090 {
6091 .name = "memsw.limit_in_bytes",
6092 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6093 .write_string = mem_cgroup_write,
6094 .read = mem_cgroup_read,
6095 },
6096 {
6097 .name = "memsw.failcnt",
6098 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6099 .trigger = mem_cgroup_reset,
6100 .read = mem_cgroup_read,
6101 },
6102 { }, /* terminate */
6103};
6104#endif
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006105static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006106{
6107 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006108 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006109 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006110 /*
6111 * This routine is called against possible nodes.
6112 * But it's BUG to call kmalloc() against offline node.
6113 *
6114 * TODO: this routine can waste much memory for nodes which will
6115 * never be onlined. It's better to use memory hotplug callback
6116 * function.
6117 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07006118 if (!node_state(node, N_NORMAL_MEMORY))
6119 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08006120 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006121 if (!pn)
6122 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006123
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006124 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6125 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08006126 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006127 mz->usage_in_excess = 0;
6128 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006129 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006130 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006131 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006132 return 0;
6133}
6134
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006135static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006136{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07006137 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08006138}
6139
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006140static struct mem_cgroup *mem_cgroup_alloc(void)
6141{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006142 struct mem_cgroup *memcg;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006143 size_t size = memcg_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006144
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006145 /* Can be very big if nr_node_ids is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08006146 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006147 memcg = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006148 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006149 memcg = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006150
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006151 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07006152 return NULL;
6153
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006154 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
6155 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006156 goto out_free;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006157 spin_lock_init(&memcg->pcp_counter_lock);
6158 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006159
6160out_free:
6161 if (size < PAGE_SIZE)
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006162 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006163 else
Hugh Dickinsd79154b2012-03-21 16:34:18 -07006164 vfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08006165 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006166}
6167
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08006168/*
Glauber Costac8b2a362012-12-18 14:22:13 -08006169 * At destroying mem_cgroup, references from swap_cgroup can remain.
6170 * (scanning all at force_empty is too costly...)
6171 *
6172 * Instead of clearing all references at force_empty, we remember
6173 * the number of reference from swap_cgroup and free mem_cgroup when
6174 * it goes down to 0.
6175 *
6176 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07006177 */
Glauber Costac8b2a362012-12-18 14:22:13 -08006178
6179static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07006180{
Glauber Costac8b2a362012-12-18 14:22:13 -08006181 int node;
Glauber Costa45cf7eb2013-02-22 16:34:49 -08006182 size_t size = memcg_size();
Hugh Dickins59927fb2012-03-15 15:17:07 -07006183
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006184 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08006185
6186 for_each_node(node)
6187 free_mem_cgroup_per_zone_info(memcg, node);
6188
6189 free_percpu(memcg->stat);
6190
Glauber Costa3f134612012-05-29 15:07:11 -07006191 /*
6192 * We need to make sure that (at least for now), the jump label
6193 * destruction code runs outside of the cgroup lock. This is because
6194 * get_online_cpus(), which is called from the static_branch update,
6195 * can't be called inside the cgroup_lock. cpusets are the ones
6196 * enforcing this dependency, so if they ever change, we might as well.
6197 *
6198 * schedule_work() will guarantee this happens. Be careful if you need
6199 * to move this code around, and make sure it is outside
6200 * the cgroup_lock.
6201 */
Glauber Costaa8964b92012-12-18 14:22:09 -08006202 disarm_static_keys(memcg);
Glauber Costa3afe36b2012-05-29 15:07:10 -07006203 if (size < PAGE_SIZE)
6204 kfree(memcg);
6205 else
6206 vfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07006207}
Glauber Costa3afe36b2012-05-29 15:07:10 -07006208
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006209/*
6210 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
6211 */
Glauber Costae1aab162011-12-11 21:47:03 +00006212struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006213{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006214 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006215 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006216 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006217}
Glauber Costae1aab162011-12-11 21:47:03 +00006218EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07006219
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07006220static void __init mem_cgroup_soft_limit_tree_init(void)
6221{
6222 struct mem_cgroup_tree_per_node *rtpn;
6223 struct mem_cgroup_tree_per_zone *rtpz;
6224 int tmp, node, zone;
6225
6226 for_each_node(node) {
6227 tmp = node;
6228 if (!node_state(node, N_NORMAL_MEMORY))
6229 tmp = -1;
6230 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
6231 BUG_ON(!rtpn);
6232
6233 soft_limit_tree.rb_tree_per_node[node] = rtpn;
6234
6235 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
6236 rtpz = &rtpn->rb_tree_per_zone[zone];
6237 rtpz->rb_root = RB_ROOT;
6238 spin_lock_init(&rtpz->lock);
6239 }
6240 }
6241}
6242
Li Zefan0eb253e2009-01-15 13:51:25 -08006243static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04006244mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006245{
Glauber Costad142e3e2013-02-22 16:34:52 -08006246 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006247 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006248 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006249
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006250 memcg = mem_cgroup_alloc();
6251 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07006252 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006253
Bob Liu3ed28fa2012-01-12 17:19:04 -08006254 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006255 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006256 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07006257
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006258 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04006259 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08006260 root_mem_cgroup = memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08006261 res_counter_init(&memcg->res, NULL);
6262 res_counter_init(&memcg->memsw, NULL);
6263 res_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08006264 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08006265
Glauber Costad142e3e2013-02-22 16:34:52 -08006266 memcg->last_scanned_node = MAX_NUMNODES;
6267 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08006268 memcg->move_charge_at_immigrate = 0;
6269 mutex_init(&memcg->thresholds_lock);
6270 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07006271 vmpressure_init(&memcg->vmpressure);
Glauber Costad142e3e2013-02-22 16:34:52 -08006272
6273 return &memcg->css;
6274
6275free_out:
6276 __mem_cgroup_free(memcg);
6277 return ERR_PTR(error);
6278}
6279
6280static int
Tejun Heoeb954192013-08-08 20:11:23 -04006281mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08006282{
Tejun Heoeb954192013-08-08 20:11:23 -04006283 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6284 struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css));
Glauber Costad142e3e2013-02-22 16:34:52 -08006285 int error = 0;
6286
Li Zefan4219b2d2013-09-23 16:56:29 +08006287 if (css->cgroup->id > MEM_CGROUP_ID_MAX)
6288 return -ENOSPC;
6289
Tejun Heo63876982013-08-08 20:11:23 -04006290 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08006291 return 0;
6292
Glauber Costa09998212013-02-22 16:34:55 -08006293 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006294
6295 memcg->use_hierarchy = parent->use_hierarchy;
6296 memcg->oom_kill_disable = parent->oom_kill_disable;
6297 memcg->swappiness = mem_cgroup_swappiness(parent);
6298
6299 if (parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006300 res_counter_init(&memcg->res, &parent->res);
6301 res_counter_init(&memcg->memsw, &parent->memsw);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006302 res_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08006303
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006304 /*
Li Zefan8d76a972013-07-08 16:00:36 -07006305 * No need to take a reference to the parent because cgroup
6306 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08006307 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08006308 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006309 res_counter_init(&memcg->res, NULL);
6310 res_counter_init(&memcg->memsw, NULL);
Glauber Costa510fc4e2012-12-18 14:21:47 -08006311 res_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006312 /*
6313 * Deeper hierachy with use_hierarchy == false doesn't make
6314 * much sense so let cgroup subsystem know about this
6315 * unfortunate state in our controller.
6316 */
Glauber Costad142e3e2013-02-22 16:34:52 -08006317 if (parent != root_mem_cgroup)
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07006318 mem_cgroup_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08006319 }
Glauber Costacbe128e32012-04-09 19:36:34 -03006320
6321 error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
Glauber Costa09998212013-02-22 16:34:55 -08006322 mutex_unlock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08006323 return error;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006324}
6325
Michal Hocko5f578162013-04-29 15:07:17 -07006326/*
6327 * Announce all parents that a group from their hierarchy is gone.
6328 */
6329static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
6330{
6331 struct mem_cgroup *parent = memcg;
6332
6333 while ((parent = parent_mem_cgroup(parent)))
Johannes Weiner519ebea2013-07-03 15:04:51 -07006334 mem_cgroup_iter_invalidate(parent);
Michal Hocko5f578162013-04-29 15:07:17 -07006335
6336 /*
6337 * if the root memcg is not hierarchical we have to check it
6338 * explicitely.
6339 */
6340 if (!root_mem_cgroup->use_hierarchy)
Johannes Weiner519ebea2013-07-03 15:04:51 -07006341 mem_cgroup_iter_invalidate(root_mem_cgroup);
Michal Hocko5f578162013-04-29 15:07:17 -07006342}
6343
Tejun Heoeb954192013-08-08 20:11:23 -04006344static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006345{
Tejun Heoeb954192013-08-08 20:11:23 -04006346 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07006347
Li Zefan10d5ebf2013-07-08 16:00:33 -07006348 kmem_cgroup_css_offline(memcg);
6349
Michal Hocko5f578162013-04-29 15:07:17 -07006350 mem_cgroup_invalidate_reclaim_iterators(memcg);
Michal Hockoab5196c2012-10-26 13:37:32 +02006351 mem_cgroup_reparent_charges(memcg);
Glauber Costa1f458cb2012-12-18 14:22:50 -08006352 mem_cgroup_destroy_all_caches(memcg);
Michal Hocko33cb8762013-07-31 13:53:51 -07006353 vmpressure_cleanup(&memcg->vmpressure);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08006354}
6355
Tejun Heoeb954192013-08-08 20:11:23 -04006356static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006357{
Tejun Heoeb954192013-08-08 20:11:23 -04006358 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner96f1c582013-12-12 17:12:34 -08006359 /*
6360 * XXX: css_offline() would be where we should reparent all
6361 * memory to prepare the cgroup for destruction. However,
6362 * memcg does not do css_tryget() and res_counter charging
6363 * under the same RCU lock region, which means that charging
6364 * could race with offlining. Offlining only happens to
6365 * cgroups with no tasks in them but charges can show up
6366 * without any tasks from the swapin path when the target
6367 * memcg is looked up from the swapout record and not from the
6368 * current task as it usually is. A race like this can leak
6369 * charges and put pages with stale cgroup pointers into
6370 * circulation:
6371 *
6372 * #0 #1
6373 * lookup_swap_cgroup_id()
6374 * rcu_read_lock()
6375 * mem_cgroup_lookup()
6376 * css_tryget()
6377 * rcu_read_unlock()
6378 * disable css_tryget()
6379 * call_rcu()
6380 * offline_css()
6381 * reparent_charges()
6382 * res_counter_charge()
6383 * css_put()
6384 * css_free()
6385 * pc->mem_cgroup = dead memcg
6386 * add page to lru
6387 *
6388 * The bulk of the charges are still moved in offline_css() to
6389 * avoid pinning a lot of pages in case a long-term reference
6390 * like a swapout record is deferring the css_free() to long
6391 * after offlining. But this makes sure we catch any charges
6392 * made after offlining:
6393 */
6394 mem_cgroup_reparent_charges(memcg);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08006395
Li Zefan10d5ebf2013-07-08 16:00:33 -07006396 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07006397 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006398}
6399
Daisuke Nishimura02491442010-03-10 15:22:17 -08006400#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006401/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006402#define PRECHARGE_COUNT_AT_ONCE 256
6403static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006404{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006405 int ret = 0;
6406 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006407 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006408
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006409 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006410 mc.precharge += count;
6411 /* we don't need css_get for root */
6412 return ret;
6413 }
6414 /* try to charge at once */
6415 if (count > 1) {
6416 struct res_counter *dummy;
6417 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006418 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006419 * by cgroup_lock_live_cgroup() that it is not removed and we
6420 * are still under the same cgroup_mutex. So we can postpone
6421 * css_get().
6422 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006423 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006424 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006425 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006426 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006427 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006428 goto one_by_one;
6429 }
6430 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006431 return ret;
6432 }
6433one_by_one:
6434 /* fall back to one by one charge */
6435 while (count--) {
6436 if (signal_pending(current)) {
6437 ret = -EINTR;
6438 break;
6439 }
6440 if (!batch_count--) {
6441 batch_count = PRECHARGE_COUNT_AT_ONCE;
6442 cond_resched();
6443 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006444 ret = __mem_cgroup_try_charge(NULL,
6445 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006446 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006447 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08006448 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006449 mc.precharge++;
6450 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006451 return ret;
6452}
6453
6454/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006455 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006456 * @vma: the vma the pte to be checked belongs
6457 * @addr: the address corresponding to the pte to be checked
6458 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08006459 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006460 *
6461 * Returns
6462 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
6463 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
6464 * move charge. if @target is not NULL, the page is stored in target->page
6465 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08006466 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
6467 * target for charge migration. if @target is not NULL, the entry is stored
6468 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006469 *
6470 * Called with pte lock held.
6471 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006472union mc_target {
6473 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006474 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006475};
6476
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006477enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006478 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006479 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08006480 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006481};
6482
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006483static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
6484 unsigned long addr, pte_t ptent)
6485{
6486 struct page *page = vm_normal_page(vma, addr, ptent);
6487
6488 if (!page || !page_mapped(page))
6489 return NULL;
6490 if (PageAnon(page)) {
6491 /* we don't move shared anon */
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006492 if (!move_anon())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006493 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006494 } else if (!move_file())
6495 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006496 return NULL;
6497 if (!get_page_unless_zero(page))
6498 return NULL;
6499
6500 return page;
6501}
6502
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006503#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006504static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6505 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6506{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006507 struct page *page = NULL;
6508 swp_entry_t ent = pte_to_swp_entry(ptent);
6509
6510 if (!move_anon() || non_swap_entry(ent))
6511 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006512 /*
6513 * Because lookup_swap_cache() updates some statistics counter,
6514 * we call find_get_page() with swapper_space directly.
6515 */
Shaohua Li33806f02013-02-22 16:34:37 -08006516 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006517 if (do_swap_account)
6518 entry->val = ent.val;
6519
6520 return page;
6521}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07006522#else
6523static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
6524 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6525{
6526 return NULL;
6527}
6528#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006529
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006530static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
6531 unsigned long addr, pte_t ptent, swp_entry_t *entry)
6532{
6533 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006534 struct address_space *mapping;
6535 pgoff_t pgoff;
6536
6537 if (!vma->vm_file) /* anonymous vma */
6538 return NULL;
6539 if (!move_file())
6540 return NULL;
6541
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006542 mapping = vma->vm_file->f_mapping;
6543 if (pte_none(ptent))
6544 pgoff = linear_page_index(vma, addr);
6545 else /* pte_file(ptent) is true */
6546 pgoff = pte_to_pgoff(ptent);
6547
6548 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006549 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006550
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006551#ifdef CONFIG_SWAP
6552 /* shmem/tmpfs may report page out on swap: account for that too. */
6553 if (radix_tree_exceptional_entry(page)) {
6554 swp_entry_t swap = radix_to_swp_entry(page);
6555 if (do_swap_account)
6556 *entry = swap;
Shaohua Li33806f02013-02-22 16:34:37 -08006557 page = find_get_page(swap_address_space(swap), swap.val);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07006558 }
6559#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006560 return page;
6561}
6562
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006563static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006564 unsigned long addr, pte_t ptent, union mc_target *target)
6565{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006566 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006567 struct page_cgroup *pc;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006568 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006569 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006570
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006571 if (pte_present(ptent))
6572 page = mc_handle_present_pte(vma, addr, ptent);
6573 else if (is_swap_pte(ptent))
6574 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07006575 else if (pte_none(ptent) || pte_file(ptent))
6576 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006577
6578 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006579 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006580 if (page) {
6581 pc = lookup_page_cgroup(page);
6582 /*
6583 * Do only loose check w/o page_cgroup lock.
6584 * mem_cgroup_move_account() checks the pc is valid or not under
6585 * the lock.
6586 */
6587 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6588 ret = MC_TARGET_PAGE;
6589 if (target)
6590 target->page = page;
6591 }
6592 if (!ret || !target)
6593 put_page(page);
6594 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006595 /* There is a swap entry and a page doesn't exist or isn't charged */
6596 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08006597 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006598 ret = MC_TARGET_SWAP;
6599 if (target)
6600 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006601 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006602 return ret;
6603}
6604
Naoya Horiguchi12724852012-03-21 16:34:28 -07006605#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6606/*
6607 * We don't consider swapping or file mapped pages because THP does not
6608 * support them for now.
6609 * Caller should make sure that pmd_trans_huge(pmd) is true.
6610 */
6611static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6612 unsigned long addr, pmd_t pmd, union mc_target *target)
6613{
6614 struct page *page = NULL;
6615 struct page_cgroup *pc;
6616 enum mc_target_type ret = MC_TARGET_NONE;
6617
6618 page = pmd_page(pmd);
6619 VM_BUG_ON(!page || !PageHead(page));
6620 if (!move_anon())
6621 return ret;
6622 pc = lookup_page_cgroup(page);
6623 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
6624 ret = MC_TARGET_PAGE;
6625 if (target) {
6626 get_page(page);
6627 target->page = page;
6628 }
6629 }
6630 return ret;
6631}
6632#else
6633static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6634 unsigned long addr, pmd_t pmd, union mc_target *target)
6635{
6636 return MC_TARGET_NONE;
6637}
6638#endif
6639
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006640static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6641 unsigned long addr, unsigned long end,
6642 struct mm_walk *walk)
6643{
6644 struct vm_area_struct *vma = walk->private;
6645 pte_t *pte;
6646 spinlock_t *ptl;
6647
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006648 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006649 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6650 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006651 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006652 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006653 }
Dave Hansen03319322011-03-22 16:32:56 -07006654
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006655 if (pmd_trans_unstable(pmd))
6656 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006657 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6658 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006659 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006660 mc.precharge++; /* increment precharge temporarily */
6661 pte_unmap_unlock(pte - 1, ptl);
6662 cond_resched();
6663
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006664 return 0;
6665}
6666
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006667static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6668{
6669 unsigned long precharge;
6670 struct vm_area_struct *vma;
6671
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006672 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006673 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6674 struct mm_walk mem_cgroup_count_precharge_walk = {
6675 .pmd_entry = mem_cgroup_count_precharge_pte_range,
6676 .mm = mm,
6677 .private = vma,
6678 };
6679 if (is_vm_hugetlb_page(vma))
6680 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006681 walk_page_range(vma->vm_start, vma->vm_end,
6682 &mem_cgroup_count_precharge_walk);
6683 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006684 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006685
6686 precharge = mc.precharge;
6687 mc.precharge = 0;
6688
6689 return precharge;
6690}
6691
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006692static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6693{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006694 unsigned long precharge = mem_cgroup_count_precharge(mm);
6695
6696 VM_BUG_ON(mc.moving_task);
6697 mc.moving_task = current;
6698 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006699}
6700
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006701/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6702static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006703{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006704 struct mem_cgroup *from = mc.from;
6705 struct mem_cgroup *to = mc.to;
Li Zefan40503772013-07-08 16:00:34 -07006706 int i;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006707
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006708 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006709 if (mc.precharge) {
6710 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
6711 mc.precharge = 0;
6712 }
6713 /*
6714 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6715 * we must uncharge here.
6716 */
6717 if (mc.moved_charge) {
6718 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
6719 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006720 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006721 /* we must fixup refcnts and charges */
6722 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006723 /* uncharge swap account from the old cgroup */
6724 if (!mem_cgroup_is_root(mc.from))
6725 res_counter_uncharge(&mc.from->memsw,
6726 PAGE_SIZE * mc.moved_swap);
Li Zefan40503772013-07-08 16:00:34 -07006727
6728 for (i = 0; i < mc.moved_swap; i++)
6729 css_put(&mc.from->css);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006730
6731 if (!mem_cgroup_is_root(mc.to)) {
6732 /*
6733 * we charged both to->res and to->memsw, so we should
6734 * uncharge to->res.
6735 */
6736 res_counter_uncharge(&mc.to->res,
6737 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006738 }
Li Zefan40503772013-07-08 16:00:34 -07006739 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006740 mc.moved_swap = 0;
6741 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006742 memcg_oom_recover(from);
6743 memcg_oom_recover(to);
6744 wake_up_all(&mc.waitq);
6745}
6746
6747static void mem_cgroup_clear_mc(void)
6748{
6749 struct mem_cgroup *from = mc.from;
6750
6751 /*
6752 * we must clear moving_task before waking up waiters at the end of
6753 * task migration.
6754 */
6755 mc.moving_task = NULL;
6756 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006757 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006758 mc.from = NULL;
6759 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006760 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006761 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006762}
6763
Tejun Heoeb954192013-08-08 20:11:23 -04006764static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006765 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006766{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006767 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006768 int ret = 0;
Tejun Heoeb954192013-08-08 20:11:23 -04006769 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costaee5e8472013-02-22 16:34:50 -08006770 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006771
Glauber Costaee5e8472013-02-22 16:34:50 -08006772 /*
6773 * We are now commited to this value whatever it is. Changes in this
6774 * tunable will only affect upcoming migrations, not the current one.
6775 * So we need to save it, and keep it going.
6776 */
6777 move_charge_at_immigrate = memcg->move_charge_at_immigrate;
6778 if (move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006779 struct mm_struct *mm;
6780 struct mem_cgroup *from = mem_cgroup_from_task(p);
6781
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006782 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006783
6784 mm = get_task_mm(p);
6785 if (!mm)
6786 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006787 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006788 if (mm->owner == p) {
6789 VM_BUG_ON(mc.from);
6790 VM_BUG_ON(mc.to);
6791 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006792 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006793 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07006794 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006795 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006796 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07006797 mc.to = memcg;
Glauber Costaee5e8472013-02-22 16:34:50 -08006798 mc.immigrate_flags = move_charge_at_immigrate;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006799 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006800 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006801
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006802 ret = mem_cgroup_precharge_mc(mm);
6803 if (ret)
6804 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006805 }
6806 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006807 }
6808 return ret;
6809}
6810
Tejun Heoeb954192013-08-08 20:11:23 -04006811static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006812 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006813{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006814 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006815}
6816
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006817static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6818 unsigned long addr, unsigned long end,
6819 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006820{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006821 int ret = 0;
6822 struct vm_area_struct *vma = walk->private;
6823 pte_t *pte;
6824 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006825 enum mc_target_type target_type;
6826 union mc_target target;
6827 struct page *page;
6828 struct page_cgroup *pc;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006829
Naoya Horiguchi12724852012-03-21 16:34:28 -07006830 /*
6831 * We don't take compound_lock() here but no race with splitting thp
6832 * happens because:
6833 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
6834 * under splitting, which means there's no concurrent thp split,
6835 * - if another thread runs into split_huge_page() just after we
6836 * entered this if-block, the thread must wait for page table lock
6837 * to be unlocked in __split_huge_page_splitting(), where the main
6838 * part of thp split is not executed yet.
6839 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006840 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006841 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006842 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006843 return 0;
6844 }
6845 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6846 if (target_type == MC_TARGET_PAGE) {
6847 page = target.page;
6848 if (!isolate_lru_page(page)) {
6849 pc = lookup_page_cgroup(page);
6850 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006851 pc, mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006852 mc.precharge -= HPAGE_PMD_NR;
6853 mc.moved_charge += HPAGE_PMD_NR;
6854 }
6855 putback_lru_page(page);
6856 }
6857 put_page(page);
6858 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006859 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006860 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006861 }
6862
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006863 if (pmd_trans_unstable(pmd))
6864 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006865retry:
6866 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6867 for (; addr != end; addr += PAGE_SIZE) {
6868 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006869 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006870
6871 if (!mc.precharge)
6872 break;
6873
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006874 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006875 case MC_TARGET_PAGE:
6876 page = target.page;
6877 if (isolate_lru_page(page))
6878 goto put;
6879 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07006880 if (!mem_cgroup_move_account(page, 1, pc,
KAMEZAWA Hiroyuki2f3479b2012-05-29 15:07:04 -07006881 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006882 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006883 /* we uncharge from mc.from later. */
6884 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006885 }
6886 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006887put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006888 put_page(page);
6889 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006890 case MC_TARGET_SWAP:
6891 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006892 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006893 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006894 /* we fixup refcnts and charges later. */
6895 mc.moved_swap++;
6896 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006897 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006898 default:
6899 break;
6900 }
6901 }
6902 pte_unmap_unlock(pte - 1, ptl);
6903 cond_resched();
6904
6905 if (addr != end) {
6906 /*
6907 * We have consumed all precharges we got in can_attach().
6908 * We try charge one by one, but don't do any additional
6909 * charges to mc.to if we have failed in charge once in attach()
6910 * phase.
6911 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006912 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006913 if (!ret)
6914 goto retry;
6915 }
6916
6917 return ret;
6918}
6919
6920static void mem_cgroup_move_charge(struct mm_struct *mm)
6921{
6922 struct vm_area_struct *vma;
6923
6924 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006925retry:
6926 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
6927 /*
6928 * Someone who are holding the mmap_sem might be waiting in
6929 * waitq. So we cancel all extra charges, wake up all waiters,
6930 * and retry. Because we cancel precharges, we might not be able
6931 * to move enough charges, but moving charge is a best-effort
6932 * feature anyway, so it wouldn't be a big problem.
6933 */
6934 __mem_cgroup_clear_mc();
6935 cond_resched();
6936 goto retry;
6937 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006938 for (vma = mm->mmap; vma; vma = vma->vm_next) {
6939 int ret;
6940 struct mm_walk mem_cgroup_move_charge_walk = {
6941 .pmd_entry = mem_cgroup_move_charge_pte_range,
6942 .mm = mm,
6943 .private = vma,
6944 };
6945 if (is_vm_hugetlb_page(vma))
6946 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006947 ret = walk_page_range(vma->vm_start, vma->vm_end,
6948 &mem_cgroup_move_charge_walk);
6949 if (ret)
6950 /*
6951 * means we have consumed all precharges and failed in
6952 * doing additional charge. Just abandon here.
6953 */
6954 break;
6955 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006956 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006957}
6958
Tejun Heoeb954192013-08-08 20:11:23 -04006959static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006960 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08006961{
Tejun Heo2f7ee562011-12-12 18:12:21 -08006962 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006963 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006964
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006965 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006966 if (mc.to)
6967 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006968 mmput(mm);
6969 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006970 if (mc.to)
6971 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08006972}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006973#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04006974static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006975 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006976{
6977 return 0;
6978}
Tejun Heoeb954192013-08-08 20:11:23 -04006979static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006980 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006981{
6982}
Tejun Heoeb954192013-08-08 20:11:23 -04006983static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08006984 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006985{
6986}
6987#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006988
Tejun Heof00baae2013-04-15 13:41:15 -07006989/*
6990 * Cgroup retains root cgroups across [un]mount cycles making it necessary
6991 * to verify sane_behavior flag on each mount attempt.
6992 */
Tejun Heoeb954192013-08-08 20:11:23 -04006993static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006994{
6995 /*
6996 * use_hierarchy is forced with sane_behavior. cgroup core
6997 * guarantees that @root doesn't have any children, so turning it
6998 * on for the root memcg is enough.
6999 */
Tejun Heoeb954192013-08-08 20:11:23 -04007000 if (cgroup_sane_behavior(root_css->cgroup))
7001 mem_cgroup_from_css(root_css)->use_hierarchy = true;
Tejun Heof00baae2013-04-15 13:41:15 -07007002}
7003
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007004struct cgroup_subsys mem_cgroup_subsys = {
7005 .name = "memory",
7006 .subsys_id = mem_cgroup_subsys_id,
Tejun Heo92fb9742012-11-19 08:13:38 -08007007 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08007008 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08007009 .css_offline = mem_cgroup_css_offline,
7010 .css_free = mem_cgroup_css_free,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007011 .can_attach = mem_cgroup_can_attach,
7012 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08007013 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07007014 .bind = mem_cgroup_bind,
Tejun Heo6bc10342012-04-01 12:09:55 -07007015 .base_cftypes = mem_cgroup_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08007016 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007017};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007018
Andrew Mortonc255a452012-07-31 16:43:02 -07007019#ifdef CONFIG_MEMCG_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08007020static int __init enable_swap_account(char *s)
7021{
Michal Hockoa2c89902011-05-24 17:12:50 -07007022 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007023 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07007024 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08007025 really_do_swap_account = 0;
7026 return 1;
7027}
Michal Hockoa2c89902011-05-24 17:12:50 -07007028__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007029
Michal Hocko2d110852013-02-22 16:34:43 -08007030static void __init memsw_file_init(void)
7031{
Michal Hocko6acc8b02013-02-22 16:34:45 -08007032 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files));
Michal Hocko2d110852013-02-22 16:34:43 -08007033}
Michal Hocko6acc8b02013-02-22 16:34:45 -08007034
7035static void __init enable_swap_cgroup(void)
7036{
7037 if (!mem_cgroup_disabled() && really_do_swap_account) {
7038 do_swap_account = 1;
7039 memsw_file_init();
7040 }
7041}
7042
Michal Hocko2d110852013-02-22 16:34:43 -08007043#else
Michal Hocko6acc8b02013-02-22 16:34:45 -08007044static void __init enable_swap_cgroup(void)
Michal Hocko2d110852013-02-22 16:34:43 -08007045{
7046}
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007047#endif
Michal Hocko2d110852013-02-22 16:34:43 -08007048
7049/*
Michal Hocko10813122013-02-22 16:35:41 -08007050 * subsys_initcall() for memory controller.
7051 *
7052 * Some parts like hotcpu_notifier() have to be initialized from this context
7053 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
7054 * everything that doesn't depend on a specific mem_cgroup structure should
7055 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007056 */
7057static int __init mem_cgroup_init(void)
7058{
7059 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Michal Hocko6acc8b02013-02-22 16:34:45 -08007060 enable_swap_cgroup();
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07007061 mem_cgroup_soft_limit_tree_init();
Michal Hockoe4777492013-02-22 16:35:40 -08007062 memcg_stock_init();
Michal Hocko2d110852013-02-22 16:34:43 -08007063 return 0;
7064}
7065subsys_initcall(mem_cgroup_init);