blob: 750ed1449955885bfac56189075fe42ecee767fd [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 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080013 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 */
23
24#include <linux/res_counter.h>
25#include <linux/memcontrol.h>
26#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080027#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080028#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080029#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080030#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070035#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040036#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080037#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070038#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080041#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080043#include <linux/eventfd.h>
44#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080045#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080046#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070047#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070048#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080050#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070051#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080052#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000053#include <net/sock.h>
54#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080055
Balbir Singh8697d332008-02-07 00:13:59 -080056#include <asm/uaccess.h>
57
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070058#include <trace/events/vmscan.h>
59
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070060struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070061#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070062struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070065/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080066int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080067
68/* for remember boot option*/
69#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70static int really_do_swap_account __initdata = 1;
71#else
72static int really_do_swap_account __initdata = 0;
73#endif
74
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075#else
76#define do_swap_account (0)
77#endif
78
79
Balbir Singh8cdea7c2008-02-07 00:13:50 -080080/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080081 * Statistics for memory cgroup.
82 */
83enum mem_cgroup_stat_index {
84 /*
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 */
87 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070088 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080089 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070090 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -070091 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -070092 MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080093 MEM_CGROUP_STAT_NSTATS,
94};
95
Johannes Weinere9f89742011-03-23 16:42:37 -070096enum mem_cgroup_events_index {
97 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
98 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
99 MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
Ying Han456f9982011-05-26 16:25:38 -0700100 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700102 MEM_CGROUP_EVENTS_NSTATS,
103};
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700104/*
105 * Per memcg event counter is incremented at every pagein/pageout. With THP,
106 * it will be incremated by the number of pages. This counter is used for
107 * for trigger some periodic events. This is straightforward and better
108 * than using jiffies etc. to handle periodic memcg event.
109 */
110enum mem_cgroup_events_target {
111 MEM_CGROUP_TARGET_THRESH,
112 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700113 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700114 MEM_CGROUP_NTARGETS,
115};
116#define THRESHOLDS_EVENTS_TARGET (128)
117#define SOFTLIMIT_EVENTS_TARGET (1024)
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700118#define NUMAINFO_EVENTS_TARGET (1024)
Johannes Weinere9f89742011-03-23 16:42:37 -0700119
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800120struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700121 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700122 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700123 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800124};
125
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800126struct mem_cgroup_reclaim_iter {
127 /* css_id of the last scanned hierarchy member */
128 int position;
129 /* scan generation, increased every round-trip */
130 unsigned int generation;
131};
132
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800133/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800134 * per-zone information in memory controller.
135 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800136struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800137 /*
138 * spin_lock to protect the per cgroup LRU
139 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700140 struct list_head lists[NR_LRU_LISTS];
141 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800142
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800143 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
144
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800145 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700146 struct rb_node tree_node; /* RB tree node */
147 unsigned long long usage_in_excess;/* Set to the value by which */
148 /* the soft limit is exceeded*/
149 bool on_tree;
Balbir Singh4e416952009-09-23 15:56:39 -0700150 struct mem_cgroup *mem; /* Back pointer, we cannot */
151 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800152};
153/* Macro for accessing counter */
154#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
155
156struct mem_cgroup_per_node {
157 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
158};
159
160struct mem_cgroup_lru_info {
161 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
162};
163
164/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700165 * Cgroups above their limits are maintained in a RB-Tree, independent of
166 * their hierarchy representation
167 */
168
169struct mem_cgroup_tree_per_zone {
170 struct rb_root rb_root;
171 spinlock_t lock;
172};
173
174struct mem_cgroup_tree_per_node {
175 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
176};
177
178struct mem_cgroup_tree {
179 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
180};
181
182static struct mem_cgroup_tree soft_limit_tree __read_mostly;
183
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800184struct mem_cgroup_threshold {
185 struct eventfd_ctx *eventfd;
186 u64 threshold;
187};
188
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700189/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800190struct mem_cgroup_threshold_ary {
191 /* An array index points to threshold just below usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700192 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800193 /* Size of entries[] */
194 unsigned int size;
195 /* Array of thresholds */
196 struct mem_cgroup_threshold entries[0];
197};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700198
199struct mem_cgroup_thresholds {
200 /* Primary thresholds array */
201 struct mem_cgroup_threshold_ary *primary;
202 /*
203 * Spare threshold array.
204 * This is needed to make mem_cgroup_unregister_event() "never fail".
205 * It must be able to store at least primary->size - 1 entries.
206 */
207 struct mem_cgroup_threshold_ary *spare;
208};
209
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700210/* for OOM */
211struct mem_cgroup_eventfd_list {
212 struct list_head list;
213 struct eventfd_ctx *eventfd;
214};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800215
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700216static void mem_cgroup_threshold(struct mem_cgroup *memcg);
217static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800218
Balbir Singhf64c3f52009-09-23 15:56:37 -0700219/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800220 * The memory controller data structure. The memory controller controls both
221 * page cache and RSS per cgroup. We would eventually like to provide
222 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
223 * to help the administrator determine what knobs to tune.
224 *
225 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800226 * we hit the water mark. May be even add a low water mark, such that
227 * no reclaim occurs from a cgroup at it's low water mark, this is
228 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800229 */
230struct mem_cgroup {
231 struct cgroup_subsys_state css;
232 /*
233 * the counter to account for memory usage
234 */
235 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800236 /*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800237 * the counter to account for mem+swap usage.
238 */
239 struct res_counter memsw;
240 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800241 * Per cgroup active and inactive list, similar to the
242 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800243 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800244 struct mem_cgroup_lru_info info;
Ying Han889976d2011-05-26 16:25:33 -0700245 int last_scanned_node;
246#if MAX_NUMNODES > 1
247 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700248 atomic_t numainfo_events;
249 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700250#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800251 /*
252 * Should the accounting and control be hierarchical, per subtree?
253 */
254 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700255
256 bool oom_lock;
257 atomic_t under_oom;
258
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800259 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800260
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700261 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700262 /* OOM-Killer disable */
263 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800264
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700265 /* set when res.limit == memsw.limit */
266 bool memsw_is_minimum;
267
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800268 /* protect arrays of thresholds */
269 struct mutex thresholds_lock;
270
271 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700272 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700273
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800274 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700275 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700276
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700277 /* For oom notifier event fd */
278 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700279
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800280 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800281 * Should we move charges of a task when a task is moved into this
282 * mem_cgroup ? And what type of charges should we move ?
283 */
284 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800285 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800286 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800287 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800288 struct mem_cgroup_stat_cpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700289 /*
290 * used when a cpu is offlined or other synchronizations
291 * See mem_cgroup_read_stat().
292 */
293 struct mem_cgroup_stat_cpu nocpu_base;
294 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000295
296#ifdef CONFIG_INET
297 struct tcp_memcontrol tcp_mem;
298#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800299};
300
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800301/* Stuffs for move charges at task migration. */
302/*
303 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
304 * left-shifted bitmap of these types.
305 */
306enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800307 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700308 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800309 NR_MOVE_TYPE,
310};
311
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800312/* "mc" and its members are protected by cgroup_mutex */
313static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800314 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800315 struct mem_cgroup *from;
316 struct mem_cgroup *to;
317 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800318 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800319 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800320 struct task_struct *moving_task; /* a task moving charges */
321 wait_queue_head_t waitq; /* a waitq for other context */
322} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700323 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800324 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
325};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800326
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700327static bool move_anon(void)
328{
329 return test_bit(MOVE_CHARGE_TYPE_ANON,
330 &mc.to->move_charge_at_immigrate);
331}
332
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700333static bool move_file(void)
334{
335 return test_bit(MOVE_CHARGE_TYPE_FILE,
336 &mc.to->move_charge_at_immigrate);
337}
338
Balbir Singh4e416952009-09-23 15:56:39 -0700339/*
340 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
341 * limit reclaim to prevent infinite loops, if they ever occur.
342 */
343#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
344#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
345
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800346enum charge_type {
347 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
348 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700349 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700350 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800351 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700352 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700353 NR_CHARGE_TYPE,
354};
355
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800356/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000357#define _MEM (0)
358#define _MEMSWAP (1)
359#define _OOM_TYPE (2)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800360#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
361#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
362#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700363/* Used for OOM nofiier */
364#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800365
Balbir Singh75822b42009-09-23 15:56:38 -0700366/*
367 * Reclaim flags for mem_cgroup_hierarchical_reclaim
368 */
369#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
370#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
371#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
372#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
373
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700374static void mem_cgroup_get(struct mem_cgroup *memcg);
375static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000376
377/* Writing them here to avoid exposing memcg's inner layout */
378#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
379#ifdef CONFIG_INET
380#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000381#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000382
383static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
384void sock_update_memcg(struct sock *sk)
385{
Glauber Costae1aab162011-12-11 21:47:03 +0000386 if (static_branch(&memcg_socket_limit_enabled)) {
387 struct mem_cgroup *memcg;
388
389 BUG_ON(!sk->sk_prot->proto_cgroup);
390
Glauber Costaf3f511e2012-01-05 20:16:39 +0000391 /* Socket cloning can throw us here with sk_cgrp already
392 * filled. It won't however, necessarily happen from
393 * process context. So the test for root memcg given
394 * the current task's memcg won't help us in this case.
395 *
396 * Respecting the original socket's memcg is a better
397 * decision in this case.
398 */
399 if (sk->sk_cgrp) {
400 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
401 mem_cgroup_get(sk->sk_cgrp->memcg);
402 return;
403 }
404
Glauber Costae1aab162011-12-11 21:47:03 +0000405 rcu_read_lock();
406 memcg = mem_cgroup_from_task(current);
407 if (!mem_cgroup_is_root(memcg)) {
408 mem_cgroup_get(memcg);
409 sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg);
410 }
411 rcu_read_unlock();
412 }
413}
414EXPORT_SYMBOL(sock_update_memcg);
415
416void sock_release_memcg(struct sock *sk)
417{
418 if (static_branch(&memcg_socket_limit_enabled) && sk->sk_cgrp) {
419 struct mem_cgroup *memcg;
420 WARN_ON(!sk->sk_cgrp->memcg);
421 memcg = sk->sk_cgrp->memcg;
422 mem_cgroup_put(memcg);
423 }
424}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000425
426struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
427{
428 if (!memcg || mem_cgroup_is_root(memcg))
429 return NULL;
430
431 return &memcg->tcp_mem.cg_proto;
432}
433EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000434#endif /* CONFIG_INET */
435#endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
436
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700437static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800438
Balbir Singhf64c3f52009-09-23 15:56:37 -0700439static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700440mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700441{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700442 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700443}
444
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700445struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100446{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700447 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100448}
449
Balbir Singhf64c3f52009-09-23 15:56:37 -0700450static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700451page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700452{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700453 int nid = page_to_nid(page);
454 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700455
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700456 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700457}
458
459static struct mem_cgroup_tree_per_zone *
460soft_limit_tree_node_zone(int nid, int zid)
461{
462 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
463}
464
465static struct mem_cgroup_tree_per_zone *
466soft_limit_tree_from_page(struct page *page)
467{
468 int nid = page_to_nid(page);
469 int zid = page_zonenum(page);
470
471 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
472}
473
474static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700475__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700476 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700477 struct mem_cgroup_tree_per_zone *mctz,
478 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700479{
480 struct rb_node **p = &mctz->rb_root.rb_node;
481 struct rb_node *parent = NULL;
482 struct mem_cgroup_per_zone *mz_node;
483
484 if (mz->on_tree)
485 return;
486
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700487 mz->usage_in_excess = new_usage_in_excess;
488 if (!mz->usage_in_excess)
489 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700490 while (*p) {
491 parent = *p;
492 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
493 tree_node);
494 if (mz->usage_in_excess < mz_node->usage_in_excess)
495 p = &(*p)->rb_left;
496 /*
497 * We can't avoid mem cgroups that are over their soft
498 * limit by the same amount
499 */
500 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
501 p = &(*p)->rb_right;
502 }
503 rb_link_node(&mz->tree_node, parent, p);
504 rb_insert_color(&mz->tree_node, &mctz->rb_root);
505 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700506}
507
508static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700509__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700510 struct mem_cgroup_per_zone *mz,
511 struct mem_cgroup_tree_per_zone *mctz)
512{
513 if (!mz->on_tree)
514 return;
515 rb_erase(&mz->tree_node, &mctz->rb_root);
516 mz->on_tree = false;
517}
518
519static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700520mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700521 struct mem_cgroup_per_zone *mz,
522 struct mem_cgroup_tree_per_zone *mctz)
523{
524 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700525 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700526 spin_unlock(&mctz->lock);
527}
528
Balbir Singhf64c3f52009-09-23 15:56:37 -0700529
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700530static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700531{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700532 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700533 struct mem_cgroup_per_zone *mz;
534 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700535 int nid = page_to_nid(page);
536 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700537 mctz = soft_limit_tree_from_page(page);
538
539 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700540 * Necessary to update all ancestors when hierarchy is used.
541 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700542 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700543 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
544 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
545 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700546 /*
547 * We have to update the tree if mz is on RB-tree or
548 * mem is over its softlimit.
549 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700550 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700551 spin_lock(&mctz->lock);
552 /* if on-tree, remove it */
553 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700554 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700555 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700556 * Insert again. mz->usage_in_excess will be updated.
557 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700558 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700559 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700560 spin_unlock(&mctz->lock);
561 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700562 }
563}
564
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700565static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700566{
567 int node, zone;
568 struct mem_cgroup_per_zone *mz;
569 struct mem_cgroup_tree_per_zone *mctz;
570
571 for_each_node_state(node, N_POSSIBLE) {
572 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700573 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700574 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700575 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700576 }
577 }
578}
579
Balbir Singh4e416952009-09-23 15:56:39 -0700580static struct mem_cgroup_per_zone *
581__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
582{
583 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700584 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700585
586retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700587 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700588 rightmost = rb_last(&mctz->rb_root);
589 if (!rightmost)
590 goto done; /* Nothing to reclaim from */
591
592 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
593 /*
594 * Remove the node now but someone else can add it back,
595 * we will to add it back at the end of reclaim to its correct
596 * position in the tree.
597 */
598 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
599 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
600 !css_tryget(&mz->mem->css))
601 goto retry;
602done:
603 return mz;
604}
605
606static struct mem_cgroup_per_zone *
607mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
608{
609 struct mem_cgroup_per_zone *mz;
610
611 spin_lock(&mctz->lock);
612 mz = __mem_cgroup_largest_soft_limit_node(mctz);
613 spin_unlock(&mctz->lock);
614 return mz;
615}
616
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700617/*
618 * Implementation Note: reading percpu statistics for memcg.
619 *
620 * Both of vmstat[] and percpu_counter has threshold and do periodic
621 * synchronization to implement "quick" read. There are trade-off between
622 * reading cost and precision of value. Then, we may have a chance to implement
623 * a periodic synchronizion of counter in memcg's counter.
624 *
625 * But this _read() function is used for user interface now. The user accounts
626 * memory usage by memory cgroup and he _always_ requires exact value because
627 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
628 * have to visit all online cpus and make sum. So, for now, unnecessary
629 * synchronization is not implemented. (just implemented for cpu hotplug)
630 *
631 * If there are kernel internal actions which can make use of some not-exact
632 * value, and reading all cpu value can be performance bottleneck in some
633 * common workload, threashold and synchonization as vmstat[] should be
634 * implemented.
635 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700636static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700637 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800638{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700639 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800640 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800641
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700642 get_online_cpus();
643 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700644 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700645#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700646 spin_lock(&memcg->pcp_counter_lock);
647 val += memcg->nocpu_base.count[idx];
648 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700649#endif
650 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800651 return val;
652}
653
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700654static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700655 bool charge)
656{
657 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700658 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700659}
660
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700661void mem_cgroup_pgfault(struct mem_cgroup *memcg, int val)
Ying Han456f9982011-05-26 16:25:38 -0700662{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700663 this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val);
Ying Han456f9982011-05-26 16:25:38 -0700664}
665
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700666void mem_cgroup_pgmajfault(struct mem_cgroup *memcg, int val)
Ying Han456f9982011-05-26 16:25:38 -0700667{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700668 this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val);
Ying Han456f9982011-05-26 16:25:38 -0700669}
670
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700671static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700672 enum mem_cgroup_events_index idx)
673{
674 unsigned long val = 0;
675 int cpu;
676
677 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700678 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700679#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680 spin_lock(&memcg->pcp_counter_lock);
681 val += memcg->nocpu_base.events[idx];
682 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700683#endif
684 return val;
685}
686
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700687static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800688 bool file, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800689{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800690 preempt_disable();
691
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800692 if (file)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700693 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
694 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800695 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700696 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
697 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700698
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800699 /* pagein of a big page is an event. So, ignore page size */
700 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700701 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800702 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700703 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800704 nr_pages = -nr_pages; /* for event */
705 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800706
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700707 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800708
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800709 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800710}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800711
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700712unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700713mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700714 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700715{
716 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700717 enum lru_list l;
718 unsigned long ret = 0;
719
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700720 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700721
722 for_each_lru(l) {
723 if (BIT(l) & lru_mask)
724 ret += MEM_CGROUP_ZSTAT(mz, l);
725 }
726 return ret;
727}
728
729static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700730mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700731 int nid, unsigned int lru_mask)
732{
Ying Han889976d2011-05-26 16:25:33 -0700733 u64 total = 0;
734 int zid;
735
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700736 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700737 total += mem_cgroup_zone_nr_lru_pages(memcg,
738 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700739
Ying Han889976d2011-05-26 16:25:33 -0700740 return total;
741}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700742
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700744 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800745{
Ying Han889976d2011-05-26 16:25:33 -0700746 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800747 u64 total = 0;
748
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700749 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700750 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800751 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800752}
753
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700754static bool __memcg_event_check(struct mem_cgroup *memcg, int target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800755{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700756 unsigned long val, next;
757
Steven Rostedt47994012011-11-02 13:38:33 -0700758 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
759 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700760 /* from time_after() in jiffies.h */
761 return ((long)next - (long)val < 0);
762}
763
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764static void __mem_cgroup_target_update(struct mem_cgroup *memcg, int target)
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700765{
766 unsigned long val, next;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800767
Steven Rostedt47994012011-11-02 13:38:33 -0700768 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800769
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700770 switch (target) {
771 case MEM_CGROUP_TARGET_THRESH:
772 next = val + THRESHOLDS_EVENTS_TARGET;
773 break;
774 case MEM_CGROUP_TARGET_SOFTLIMIT:
775 next = val + SOFTLIMIT_EVENTS_TARGET;
776 break;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700777 case MEM_CGROUP_TARGET_NUMAINFO:
778 next = val + NUMAINFO_EVENTS_TARGET;
779 break;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700780 default:
781 return;
782 }
783
Steven Rostedt47994012011-11-02 13:38:33 -0700784 __this_cpu_write(memcg->stat->targets[target], next);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800785}
786
787/*
788 * Check events in order.
789 *
790 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700791static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800792{
Steven Rostedt47994012011-11-02 13:38:33 -0700793 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800794 /* threshold event is triggered in finer grain than soft limit */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700795 if (unlikely(__memcg_event_check(memcg, MEM_CGROUP_TARGET_THRESH))) {
796 mem_cgroup_threshold(memcg);
797 __mem_cgroup_target_update(memcg, MEM_CGROUP_TARGET_THRESH);
798 if (unlikely(__memcg_event_check(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700799 MEM_CGROUP_TARGET_SOFTLIMIT))) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700800 mem_cgroup_update_tree(memcg, page);
801 __mem_cgroup_target_update(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700802 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700803 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700804#if MAX_NUMNODES > 1
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700805 if (unlikely(__memcg_event_check(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700806 MEM_CGROUP_TARGET_NUMAINFO))) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700807 atomic_inc(&memcg->numainfo_events);
808 __mem_cgroup_target_update(memcg,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700809 MEM_CGROUP_TARGET_NUMAINFO);
810 }
811#endif
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800812 }
Steven Rostedt47994012011-11-02 13:38:33 -0700813 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800814}
815
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000816struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800817{
818 return container_of(cgroup_subsys_state(cont,
819 mem_cgroup_subsys_id), struct mem_cgroup,
820 css);
821}
822
Balbir Singhcf475ad2008-04-29 01:00:16 -0700823struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800824{
Balbir Singh31a78f22008-09-28 23:09:31 +0100825 /*
826 * mm_update_next_owner() may clear mm->owner to NULL
827 * if it races with swapoff, page migration, etc.
828 * So this can be called with p == NULL.
829 */
830 if (unlikely(!p))
831 return NULL;
832
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800833 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
834 struct mem_cgroup, css);
835}
836
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700837struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800838{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700839 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700840
841 if (!mm)
842 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800843 /*
844 * Because we have no locks, mm->owner's may be being moved to other
845 * cgroup. We use css_tryget() here even if this looks
846 * pessimistic (rather than adding locks here).
847 */
848 rcu_read_lock();
849 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700850 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
851 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800852 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700853 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800854 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700855 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800856}
857
Johannes Weiner56600482012-01-12 17:17:59 -0800858/**
859 * mem_cgroup_iter - iterate over memory cgroup hierarchy
860 * @root: hierarchy root
861 * @prev: previously returned memcg, NULL on first invocation
862 * @reclaim: cookie for shared reclaim walks, NULL for full walks
863 *
864 * Returns references to children of the hierarchy below @root, or
865 * @root itself, or %NULL after a full round-trip.
866 *
867 * Caller must pass the return value in @prev on subsequent
868 * invocations for reference counting, or use mem_cgroup_iter_break()
869 * to cancel a hierarchy walk before the round-trip is complete.
870 *
871 * Reclaimers can specify a zone and a priority level in @reclaim to
872 * divide up the memcgs in the hierarchy among all concurrent
873 * reclaimers operating on the same zone and priority.
874 */
875struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
876 struct mem_cgroup *prev,
877 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700878{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800879 struct mem_cgroup *memcg = NULL;
880 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700881
Johannes Weiner56600482012-01-12 17:17:59 -0800882 if (mem_cgroup_disabled())
883 return NULL;
884
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700885 if (!root)
886 root = root_mem_cgroup;
887
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800888 if (prev && !reclaim)
889 id = css_id(&prev->css);
890
891 if (prev && prev != root)
892 css_put(&prev->css);
893
894 if (!root->use_hierarchy && root != root_mem_cgroup) {
895 if (prev)
896 return NULL;
897 return root;
898 }
899
900 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800901 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800902 struct cgroup_subsys_state *css;
903
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800904 if (reclaim) {
905 int nid = zone_to_nid(reclaim->zone);
906 int zid = zone_idx(reclaim->zone);
907 struct mem_cgroup_per_zone *mz;
908
909 mz = mem_cgroup_zoneinfo(root, nid, zid);
910 iter = &mz->reclaim_iter[reclaim->priority];
911 if (prev && reclaim->generation != iter->generation)
912 return NULL;
913 id = iter->position;
914 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800915
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700916 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800917 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
918 if (css) {
919 if (css == &root->css || css_tryget(css))
920 memcg = container_of(css,
921 struct mem_cgroup, css);
922 } else
923 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700924 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700925
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800926 if (reclaim) {
927 iter->position = id;
928 if (!css)
929 iter->generation++;
930 else if (!prev && memcg)
931 reclaim->generation = iter->generation;
932 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800933
934 if (prev && !css)
935 return NULL;
936 }
937 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700938}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800939
Johannes Weiner56600482012-01-12 17:17:59 -0800940/**
941 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
942 * @root: hierarchy root
943 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
944 */
945void mem_cgroup_iter_break(struct mem_cgroup *root,
946 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800947{
948 if (!root)
949 root = root_mem_cgroup;
950 if (prev && prev != root)
951 css_put(&prev->css);
952}
953
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700954/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800955 * Iteration constructs for visiting all cgroups (under a tree). If
956 * loops are exited prematurely (break), mem_cgroup_iter_break() must
957 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700958 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800959#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800960 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800961 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800962 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700963
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800964#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800965 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800966 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800967 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700968
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700969static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700970{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700971 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700972}
973
Ying Han456f9982011-05-26 16:25:38 -0700974void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
975{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700976 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700977
978 if (!mm)
979 return;
980
981 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700982 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
983 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700984 goto out;
985
986 switch (idx) {
987 case PGMAJFAULT:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700988 mem_cgroup_pgmajfault(memcg, 1);
Ying Han456f9982011-05-26 16:25:38 -0700989 break;
990 case PGFAULT:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700991 mem_cgroup_pgfault(memcg, 1);
Ying Han456f9982011-05-26 16:25:38 -0700992 break;
993 default:
994 BUG();
995 }
996out:
997 rcu_read_unlock();
998}
999EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1000
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001001/*
1002 * Following LRU functions are allowed to be used without PCG_LOCK.
1003 * Operations are called by routine of global LRU independently from memcg.
1004 * What we have to take care of here is validness of pc->mem_cgroup.
1005 *
1006 * Changes to pc->mem_cgroup happens when
1007 * 1. charge
1008 * 2. moving account
1009 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1010 * It is added to LRU before charge.
1011 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1012 * When moving account, the page is not on LRU. It's isolated.
1013 */
1014
1015void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001016{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001017 struct page_cgroup *pc;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001018 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001019
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001020 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001021 return;
1022 pc = lookup_page_cgroup(page);
1023 /* can happen while we handle swapcache. */
Balbir Singh4b3bde42009-09-23 15:56:32 -07001024 if (!TestClearPageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001025 return;
Balbir Singh4b3bde42009-09-23 15:56:32 -07001026 VM_BUG_ON(!pc->mem_cgroup);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001027 /*
1028 * We don't check PCG_USED bit. It's cleared when the "page" is finally
1029 * removed from global LRU.
1030 */
Johannes Weiner97a6c372011-03-23 16:42:27 -07001031 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08001032 /* huge page split is done under lru_lock. so, we have no races. */
1033 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001034 if (mem_cgroup_is_root(pc->mem_cgroup))
1035 return;
1036 VM_BUG_ON(list_empty(&pc->lru));
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001037 list_del_init(&pc->lru);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001038}
1039
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001040void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001041{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001042 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001043}
1044
Minchan Kim3f58a822011-03-22 16:32:53 -07001045/*
1046 * Writeback is about to end against a page which has been marked for immediate
1047 * reclaim. If it still appears to be reclaimable, move it to the tail of the
1048 * inactive list.
1049 */
1050void mem_cgroup_rotate_reclaimable_page(struct page *page)
1051{
1052 struct mem_cgroup_per_zone *mz;
1053 struct page_cgroup *pc;
1054 enum lru_list lru = page_lru(page);
1055
1056 if (mem_cgroup_disabled())
1057 return;
1058
1059 pc = lookup_page_cgroup(page);
1060 /* unused or root page is not rotated. */
1061 if (!PageCgroupUsed(pc))
1062 return;
1063 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1064 smp_rmb();
1065 if (mem_cgroup_is_root(pc->mem_cgroup))
1066 return;
Johannes Weiner97a6c372011-03-23 16:42:27 -07001067 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
Minchan Kim3f58a822011-03-22 16:32:53 -07001068 list_move_tail(&pc->lru, &mz->lists[lru]);
1069}
1070
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001071void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -08001072{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001073 struct mem_cgroup_per_zone *mz;
1074 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001075
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001076 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001077 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001078
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001079 pc = lookup_page_cgroup(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001080 /* unused or root page is not rotated. */
Johannes Weiner713735b2011-01-20 14:44:31 -08001081 if (!PageCgroupUsed(pc))
1082 return;
1083 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1084 smp_rmb();
1085 if (mem_cgroup_is_root(pc->mem_cgroup))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001086 return;
Johannes Weiner97a6c372011-03-23 16:42:27 -07001087 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001088 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -08001089}
1090
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001091void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
1092{
1093 struct page_cgroup *pc;
1094 struct mem_cgroup_per_zone *mz;
1095
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001096 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001097 return;
1098 pc = lookup_page_cgroup(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001099 VM_BUG_ON(PageCgroupAcctLRU(pc));
Johannes Weinera61ed3c2011-11-02 13:38:29 -07001100 /*
1101 * putback: charge:
1102 * SetPageLRU SetPageCgroupUsed
1103 * smp_mb smp_mb
1104 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1105 *
1106 * Ensure that one of the two sides adds the page to the memcg
1107 * LRU during a race.
1108 */
1109 smp_mb();
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001110 if (!PageCgroupUsed(pc))
1111 return;
Johannes Weiner713735b2011-01-20 14:44:31 -08001112 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1113 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001114 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08001115 /* huge page split is done under lru_lock. so, we have no races. */
1116 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -07001117 SetPageCgroupAcctLRU(pc);
1118 if (mem_cgroup_is_root(pc->mem_cgroup))
1119 return;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001120 list_add(&pc->lru, &mz->lists[lru]);
1121}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001122
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001123/*
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001124 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1125 * while it's linked to lru because the page may be reused after it's fully
1126 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1127 * It's done under lock_page and expected that zone->lru_lock isnever held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001128 */
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001129static void mem_cgroup_lru_del_before_commit(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001130{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001131 unsigned long flags;
1132 struct zone *zone = page_zone(page);
1133 struct page_cgroup *pc = lookup_page_cgroup(page);
1134
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001135 /*
1136 * Doing this check without taking ->lru_lock seems wrong but this
1137 * is safe. Because if page_cgroup's USED bit is unset, the page
1138 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1139 * set, the commit after this will fail, anyway.
1140 * This all charge/uncharge is done under some mutual execustion.
1141 * So, we don't need to taking care of changes in USED bit.
1142 */
1143 if (likely(!PageLRU(page)))
1144 return;
1145
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001146 spin_lock_irqsave(&zone->lru_lock, flags);
1147 /*
1148 * Forget old LRU when this page_cgroup is *not* used. This Used bit
1149 * is guarded by lock_page() because the page is SwapCache.
1150 */
1151 if (!PageCgroupUsed(pc))
1152 mem_cgroup_del_lru_list(page, page_lru(page));
1153 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001154}
1155
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001156static void mem_cgroup_lru_add_after_commit(struct page *page)
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001157{
1158 unsigned long flags;
1159 struct zone *zone = page_zone(page);
1160 struct page_cgroup *pc = lookup_page_cgroup(page);
Johannes Weinera61ed3c2011-11-02 13:38:29 -07001161 /*
1162 * putback: charge:
1163 * SetPageLRU SetPageCgroupUsed
1164 * smp_mb smp_mb
1165 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1166 *
1167 * Ensure that one of the two sides adds the page to the memcg
1168 * LRU during a race.
1169 */
1170 smp_mb();
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07001171 /* taking care of that the page is added to LRU while we commit it */
1172 if (likely(!PageLRU(page)))
1173 return;
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001174 spin_lock_irqsave(&zone->lru_lock, flags);
1175 /* link when the page is linked to LRU but page_cgroup isn't */
Balbir Singh4b3bde42009-09-23 15:56:32 -07001176 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001177 mem_cgroup_add_lru_list(page, page_lru(page));
1178 spin_unlock_irqrestore(&zone->lru_lock, flags);
1179}
1180
1181
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001182void mem_cgroup_move_lists(struct page *page,
1183 enum lru_list from, enum lru_list to)
1184{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001185 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001186 return;
1187 mem_cgroup_del_lru_list(page, from);
1188 mem_cgroup_add_lru_list(page, to);
1189}
1190
Michal Hocko3e920412011-07-26 16:08:29 -07001191/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001192 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001193 * hierarchy subtree
1194 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001195static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1196 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001197{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001198 if (root_memcg != memcg) {
1199 return (root_memcg->use_hierarchy &&
1200 css_is_ancestor(&memcg->css, &root_memcg->css));
Michal Hocko3e920412011-07-26 16:08:29 -07001201 }
1202
1203 return true;
1204}
1205
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001206int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001207{
1208 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001209 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001210 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001211
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001212 p = find_lock_task_mm(task);
1213 if (!p)
1214 return 0;
1215 curr = try_get_mem_cgroup_from_mm(p->mm);
1216 task_unlock(p);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001217 if (!curr)
1218 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001219 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001220 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001221 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001222 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1223 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001224 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001225 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001226 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001227 return ret;
1228}
1229
Johannes Weiner9b272972011-11-02 13:38:23 -07001230int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001231{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001232 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001233 int nid = zone_to_nid(zone);
1234 int zid = zone_idx(zone);
1235 unsigned long inactive;
1236 unsigned long active;
1237 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001238
Johannes Weiner9b272972011-11-02 13:38:23 -07001239 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1240 BIT(LRU_INACTIVE_ANON));
1241 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1242 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001243
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001244 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1245 if (gb)
1246 inactive_ratio = int_sqrt(10 * gb);
1247 else
1248 inactive_ratio = 1;
1249
Johannes Weiner9b272972011-11-02 13:38:23 -07001250 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001251}
1252
Johannes Weiner9b272972011-11-02 13:38:23 -07001253int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001254{
1255 unsigned long active;
1256 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001257 int zid = zone_idx(zone);
1258 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001259
Johannes Weiner9b272972011-11-02 13:38:23 -07001260 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1261 BIT(LRU_INACTIVE_FILE));
1262 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1263 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001264
1265 return (active > inactive);
1266}
1267
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001268struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1269 struct zone *zone)
1270{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001271 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001272 int zid = zone_idx(zone);
1273 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1274
1275 return &mz->reclaim_stat;
1276}
1277
1278struct zone_reclaim_stat *
1279mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1280{
1281 struct page_cgroup *pc;
1282 struct mem_cgroup_per_zone *mz;
1283
1284 if (mem_cgroup_disabled())
1285 return NULL;
1286
1287 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001288 if (!PageCgroupUsed(pc))
1289 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001290 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1291 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001292 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001293 return &mz->reclaim_stat;
1294}
1295
Balbir Singh66e17072008-02-07 00:13:56 -08001296unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
1297 struct list_head *dst,
1298 unsigned long *scanned, int order,
Minchan Kim4356f212011-10-31 17:06:47 -07001299 isolate_mode_t mode,
1300 struct zone *z,
Balbir Singh66e17072008-02-07 00:13:56 -08001301 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001302 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -08001303{
1304 unsigned long nr_taken = 0;
1305 struct page *page;
1306 unsigned long scan;
1307 LIST_HEAD(pc_list);
1308 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001309 struct page_cgroup *pc, *tmp;
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001310 int nid = zone_to_nid(z);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001311 int zid = zone_idx(z);
1312 struct mem_cgroup_per_zone *mz;
Johannes Weinerb7c46d12009-09-21 17:02:56 -07001313 int lru = LRU_FILE * file + active;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001314 int ret;
Balbir Singh66e17072008-02-07 00:13:56 -08001315
Balbir Singhcf475ad2008-04-29 01:00:16 -07001316 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001317 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001318 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -08001319
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001320 scan = 0;
1321 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -08001322 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001323 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001324
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001325 if (unlikely(!PageCgroupUsed(pc)))
1326 continue;
Johannes Weiner5564e882011-03-23 16:42:29 -07001327
Johannes Weiner6b3ae582011-03-23 16:42:30 -07001328 page = lookup_cgroup_page(pc);
Johannes Weiner5564e882011-03-23 16:42:29 -07001329
Hugh Dickins436c65412008-02-07 00:14:12 -08001330 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001331 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -08001332
Hugh Dickins436c65412008-02-07 00:14:12 -08001333 scan++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001334 ret = __isolate_lru_page(page, mode, file);
1335 switch (ret) {
1336 case 0:
Balbir Singh66e17072008-02-07 00:13:56 -08001337 list_move(&page->lru, dst);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001338 mem_cgroup_del_lru(page);
Rik van Riel2c888cf2011-01-13 15:47:13 -08001339 nr_taken += hpage_nr_pages(page);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -07001340 break;
1341 case -EBUSY:
1342 /* we don't affect global LRU but rotate in our LRU */
1343 mem_cgroup_rotate_lru_list(page, page_lru(page));
1344 break;
1345 default:
1346 break;
Balbir Singh66e17072008-02-07 00:13:56 -08001347 }
1348 }
1349
Balbir Singh66e17072008-02-07 00:13:56 -08001350 *scanned = scan;
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -07001351
1352 trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
1353 0, 0, 0, mode);
1354
Balbir Singh66e17072008-02-07 00:13:56 -08001355 return nr_taken;
1356}
1357
Balbir Singh6d61ef42009-01-07 18:08:06 -08001358#define mem_cgroup_from_res_counter(counter, member) \
1359 container_of(counter, struct mem_cgroup, member)
1360
Johannes Weiner19942822011-02-01 15:52:43 -08001361/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001362 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1363 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001364 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001365 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001366 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001367 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001368static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001369{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001370 unsigned long long margin;
1371
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001372 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001373 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001374 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001375 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001376}
1377
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001378int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001379{
1380 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001381
1382 /* root ? */
1383 if (cgrp->parent == NULL)
1384 return vm_swappiness;
1385
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001386 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001387}
1388
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001389static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001390{
1391 int cpu;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001392
1393 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001394 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001395 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001396 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1397 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1398 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001399 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001400
1401 synchronize_rcu();
1402}
1403
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001404static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001405{
1406 int cpu;
1407
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001408 if (!memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001409 return;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001410 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001411 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001412 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001413 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1414 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1415 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001416 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001417}
1418/*
1419 * 2 routines for checking "mem" is under move_account() or not.
1420 *
1421 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1422 * for avoiding race in accounting. If true,
1423 * pc->mem_cgroup may be overwritten.
1424 *
1425 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1426 * under hierarchy of moving cgroups. This is for
1427 * waiting at hith-memory prressure caused by "move".
1428 */
1429
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001430static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001431{
1432 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001433 return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001434}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001435
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001436static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001437{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001438 struct mem_cgroup *from;
1439 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001440 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001441 /*
1442 * Unlike task_move routines, we access mc.to, mc.from not under
1443 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1444 */
1445 spin_lock(&mc.lock);
1446 from = mc.from;
1447 to = mc.to;
1448 if (!from)
1449 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001450
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001451 ret = mem_cgroup_same_or_subtree(memcg, from)
1452 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001453unlock:
1454 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001455 return ret;
1456}
1457
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001458static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001459{
1460 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001461 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001462 DEFINE_WAIT(wait);
1463 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1464 /* moving charge context might have finished. */
1465 if (mc.moving_task)
1466 schedule();
1467 finish_wait(&mc.waitq, &wait);
1468 return true;
1469 }
1470 }
1471 return false;
1472}
1473
Balbir Singhe2224322009-04-02 16:57:39 -07001474/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001475 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001476 * @memcg: The memory cgroup that went over limit
1477 * @p: Task that is going to be killed
1478 *
1479 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1480 * enabled
1481 */
1482void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1483{
1484 struct cgroup *task_cgrp;
1485 struct cgroup *mem_cgrp;
1486 /*
1487 * Need a buffer in BSS, can't rely on allocations. The code relies
1488 * on the assumption that OOM is serialized for memory controller.
1489 * If this assumption is broken, revisit this code.
1490 */
1491 static char memcg_name[PATH_MAX];
1492 int ret;
1493
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001494 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001495 return;
1496
1497
1498 rcu_read_lock();
1499
1500 mem_cgrp = memcg->css.cgroup;
1501 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1502
1503 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1504 if (ret < 0) {
1505 /*
1506 * Unfortunately, we are unable to convert to a useful name
1507 * But we'll still print out the usage information
1508 */
1509 rcu_read_unlock();
1510 goto done;
1511 }
1512 rcu_read_unlock();
1513
1514 printk(KERN_INFO "Task in %s killed", memcg_name);
1515
1516 rcu_read_lock();
1517 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1518 if (ret < 0) {
1519 rcu_read_unlock();
1520 goto done;
1521 }
1522 rcu_read_unlock();
1523
1524 /*
1525 * Continues from above, so we don't need an KERN_ level
1526 */
1527 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1528done:
1529
1530 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1531 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1532 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1533 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1534 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1535 "failcnt %llu\n",
1536 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1537 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1538 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1539}
1540
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001541/*
1542 * This function returns the number of memcg under hierarchy tree. Returns
1543 * 1(self count) if no children.
1544 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001545static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001546{
1547 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001548 struct mem_cgroup *iter;
1549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001550 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001551 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001552 return num;
1553}
1554
Balbir Singh6d61ef42009-01-07 18:08:06 -08001555/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001556 * Return the memory (and swap, if configured) limit for a memcg.
1557 */
1558u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1559{
1560 u64 limit;
1561 u64 memsw;
1562
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001563 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1564 limit += total_swap_pages << PAGE_SHIFT;
1565
David Rientjesa63d83f2010-08-09 17:19:46 -07001566 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1567 /*
1568 * If memsw is finite and limits the amount of swap space available
1569 * to this memcg, return that limit.
1570 */
1571 return min(limit, memsw);
1572}
1573
Johannes Weiner56600482012-01-12 17:17:59 -08001574static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1575 gfp_t gfp_mask,
1576 unsigned long flags)
1577{
1578 unsigned long total = 0;
1579 bool noswap = false;
1580 int loop;
1581
1582 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1583 noswap = true;
1584 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1585 noswap = true;
1586
1587 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1588 if (loop)
1589 drain_all_stock_async(memcg);
1590 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1591 /*
1592 * Allow limit shrinkers, which are triggered directly
1593 * by userspace, to catch signals and stop reclaim
1594 * after minimal progress, regardless of the margin.
1595 */
1596 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1597 break;
1598 if (mem_cgroup_margin(memcg))
1599 break;
1600 /*
1601 * If nothing was reclaimed after two attempts, there
1602 * may be no reclaimable pages in this hierarchy.
1603 */
1604 if (loop && !total)
1605 break;
1606 }
1607 return total;
1608}
1609
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001610/**
1611 * test_mem_cgroup_node_reclaimable
1612 * @mem: the target memcg
1613 * @nid: the node ID to be checked.
1614 * @noswap : specify true here if the user wants flle only information.
1615 *
1616 * This function returns whether the specified memcg contains any
1617 * reclaimable pages on a node. Returns true if there are any reclaimable
1618 * pages in the node.
1619 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001620static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001621 int nid, bool noswap)
1622{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001623 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001624 return true;
1625 if (noswap || !total_swap_pages)
1626 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001628 return true;
1629 return false;
1630
1631}
Ying Han889976d2011-05-26 16:25:33 -07001632#if MAX_NUMNODES > 1
1633
1634/*
1635 * Always updating the nodemask is not very good - even if we have an empty
1636 * list or the wrong list here, we can start from some node and traverse all
1637 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1638 *
1639 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001640static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001641{
1642 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001643 /*
1644 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1645 * pagein/pageout changes since the last update.
1646 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001647 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001648 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001649 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001650 return;
1651
Ying Han889976d2011-05-26 16:25:33 -07001652 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001653 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001654
1655 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1656
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001657 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1658 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001659 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001660
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001661 atomic_set(&memcg->numainfo_events, 0);
1662 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001663}
1664
1665/*
1666 * Selecting a node where we start reclaim from. Because what we need is just
1667 * reducing usage counter, start from anywhere is O,K. Considering
1668 * memory reclaim from current node, there are pros. and cons.
1669 *
1670 * Freeing memory from current node means freeing memory from a node which
1671 * we'll use or we've used. So, it may make LRU bad. And if several threads
1672 * hit limits, it will see a contention on a node. But freeing from remote
1673 * node means more costs for memory reclaim because of memory latency.
1674 *
1675 * Now, we use round-robin. Better algorithm is welcomed.
1676 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001677int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001678{
1679 int node;
1680
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001681 mem_cgroup_may_update_nodemask(memcg);
1682 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001683
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001684 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001685 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001686 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001687 /*
1688 * We call this when we hit limit, not when pages are added to LRU.
1689 * No LRU may hold pages because all pages are UNEVICTABLE or
1690 * memcg is too small and all pages are not on LRU. In that case,
1691 * we use curret node.
1692 */
1693 if (unlikely(node == MAX_NUMNODES))
1694 node = numa_node_id();
1695
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001696 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001697 return node;
1698}
1699
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001700/*
1701 * Check all nodes whether it contains reclaimable pages or not.
1702 * For quick scan, we make use of scan_nodes. This will allow us to skip
1703 * unused nodes. But scan_nodes is lazily updated and may not cotain
1704 * enough new information. We need to do double check.
1705 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001706bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001707{
1708 int nid;
1709
1710 /*
1711 * quick check...making use of scan_node.
1712 * We can skip unused nodes.
1713 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001714 if (!nodes_empty(memcg->scan_nodes)) {
1715 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001716 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001717 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001718
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001719 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001720 return true;
1721 }
1722 }
1723 /*
1724 * Check rest of nodes.
1725 */
1726 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001727 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001728 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001729 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001730 return true;
1731 }
1732 return false;
1733}
1734
Ying Han889976d2011-05-26 16:25:33 -07001735#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001736int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001737{
1738 return 0;
1739}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001740
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001741bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001742{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001743 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001744}
Ying Han889976d2011-05-26 16:25:33 -07001745#endif
1746
Johannes Weiner56600482012-01-12 17:17:59 -08001747static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1748 struct zone *zone,
1749 gfp_t gfp_mask,
1750 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001751{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001752 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001753 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001754 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001755 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001756 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001757 struct mem_cgroup_reclaim_cookie reclaim = {
1758 .zone = zone,
1759 .priority = 0,
1760 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001761
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001762 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001763
Balbir Singh4e416952009-09-23 15:56:39 -07001764 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001765 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001766 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001767 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001768 if (loop >= 2) {
1769 /*
1770 * If we have not been able to reclaim
1771 * anything, it might because there are
1772 * no reclaimable pages under this hierarchy
1773 */
Johannes Weiner56600482012-01-12 17:17:59 -08001774 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001775 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001776 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001777 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001778 * excess >> 2 is not to excessive so as to
1779 * reclaim too much, nor too less that we keep
1780 * coming back to reclaim from this cgroup
1781 */
1782 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001783 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001784 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001785 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001786 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001787 }
Johannes Weiner56600482012-01-12 17:17:59 -08001788 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001789 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001790 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1791 zone, &nr_scanned);
1792 *total_scanned += nr_scanned;
1793 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001794 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001795 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001796 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001797 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001798}
1799
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001800/*
1801 * Check OOM-Killer is already running under our hierarchy.
1802 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001803 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001804 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001805static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001806{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001807 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001808
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001809 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001810 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001811 /*
1812 * this subtree of our hierarchy is already locked
1813 * so we cannot give a lock.
1814 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001815 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001816 mem_cgroup_iter_break(memcg, iter);
1817 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001818 } else
1819 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001820 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001821
Michal Hocko79dfdac2011-07-26 16:08:23 -07001822 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001823 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001824
1825 /*
1826 * OK, we failed to lock the whole subtree so we have to clean up
1827 * what we set up to the failing subtree
1828 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001829 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001830 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001831 mem_cgroup_iter_break(memcg, iter);
1832 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001833 }
1834 iter->oom_lock = false;
1835 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001836 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001837}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001838
Michal Hocko79dfdac2011-07-26 16:08:23 -07001839/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001840 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001841 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001842static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001843{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001844 struct mem_cgroup *iter;
1845
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001846 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001847 iter->oom_lock = false;
1848 return 0;
1849}
1850
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001851static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001852{
1853 struct mem_cgroup *iter;
1854
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001855 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001856 atomic_inc(&iter->under_oom);
1857}
1858
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001859static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001860{
1861 struct mem_cgroup *iter;
1862
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001863 /*
1864 * When a new child is created while the hierarchy is under oom,
1865 * mem_cgroup_oom_lock() may not be called. We have to use
1866 * atomic_add_unless() here.
1867 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001868 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001869 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001870}
1871
Michal Hocko1af8efe2011-07-26 16:08:24 -07001872static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001873static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1874
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001875struct oom_wait_info {
1876 struct mem_cgroup *mem;
1877 wait_queue_t wait;
1878};
1879
1880static int memcg_oom_wake_function(wait_queue_t *wait,
1881 unsigned mode, int sync, void *arg)
1882{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001883 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg,
1884 *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001885 struct oom_wait_info *oom_wait_info;
1886
1887 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001888 oom_wait_memcg = oom_wait_info->mem;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001889
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001890 /*
1891 * Both of oom_wait_info->mem and wake_mem are stable under us.
1892 * Then we can use css_is_ancestor without taking care of RCU.
1893 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001894 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1895 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001896 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001897 return autoremove_wake_function(wait, mode, sync, arg);
1898}
1899
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001900static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001901{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001902 /* for filtering, pass "memcg" as argument. */
1903 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001904}
1905
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001906static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001907{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001908 if (memcg && atomic_read(&memcg->under_oom))
1909 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001910}
1911
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001912/*
1913 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1914 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001915bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001916{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001917 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001918 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001919
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001920 owait.mem = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001921 owait.wait.flags = 0;
1922 owait.wait.func = memcg_oom_wake_function;
1923 owait.wait.private = current;
1924 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001925 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001926 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001927
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001928 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001929 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001930 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001931 /*
1932 * Even if signal_pending(), we can't quit charge() loop without
1933 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1934 * under OOM is always welcomed, use TASK_KILLABLE here.
1935 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001936 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001937 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001938 need_to_kill = false;
1939 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001940 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001941 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001942
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001943 if (need_to_kill) {
1944 finish_wait(&memcg_oom_waitq, &owait.wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001945 mem_cgroup_out_of_memory(memcg, mask);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001946 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001947 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001948 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001949 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001950 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001951 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001952 mem_cgroup_oom_unlock(memcg);
1953 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001954 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001955
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001956 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001957
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001958 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1959 return false;
1960 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001961 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001962 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001963}
1964
Balbir Singhd69b0422009-06-17 16:26:34 -07001965/*
1966 * Currently used to update mapped file statistics, but the routine can be
1967 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001968 *
1969 * Notes: Race condition
1970 *
1971 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1972 * it tends to be costly. But considering some conditions, we doesn't need
1973 * to do so _always_.
1974 *
1975 * Considering "charge", lock_page_cgroup() is not required because all
1976 * file-stat operations happen after a page is attached to radix-tree. There
1977 * are no race with "charge".
1978 *
1979 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1980 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1981 * if there are race with "uncharge". Statistics itself is properly handled
1982 * by flags.
1983 *
1984 * Considering "move", this is an only case we see a race. To make the race
1985 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1986 * possibility of race condition. If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001987 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001988
Greg Thelen2a7106f2011-01-13 15:47:37 -08001989void mem_cgroup_update_page_stat(struct page *page,
1990 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001991{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001992 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001993 struct page_cgroup *pc = lookup_page_cgroup(page);
1994 bool need_unlock = false;
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001995 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001996
Balbir Singhd69b0422009-06-17 16:26:34 -07001997 if (unlikely(!pc))
1998 return;
1999
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002000 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002001 memcg = pc->mem_cgroup;
2002 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002003 goto out;
2004 /* pc->mem_cgroup is unstable ? */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002005 if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) {
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002006 /* take a lock against to access pc->mem_cgroup */
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002007 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002008 need_unlock = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002009 memcg = pc->mem_cgroup;
2010 if (!memcg || !PageCgroupUsed(pc))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002011 goto out;
2012 }
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002013
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002014 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08002015 case MEMCG_NR_FILE_MAPPED:
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002016 if (val > 0)
2017 SetPageCgroupFileMapped(pc);
2018 else if (!page_mapped(page))
KAMEZAWA Hiroyuki0c270f82010-10-27 15:33:39 -07002019 ClearPageCgroupFileMapped(pc);
Greg Thelen2a7106f2011-01-13 15:47:37 -08002020 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002021 break;
2022 default:
2023 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002024 }
Balbir Singhd69b0422009-06-17 16:26:34 -07002025
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002026 this_cpu_add(memcg->stat->count[idx], val);
Greg Thelen2a7106f2011-01-13 15:47:37 -08002027
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002028out:
2029 if (unlikely(need_unlock))
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08002030 move_unlock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002031 rcu_read_unlock();
2032 return;
Balbir Singhd69b0422009-06-17 16:26:34 -07002033}
Greg Thelen2a7106f2011-01-13 15:47:37 -08002034EXPORT_SYMBOL(mem_cgroup_update_page_stat);
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07002035
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002036/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002037 * size of first charge trial. "32" comes from vmscan.c's magic value.
2038 * TODO: maybe necessary to use big numbers in big irons.
2039 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002040#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002041struct memcg_stock_pcp {
2042 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002043 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002044 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002045 unsigned long flags;
2046#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002047};
2048static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002049static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002050
2051/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002052 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002053 * from local stock and true is returned. If the stock is 0 or charges from a
2054 * cgroup which is not current target, returns false. This stock will be
2055 * refilled.
2056 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002057static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002058{
2059 struct memcg_stock_pcp *stock;
2060 bool ret = true;
2061
2062 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002063 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002064 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002065 else /* need to call res_counter_charge */
2066 ret = false;
2067 put_cpu_var(memcg_stock);
2068 return ret;
2069}
2070
2071/*
2072 * Returns stocks cached in percpu to res_counter and reset cached information.
2073 */
2074static void drain_stock(struct memcg_stock_pcp *stock)
2075{
2076 struct mem_cgroup *old = stock->cached;
2077
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002078 if (stock->nr_pages) {
2079 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2080
2081 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002082 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002083 res_counter_uncharge(&old->memsw, bytes);
2084 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002085 }
2086 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002087}
2088
2089/*
2090 * This must be called under preempt disabled or must be called by
2091 * a thread which is pinned to local cpu.
2092 */
2093static void drain_local_stock(struct work_struct *dummy)
2094{
2095 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2096 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002097 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002098}
2099
2100/*
2101 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002102 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002103 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002104static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002105{
2106 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2107
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002108 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002109 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002110 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002111 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002112 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002113 put_cpu_var(memcg_stock);
2114}
2115
2116/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002117 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07002118 * of the hierarchy under it. sync flag says whether we should block
2119 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002120 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002121static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002122{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002123 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002124
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002125 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002126 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07002127 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002128 for_each_online_cpu(cpu) {
2129 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002130 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002131
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002132 memcg = stock->cached;
2133 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002134 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002135 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002136 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002137 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2138 if (cpu == curcpu)
2139 drain_local_stock(&stock->work);
2140 else
2141 schedule_work_on(cpu, &stock->work);
2142 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002143 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002144 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002145
2146 if (!sync)
2147 goto out;
2148
2149 for_each_online_cpu(cpu) {
2150 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002151 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002152 flush_work(&stock->work);
2153 }
2154out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002155 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002156}
2157
2158/*
2159 * Tries to drain stocked charges in other cpus. This function is asynchronous
2160 * and just put a work per cpu for draining localy on each cpu. Caller can
2161 * expects some charges will be back to res_counter later but cannot wait for
2162 * it.
2163 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002164static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002165{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002166 /*
2167 * If someone calls draining, avoid adding more kworker runs.
2168 */
2169 if (!mutex_trylock(&percpu_charge_mutex))
2170 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002171 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002172 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002173}
2174
2175/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002176static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002177{
2178 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002179 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002180 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002181 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002182}
2183
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002184/*
2185 * This function drains percpu counter value from DEAD cpu and
2186 * move it to local cpu. Note that this function can be preempted.
2187 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002188static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002189{
2190 int i;
2191
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002192 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002193 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002194 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002195
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002196 per_cpu(memcg->stat->count[i], cpu) = 0;
2197 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002198 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002199 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002200 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002201
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002202 per_cpu(memcg->stat->events[i], cpu) = 0;
2203 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002204 }
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002205 /* need to clear ON_MOVE value, works as a kind of lock. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002206 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
2207 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002208}
2209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002210static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002211{
2212 int idx = MEM_CGROUP_ON_MOVE;
2213
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002214 spin_lock(&memcg->pcp_counter_lock);
2215 per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx];
2216 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002217}
2218
2219static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002220 unsigned long action,
2221 void *hcpu)
2222{
2223 int cpu = (unsigned long)hcpu;
2224 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002225 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002226
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002227 if ((action == CPU_ONLINE)) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002228 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002229 synchronize_mem_cgroup_on_move(iter, cpu);
2230 return NOTIFY_OK;
2231 }
2232
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002233 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002234 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002235
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002236 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002237 mem_cgroup_drain_pcp_counter(iter, cpu);
2238
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002239 stock = &per_cpu(memcg_stock, cpu);
2240 drain_stock(stock);
2241 return NOTIFY_OK;
2242}
2243
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002244
2245/* See __mem_cgroup_try_charge() for details */
2246enum {
2247 CHARGE_OK, /* success */
2248 CHARGE_RETRY, /* need to retry but retry is not bad */
2249 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2250 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2251 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2252};
2253
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002254static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002255 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002256{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002257 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002258 struct mem_cgroup *mem_over_limit;
2259 struct res_counter *fail_res;
2260 unsigned long flags = 0;
2261 int ret;
2262
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002263 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002264
2265 if (likely(!ret)) {
2266 if (!do_swap_account)
2267 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002268 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002269 if (likely(!ret))
2270 return CHARGE_OK;
2271
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002272 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002273 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2274 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2275 } else
2276 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002277 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002278 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2279 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002280 *
2281 * Never reclaim on behalf of optional batching, retry with a
2282 * single page instead.
2283 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002284 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002285 return CHARGE_RETRY;
2286
2287 if (!(gfp_mask & __GFP_WAIT))
2288 return CHARGE_WOULDBLOCK;
2289
Johannes Weiner56600482012-01-12 17:17:59 -08002290 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002291 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002292 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002293 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002294 * Even though the limit is exceeded at this point, reclaim
2295 * may have been able to free some pages. Retry the charge
2296 * before killing the task.
2297 *
2298 * Only for regular pages, though: huge pages are rather
2299 * unlikely to succeed so close to the limit, and we fall back
2300 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002301 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002302 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002303 return CHARGE_RETRY;
2304
2305 /*
2306 * At task move, charge accounts can be doubly counted. So, it's
2307 * better to wait until the end of task_move if something is going on.
2308 */
2309 if (mem_cgroup_wait_acct_move(mem_over_limit))
2310 return CHARGE_RETRY;
2311
2312 /* If we don't need to call oom-killer at el, return immediately */
2313 if (!oom_check)
2314 return CHARGE_NOMEM;
2315 /* check OOM */
2316 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
2317 return CHARGE_OOM_DIE;
2318
2319 return CHARGE_RETRY;
2320}
2321
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002322/*
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002323 * Unlike exported interface, "oom" parameter is added. if oom==true,
2324 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002325 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002326static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002327 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002328 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002329 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002330 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002331{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002332 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002333 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002334 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002335 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002336
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002337 /*
2338 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2339 * in system level. So, allow to go ahead dying process in addition to
2340 * MEMDIE process.
2341 */
2342 if (unlikely(test_thread_flag(TIF_MEMDIE)
2343 || fatal_signal_pending(current)))
2344 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002345
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002346 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002347 * We always charge the cgroup the mm_struct belongs to.
2348 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002349 * thread group leader migrates. It's possible that mm is not
2350 * set, if so charge the init_mm (happens for pagecache usage).
2351 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002352 if (!*ptr && !mm)
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002353 goto bypass;
2354again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002355 if (*ptr) { /* css should be a valid one */
2356 memcg = *ptr;
2357 VM_BUG_ON(css_is_removed(&memcg->css));
2358 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002359 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002360 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002361 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002362 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002363 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002364 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002365
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002366 rcu_read_lock();
2367 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002368 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002369 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002370 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002371 * race with swapoff. Then, we have small risk of mis-accouning.
2372 * But such kind of mis-account by race always happens because
2373 * we don't have cgroup_mutex(). It's overkill and we allo that
2374 * small race, here.
2375 * (*) swapoff at el will charge against mm-struct not against
2376 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002377 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002378 memcg = mem_cgroup_from_task(p);
2379 if (!memcg || mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002380 rcu_read_unlock();
2381 goto done;
2382 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002383 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002384 /*
2385 * It seems dagerous to access memcg without css_get().
2386 * But considering how consume_stok works, it's not
2387 * necessary. If consume_stock success, some charges
2388 * from this memcg are cached on this cpu. So, we
2389 * don't need to call css_get()/css_tryget() before
2390 * calling consume_stock().
2391 */
2392 rcu_read_unlock();
2393 goto done;
2394 }
2395 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002396 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002397 rcu_read_unlock();
2398 goto again;
2399 }
2400 rcu_read_unlock();
2401 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002402
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002403 do {
2404 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002405
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002406 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002407 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002408 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002409 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002410 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002411
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002412 oom_check = false;
2413 if (oom && !nr_oom_retries) {
2414 oom_check = true;
2415 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2416 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002417
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002418 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002419 switch (ret) {
2420 case CHARGE_OK:
2421 break;
2422 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002423 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002424 css_put(&memcg->css);
2425 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002426 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002427 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002428 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002429 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002430 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002431 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002432 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002433 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002434 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002435 /* If oom, we never return -ENOMEM */
2436 nr_oom_retries--;
2437 break;
2438 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002439 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002440 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002441 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002442 } while (ret != CHARGE_OK);
2443
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002444 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002445 refill_stock(memcg, batch - nr_pages);
2446 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002447done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002448 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002449 return 0;
2450nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002451 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002452 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002453bypass:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002454 *ptr = NULL;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002455 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002456}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002457
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002458/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002459 * Somemtimes we have to undo a charge we got by try_charge().
2460 * This function is for that and do uncharge, put css's refcnt.
2461 * gotten by try_charge().
2462 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002463static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b82011-03-23 16:42:33 -07002464 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002465{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002466 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b82011-03-23 16:42:33 -07002467 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002468
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002469 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002470 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002471 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002472 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002473}
2474
2475/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002476 * A helper function to get mem_cgroup from ID. must be called under
2477 * rcu_read_lock(). The caller must check css_is_removed() or some if
2478 * it's concern. (dropping refcnt from swap can be called against removed
2479 * memcg.)
2480 */
2481static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2482{
2483 struct cgroup_subsys_state *css;
2484
2485 /* ID 0 is unused ID */
2486 if (!id)
2487 return NULL;
2488 css = css_lookup(&mem_cgroup_subsys, id);
2489 if (!css)
2490 return NULL;
2491 return container_of(css, struct mem_cgroup, css);
2492}
2493
Wu Fengguange42d9d52009-12-16 12:19:59 +01002494struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002495{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002496 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002497 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002498 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002499 swp_entry_t ent;
2500
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002501 VM_BUG_ON(!PageLocked(page));
2502
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002503 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002504 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002505 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002506 memcg = pc->mem_cgroup;
2507 if (memcg && !css_tryget(&memcg->css))
2508 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002509 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002510 ent.val = page_private(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002511 id = lookup_swap_cgroup(ent);
2512 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002513 memcg = mem_cgroup_lookup(id);
2514 if (memcg && !css_tryget(&memcg->css))
2515 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002516 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002517 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002518 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002519 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002520}
2521
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002522static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002523 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002524 unsigned int nr_pages,
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002525 struct page_cgroup *pc,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002526 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002527{
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002528 lock_page_cgroup(pc);
2529 if (unlikely(PageCgroupUsed(pc))) {
2530 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002531 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002532 return;
2533 }
2534 /*
2535 * we don't need page_cgroup_lock about tail pages, becase they are not
2536 * accessed by any other context at this point.
2537 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002538 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002539 /*
2540 * We access a page_cgroup asynchronously without lock_page_cgroup().
2541 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2542 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2543 * before USED bit, we need memory barrier here.
2544 * See mem_cgroup_add_lru_list(), etc.
2545 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002546 smp_wmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -07002547 switch (ctype) {
2548 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2549 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2550 SetPageCgroupCache(pc);
2551 SetPageCgroupUsed(pc);
2552 break;
2553 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2554 ClearPageCgroupCache(pc);
2555 SetPageCgroupUsed(pc);
2556 break;
2557 default:
2558 break;
2559 }
Hugh Dickins3be91272008-02-07 00:14:19 -08002560
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002561 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002562 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002563 /*
2564 * "charge_statistics" updated event counter. Then, check it.
2565 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2566 * if they exceeds softlimit.
2567 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002568 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002569}
2570
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002571#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2572
2573#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2574 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2575/*
2576 * Because tail pages are not marked as "used", set it. We're under
2577 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2578 */
2579void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
2580{
2581 struct page_cgroup *head_pc = lookup_page_cgroup(head);
2582 struct page_cgroup *tail_pc = lookup_page_cgroup(tail);
2583 unsigned long flags;
2584
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002585 if (mem_cgroup_disabled())
2586 return;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002587 /*
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002588 * We have no races with charge/uncharge but will have races with
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002589 * page state accounting.
2590 */
2591 move_lock_page_cgroup(head_pc, &flags);
2592
2593 tail_pc->mem_cgroup = head_pc->mem_cgroup;
2594 smp_wmb(); /* see __commit_charge() */
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002595 if (PageCgroupAcctLRU(head_pc)) {
2596 enum lru_list lru;
2597 struct mem_cgroup_per_zone *mz;
2598
2599 /*
2600 * LRU flags cannot be copied because we need to add tail
2601 *.page to LRU by generic call and our hook will be called.
2602 * We hold lru_lock, then, reduce counter directly.
2603 */
2604 lru = page_lru(head);
Johannes Weiner97a6c372011-03-23 16:42:27 -07002605 mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head);
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08002606 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
2607 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002608 tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2609 move_unlock_page_cgroup(head_pc, &flags);
2610}
2611#endif
2612
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002613/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002614 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002615 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002616 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002617 * @pc: page_cgroup of the page.
2618 * @from: mem_cgroup which the page is moved from.
2619 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002620 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002621 *
2622 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002623 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002624 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002625 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002626 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002627 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002628 * true, this function does "uncharge" from old cgroup, but it doesn't if
2629 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002630 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002631static int mem_cgroup_move_account(struct page *page,
2632 unsigned int nr_pages,
2633 struct page_cgroup *pc,
2634 struct mem_cgroup *from,
2635 struct mem_cgroup *to,
2636 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002637{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002638 unsigned long flags;
2639 int ret;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002640
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002641 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002642 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002643 /*
2644 * The page is isolated from LRU. So, collapse function
2645 * will not handle this page. But page splitting can happen.
2646 * Do this check under compound_page_lock(). The caller should
2647 * hold it.
2648 */
2649 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002650 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002651 goto out;
2652
2653 lock_page_cgroup(pc);
2654
2655 ret = -EINVAL;
2656 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2657 goto unlock;
2658
2659 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002660
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002661 if (PageCgroupFileMapped(pc)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002662 /* Update mapped_file data for mem_cgroup */
2663 preempt_disable();
2664 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2665 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2666 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002667 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002668 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002669 if (uncharge)
2670 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b82011-03-23 16:42:33 -07002671 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002672
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002673 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002674 pc->mem_cgroup = to;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002675 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002676 /*
2677 * We charges against "to" which may not have any tasks. Then, "to"
2678 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002679 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002680 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002681 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002682 */
Johannes Weinerde3638d2011-03-23 16:42:28 -07002683 move_unlock_page_cgroup(pc, &flags);
2684 ret = 0;
2685unlock:
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002686 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002687 /*
2688 * check events
2689 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002690 memcg_check_events(to, page);
2691 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002692out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002693 return ret;
2694}
2695
2696/*
2697 * move charges to its parent.
2698 */
2699
Johannes Weiner5564e882011-03-23 16:42:29 -07002700static int mem_cgroup_move_parent(struct page *page,
2701 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002702 struct mem_cgroup *child,
2703 gfp_t gfp_mask)
2704{
2705 struct cgroup *cg = child->css.cgroup;
2706 struct cgroup *pcg = cg->parent;
2707 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002708 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002709 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002710 int ret;
2711
2712 /* Is ROOT ? */
2713 if (!pcg)
2714 return -EINVAL;
2715
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002716 ret = -EBUSY;
2717 if (!get_page_unless_zero(page))
2718 goto out;
2719 if (isolate_lru_page(page))
2720 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002721
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002722 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002723
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002724 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002725 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002726 if (ret || !parent)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002727 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002728
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002729 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002730 flags = compound_lock_irqsave(page);
2731
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002732 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002733 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002734 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002735
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002736 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002737 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002738put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002739 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002740put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002741 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002742out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002743 return ret;
2744}
2745
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002746/*
2747 * Charge the memory controller for page usage.
2748 * Return
2749 * 0 if the charge was successful
2750 * < 0 if the cgroup is over its limit
2751 */
2752static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002753 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002754{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002755 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002756 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002757 struct page_cgroup *pc;
2758 bool oom = true;
2759 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002760
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002761 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002762 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002763 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002764 /*
2765 * Never OOM-kill a process for a huge page. The
2766 * fault handler will fall back to regular pages.
2767 */
2768 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002769 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002770
2771 pc = lookup_page_cgroup(page);
Johannes Weineraf4a6622011-03-23 16:42:24 -07002772 BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002773
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002774 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
2775 if (ret || !memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002776 return ret;
2777
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002778 __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002779 return 0;
2780}
2781
2782int mem_cgroup_newpage_charge(struct page *page,
2783 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002784{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002785 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002786 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002787 /*
2788 * If already mapped, we don't have to account.
2789 * If page cache, page->mapping has address_space.
2790 * But page->mapping may have out-of-use anon_vma pointer,
2791 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2792 * is NULL.
2793 */
2794 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2795 return 0;
2796 if (unlikely(!mm))
2797 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002798 return mem_cgroup_charge_common(page, mm, gfp_mask,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002799 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002800}
2801
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002802static void
2803__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2804 enum charge_type ctype);
2805
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002806static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002807__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002808 enum charge_type ctype)
2809{
2810 struct page_cgroup *pc = lookup_page_cgroup(page);
2811 /*
2812 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2813 * is already on LRU. It means the page may on some other page_cgroup's
2814 * LRU. Take care of it.
2815 */
2816 mem_cgroup_lru_del_before_commit(page);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002817 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002818 mem_cgroup_lru_add_after_commit(page);
2819 return;
2820}
2821
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002822int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2823 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002824{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002825 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002826 int ret;
2827
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002828 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002829 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002830 if (PageCompound(page))
2831 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002832
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002833 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002834 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002835
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002836 if (page_is_file_cache(page)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002837 ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &memcg, true);
2838 if (ret || !memcg)
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002839 return ret;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002840
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002841 /*
2842 * FUSE reuses pages without going through the final
2843 * put that would remove them from the LRU list, make
2844 * sure that they get relinked properly.
2845 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002846 __mem_cgroup_commit_charge_lrucare(page, memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002847 MEM_CGROUP_CHARGE_TYPE_CACHE);
2848 return ret;
2849 }
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002850 /* shmem */
2851 if (PageSwapCache(page)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002852 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002853 if (!ret)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002854 __mem_cgroup_commit_charge_swapin(page, memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002855 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2856 } else
2857 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002858 MEM_CGROUP_CHARGE_TYPE_SHMEM);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002859
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002860 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002861}
2862
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002863/*
2864 * While swap-in, try_charge -> commit or cancel, the page is locked.
2865 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002866 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002867 * "commit()" or removed by "cancel()"
2868 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002869int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2870 struct page *page,
2871 gfp_t mask, struct mem_cgroup **ptr)
2872{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002873 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002874 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002875
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002876 *ptr = NULL;
2877
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002878 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002879 return 0;
2880
2881 if (!do_swap_account)
2882 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002883 /*
2884 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002885 * the pte, and even removed page from swap cache: in those cases
2886 * do_swap_page()'s pte_same() test will fail; but there's also a
2887 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002888 */
2889 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002890 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002891 memcg = try_get_mem_cgroup_from_page(page);
2892 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002893 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002894 *ptr = memcg;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002895 ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002896 css_put(&memcg->css);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002897 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002898charge_cur_mm:
2899 if (unlikely(!mm))
2900 mm = &init_mm;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002901 return __mem_cgroup_try_charge(mm, mask, 1, ptr, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002902}
2903
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002904static void
2905__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2906 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002907{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002908 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002909 return;
2910 if (!ptr)
2911 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002912 cgroup_exclude_rmdir(&ptr->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002913
2914 __mem_cgroup_commit_charge_lrucare(page, ptr, ctype);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002915 /*
2916 * Now swap is on-memory. This means this page may be
2917 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002918 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2919 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2920 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002921 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002922 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002923 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002924 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002925 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002926
2927 id = swap_cgroup_record(ent, 0);
2928 rcu_read_lock();
2929 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002930 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002931 /*
2932 * This recorded memcg can be obsolete one. So, avoid
2933 * calling css_tryget
2934 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002935 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07002936 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002937 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002938 mem_cgroup_put(memcg);
2939 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002940 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002941 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002942 /*
2943 * At swapin, we may charge account against cgroup which has no tasks.
2944 * So, rmdir()->pre_destroy() can be called while we do this charge.
2945 * In that case, we need to call pre_destroy() again. check it here.
2946 */
2947 cgroup_release_and_wakeup_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002948}
2949
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002950void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2951{
2952 __mem_cgroup_commit_charge_swapin(page, ptr,
2953 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2954}
2955
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002956void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002957{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002958 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002959 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002960 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002961 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002962 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002963}
2964
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002965static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002966 unsigned int nr_pages,
2967 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002968{
2969 struct memcg_batch_info *batch = NULL;
2970 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002971
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002972 /* If swapout, usage of swap doesn't decrease */
2973 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2974 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002975
2976 batch = &current->memcg_batch;
2977 /*
2978 * In usual, we do css_get() when we remember memcg pointer.
2979 * But in this case, we keep res->usage until end of a series of
2980 * uncharges. Then, it's ok to ignore memcg's refcnt.
2981 */
2982 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002983 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002984 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002985 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002986 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002987 * the same cgroup and we have chance to coalesce uncharges.
2988 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2989 * because we want to do uncharge as soon as possible.
2990 */
2991
2992 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2993 goto direct_uncharge;
2994
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002995 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002996 goto direct_uncharge;
2997
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002998 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002999 * In typical case, batch->memcg == mem. This means we can
3000 * merge a series of uncharges to an uncharge of res_counter.
3001 * If not, we uncharge res_counter ony by one.
3002 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003003 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003004 goto direct_uncharge;
3005 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003006 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003007 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003008 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003009 return;
3010direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003011 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003012 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003013 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
3014 if (unlikely(batch->memcg != memcg))
3015 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003016 return;
3017}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003018
Balbir Singh8697d332008-02-07 00:13:59 -08003019/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003020 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003021 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003022static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003023__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003024{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003025 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003026 unsigned int nr_pages = 1;
3027 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003028
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003029 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003030 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07003031
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003032 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003033 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003034
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003035 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003036 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08003037 VM_BUG_ON(!PageTransHuge(page));
3038 }
Balbir Singh8697d332008-02-07 00:13:59 -08003039 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08003040 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08003041 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003042 pc = lookup_page_cgroup(page);
3043 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003044 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003045
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003046 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003047
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003048 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003049
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003050 if (!PageCgroupUsed(pc))
3051 goto unlock_out;
3052
3053 switch (ctype) {
3054 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003055 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003056 /* See mem_cgroup_prepare_migration() */
3057 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003058 goto unlock_out;
3059 break;
3060 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3061 if (!PageAnon(page)) { /* Shared memory */
3062 if (page->mapping && !page_is_file_cache(page))
3063 goto unlock_out;
3064 } else if (page_mapped(page)) /* Anon */
3065 goto unlock_out;
3066 break;
3067 default:
3068 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003069 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003070
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003071 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003072
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003073 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08003074 /*
3075 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3076 * freed from LRU. This is safe because uncharged page is expected not
3077 * to be reused (freed soon). Exception is SwapCache, it's handled by
3078 * special functions.
3079 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003080
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003081 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003082 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003083 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003084 * will never be freed.
3085 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003086 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003087 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003088 mem_cgroup_swap_statistics(memcg, true);
3089 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003090 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003091 if (!mem_cgroup_is_root(memcg))
3092 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003093
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003094 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003095
3096unlock_out:
3097 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003098 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08003099}
3100
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003101void mem_cgroup_uncharge_page(struct page *page)
3102{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003103 /* early check. */
3104 if (page_mapped(page))
3105 return;
3106 if (page->mapping && !PageAnon(page))
3107 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003108 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3109}
3110
3111void mem_cgroup_uncharge_cache_page(struct page *page)
3112{
3113 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07003114 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003115 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3116}
3117
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003118/*
3119 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3120 * In that cases, pages are freed continuously and we can expect pages
3121 * are in the same memcg. All these calls itself limits the number of
3122 * pages freed at once, then uncharge_start/end() is called properly.
3123 * This may be called prural(2) times in a context,
3124 */
3125
3126void mem_cgroup_uncharge_start(void)
3127{
3128 current->memcg_batch.do_batch++;
3129 /* We can do nest. */
3130 if (current->memcg_batch.do_batch == 1) {
3131 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003132 current->memcg_batch.nr_pages = 0;
3133 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003134 }
3135}
3136
3137void mem_cgroup_uncharge_end(void)
3138{
3139 struct memcg_batch_info *batch = &current->memcg_batch;
3140
3141 if (!batch->do_batch)
3142 return;
3143
3144 batch->do_batch--;
3145 if (batch->do_batch) /* If stacked, do nothing. */
3146 return;
3147
3148 if (!batch->memcg)
3149 return;
3150 /*
3151 * This "batch->memcg" is valid without any css_get/put etc...
3152 * bacause we hide charges behind us.
3153 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003154 if (batch->nr_pages)
3155 res_counter_uncharge(&batch->memcg->res,
3156 batch->nr_pages * PAGE_SIZE);
3157 if (batch->memsw_nr_pages)
3158 res_counter_uncharge(&batch->memcg->memsw,
3159 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003160 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003161 /* forget this pointer (for sanity check) */
3162 batch->memcg = NULL;
3163}
3164
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003165#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003166/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003167 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003168 * memcg information is recorded to swap_cgroup of "ent"
3169 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003170void
3171mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003172{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003173 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003174 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003175
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003176 if (!swapout) /* this was a swap cache but the swap is unused ! */
3177 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3178
3179 memcg = __mem_cgroup_uncharge_common(page, ctype);
3180
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003181 /*
3182 * record memcg information, if swapout && memcg != NULL,
3183 * mem_cgroup_get() was called in uncharge().
3184 */
3185 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003186 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003187}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003188#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003189
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003190#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3191/*
3192 * called from swap_entry_free(). remove record in swap_cgroup and
3193 * uncharge "memsw" account.
3194 */
3195void mem_cgroup_uncharge_swap(swp_entry_t ent)
3196{
3197 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003198 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003199
3200 if (!do_swap_account)
3201 return;
3202
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003203 id = swap_cgroup_record(ent, 0);
3204 rcu_read_lock();
3205 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003206 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003207 /*
3208 * We uncharge this because swap is freed.
3209 * This memcg can be obsolete one. We avoid calling css_tryget
3210 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003211 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003212 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003213 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003214 mem_cgroup_put(memcg);
3215 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003216 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003217}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003218
3219/**
3220 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3221 * @entry: swap entry to be moved
3222 * @from: mem_cgroup which the entry is moved from
3223 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003224 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003225 *
3226 * It succeeds only when the swap_cgroup's record for this entry is the same
3227 * as the mem_cgroup's id of @from.
3228 *
3229 * Returns 0 on success, -EINVAL on failure.
3230 *
3231 * The caller must have charged to @to, IOW, called res_counter_charge() about
3232 * both res and memsw, and called css_get().
3233 */
3234static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003235 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003236{
3237 unsigned short old_id, new_id;
3238
3239 old_id = css_id(&from->css);
3240 new_id = css_id(&to->css);
3241
3242 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003243 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003244 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003245 /*
3246 * This function is only called from task migration context now.
3247 * It postpones res_counter and refcount handling till the end
3248 * of task migration(mem_cgroup_clear_mc()) for performance
3249 * improvement. But we cannot postpone mem_cgroup_get(to)
3250 * because if the process that has been moved to @to does
3251 * swap-in, the refcount of @to might be decreased to 0.
3252 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003253 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003254 if (need_fixup) {
3255 if (!mem_cgroup_is_root(from))
3256 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3257 mem_cgroup_put(from);
3258 /*
3259 * we charged both to->res and to->memsw, so we should
3260 * uncharge to->res.
3261 */
3262 if (!mem_cgroup_is_root(to))
3263 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003264 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003265 return 0;
3266 }
3267 return -EINVAL;
3268}
3269#else
3270static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003271 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003272{
3273 return -EINVAL;
3274}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003275#endif
3276
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003277/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003278 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3279 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003280 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003281int mem_cgroup_prepare_migration(struct page *page,
Miklos Szeredief6a3c62011-03-22 16:30:52 -07003282 struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003283{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003284 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003285 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003286 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003287 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003288
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003289 *ptr = NULL;
3290
Andrea Arcangeliec168512011-01-13 15:46:56 -08003291 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003292 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003293 return 0;
3294
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003295 pc = lookup_page_cgroup(page);
3296 lock_page_cgroup(pc);
3297 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003298 memcg = pc->mem_cgroup;
3299 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003300 /*
3301 * At migrating an anonymous page, its mapcount goes down
3302 * to 0 and uncharge() will be called. But, even if it's fully
3303 * unmapped, migration may fail and this page has to be
3304 * charged again. We set MIGRATION flag here and delay uncharge
3305 * until end_migration() is called
3306 *
3307 * Corner Case Thinking
3308 * A)
3309 * When the old page was mapped as Anon and it's unmap-and-freed
3310 * while migration was ongoing.
3311 * If unmap finds the old page, uncharge() of it will be delayed
3312 * until end_migration(). If unmap finds a new page, it's
3313 * uncharged when it make mapcount to be 1->0. If unmap code
3314 * finds swap_migration_entry, the new page will not be mapped
3315 * and end_migration() will find it(mapcount==0).
3316 *
3317 * B)
3318 * When the old page was mapped but migraion fails, the kernel
3319 * remaps it. A charge for it is kept by MIGRATION flag even
3320 * if mapcount goes down to 0. We can do remap successfully
3321 * without charging it again.
3322 *
3323 * C)
3324 * The "old" page is under lock_page() until the end of
3325 * migration, so, the old page itself will not be swapped-out.
3326 * If the new page is swapped out before end_migraton, our
3327 * hook to usual swap-out path will catch the event.
3328 */
3329 if (PageAnon(page))
3330 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003331 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003332 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003333 /*
3334 * If the page is not charged at this point,
3335 * we return here.
3336 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003337 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003338 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003339
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003340 *ptr = memcg;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003341 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003342 css_put(&memcg->css);/* drop extra refcnt */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003343 if (ret || *ptr == NULL) {
3344 if (PageAnon(page)) {
3345 lock_page_cgroup(pc);
3346 ClearPageCgroupMigration(pc);
3347 unlock_page_cgroup(pc);
3348 /*
3349 * The old page may be fully unmapped while we kept it.
3350 */
3351 mem_cgroup_uncharge_page(page);
3352 }
3353 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003354 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003355 /*
3356 * We charge new page before it's used/mapped. So, even if unlock_page()
3357 * is called before end_migration, we can catch all events on this new
3358 * page. In the case new page is migrated but not remapped, new page's
3359 * mapcount will be finally 0 and we call uncharge in end_migration().
3360 */
3361 pc = lookup_page_cgroup(newpage);
3362 if (PageAnon(page))
3363 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3364 else if (page_is_file_cache(page))
3365 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3366 else
3367 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003368 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003369 return ret;
3370}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003371
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003372/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003373void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003374 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003375{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003376 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003377 struct page_cgroup *pc;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003378
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003379 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003380 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003381 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003382 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003383 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003384 used = oldpage;
3385 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003386 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003387 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003388 unused = oldpage;
3389 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003390 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003391 * We disallowed uncharge of pages under migration because mapcount
3392 * of the page goes down to zero, temporarly.
3393 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003394 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003395 pc = lookup_page_cgroup(oldpage);
3396 lock_page_cgroup(pc);
3397 ClearPageCgroupMigration(pc);
3398 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003399
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003400 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3401
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003402 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003403 * If a page is a file cache, radix-tree replacement is very atomic
3404 * and we can skip this check. When it was an Anon page, its mapcount
3405 * goes down to 0. But because we added MIGRATION flage, it's not
3406 * uncharged yet. There are several case but page->mapcount check
3407 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3408 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003409 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003410 if (PageAnon(used))
3411 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003412 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003413 * At migration, we may charge account against cgroup which has no
3414 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003415 * So, rmdir()->pre_destroy() can be called while we do this charge.
3416 * In that case, we need to call pre_destroy() again. check it here.
3417 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003418 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003419}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003420
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003421/*
3422 * At replace page cache, newpage is not under any memcg but it's on
3423 * LRU. So, this function doesn't touch res_counter but handles LRU
3424 * in correct way. Both pages are locked so we cannot race with uncharge.
3425 */
3426void mem_cgroup_replace_page_cache(struct page *oldpage,
3427 struct page *newpage)
3428{
3429 struct mem_cgroup *memcg;
3430 struct page_cgroup *pc;
3431 struct zone *zone;
3432 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
3433 unsigned long flags;
3434
3435 if (mem_cgroup_disabled())
3436 return;
3437
3438 pc = lookup_page_cgroup(oldpage);
3439 /* fix accounting on old pages */
3440 lock_page_cgroup(pc);
3441 memcg = pc->mem_cgroup;
3442 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -1);
3443 ClearPageCgroupUsed(pc);
3444 unlock_page_cgroup(pc);
3445
3446 if (PageSwapBacked(oldpage))
3447 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3448
3449 zone = page_zone(newpage);
3450 pc = lookup_page_cgroup(newpage);
3451 /*
3452 * Even if newpage->mapping was NULL before starting replacement,
3453 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3454 * LRU while we overwrite pc->mem_cgroup.
3455 */
3456 spin_lock_irqsave(&zone->lru_lock, flags);
3457 if (PageLRU(newpage))
3458 del_page_from_lru_list(zone, newpage, page_lru(newpage));
3459 __mem_cgroup_commit_charge(memcg, newpage, 1, pc, type);
3460 if (PageLRU(newpage))
3461 add_page_to_lru_list(zone, newpage, page_lru(newpage));
3462 spin_unlock_irqrestore(&zone->lru_lock, flags);
3463}
3464
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003465#ifdef CONFIG_DEBUG_VM
3466static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3467{
3468 struct page_cgroup *pc;
3469
3470 pc = lookup_page_cgroup(page);
3471 if (likely(pc) && PageCgroupUsed(pc))
3472 return pc;
3473 return NULL;
3474}
3475
3476bool mem_cgroup_bad_page_check(struct page *page)
3477{
3478 if (mem_cgroup_disabled())
3479 return false;
3480
3481 return lookup_page_cgroup_used(page) != NULL;
3482}
3483
3484void mem_cgroup_print_bad_page(struct page *page)
3485{
3486 struct page_cgroup *pc;
3487
3488 pc = lookup_page_cgroup_used(page);
3489 if (pc) {
3490 int ret = -1;
3491 char *path;
3492
3493 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3494 pc, pc->flags, pc->mem_cgroup);
3495
3496 path = kmalloc(PATH_MAX, GFP_KERNEL);
3497 if (path) {
3498 rcu_read_lock();
3499 ret = cgroup_path(pc->mem_cgroup->css.cgroup,
3500 path, PATH_MAX);
3501 rcu_read_unlock();
3502 }
3503
3504 printk(KERN_CONT "(%s)\n",
3505 (ret < 0) ? "cannot get the path" : path);
3506 kfree(path);
3507 }
3508}
3509#endif
3510
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003511static DEFINE_MUTEX(set_limit_mutex);
3512
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003513static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003514 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003515{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003516 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003517 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003518 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003519 int children = mem_cgroup_count_children(memcg);
3520 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003521 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003522
3523 /*
3524 * For keeping hierarchical_reclaim simple, how long we should retry
3525 * is depends on callers. We set our retry-count to be function
3526 * of # of children which we should visit in this loop.
3527 */
3528 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3529
3530 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003531
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003532 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003533 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003534 if (signal_pending(current)) {
3535 ret = -EINTR;
3536 break;
3537 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003538 /*
3539 * Rather than hide all in some function, I do this in
3540 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003541 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003542 */
3543 mutex_lock(&set_limit_mutex);
3544 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3545 if (memswlimit < val) {
3546 ret = -EINVAL;
3547 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003548 break;
3549 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003550
3551 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3552 if (memlimit < val)
3553 enlarge = 1;
3554
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003555 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003556 if (!ret) {
3557 if (memswlimit == val)
3558 memcg->memsw_is_minimum = true;
3559 else
3560 memcg->memsw_is_minimum = false;
3561 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003562 mutex_unlock(&set_limit_mutex);
3563
3564 if (!ret)
3565 break;
3566
Johannes Weiner56600482012-01-12 17:17:59 -08003567 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3568 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003569 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3570 /* Usage is reduced ? */
3571 if (curusage >= oldusage)
3572 retry_count--;
3573 else
3574 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003575 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003576 if (!ret && enlarge)
3577 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003578
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003579 return ret;
3580}
3581
Li Zefan338c8432009-06-17 16:27:15 -07003582static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3583 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003584{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003585 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003586 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003587 int children = mem_cgroup_count_children(memcg);
3588 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003590
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003591 /* see mem_cgroup_resize_res_limit */
3592 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3593 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003594 while (retry_count) {
3595 if (signal_pending(current)) {
3596 ret = -EINTR;
3597 break;
3598 }
3599 /*
3600 * Rather than hide all in some function, I do this in
3601 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003602 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003603 */
3604 mutex_lock(&set_limit_mutex);
3605 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3606 if (memlimit > val) {
3607 ret = -EINVAL;
3608 mutex_unlock(&set_limit_mutex);
3609 break;
3610 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003611 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3612 if (memswlimit < val)
3613 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003614 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003615 if (!ret) {
3616 if (memlimit == val)
3617 memcg->memsw_is_minimum = true;
3618 else
3619 memcg->memsw_is_minimum = false;
3620 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003621 mutex_unlock(&set_limit_mutex);
3622
3623 if (!ret)
3624 break;
3625
Johannes Weiner56600482012-01-12 17:17:59 -08003626 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3627 MEM_CGROUP_RECLAIM_NOSWAP |
3628 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003629 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003630 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003631 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003632 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003633 else
3634 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003635 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003636 if (!ret && enlarge)
3637 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003638 return ret;
3639}
3640
Balbir Singh4e416952009-09-23 15:56:39 -07003641unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003642 gfp_t gfp_mask,
3643 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003644{
3645 unsigned long nr_reclaimed = 0;
3646 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3647 unsigned long reclaimed;
3648 int loop = 0;
3649 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003650 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003651 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003652
3653 if (order > 0)
3654 return 0;
3655
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003656 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003657 /*
3658 * This loop can run a while, specially if mem_cgroup's continuously
3659 * keep exceeding their soft limit and putting the system under
3660 * pressure
3661 */
3662 do {
3663 if (next_mz)
3664 mz = next_mz;
3665 else
3666 mz = mem_cgroup_largest_soft_limit_node(mctz);
3667 if (!mz)
3668 break;
3669
Ying Han0ae5e892011-05-26 16:25:25 -07003670 nr_scanned = 0;
Johannes Weiner56600482012-01-12 17:17:59 -08003671 reclaimed = mem_cgroup_soft_reclaim(mz->mem, zone,
3672 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003673 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003674 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003675 spin_lock(&mctz->lock);
3676
3677 /*
3678 * If we failed to reclaim anything from this memory cgroup
3679 * it is time to move on to the next cgroup
3680 */
3681 next_mz = NULL;
3682 if (!reclaimed) {
3683 do {
3684 /*
3685 * Loop until we find yet another one.
3686 *
3687 * By the time we get the soft_limit lock
3688 * again, someone might have aded the
3689 * group back on the RB tree. Iterate to
3690 * make sure we get a different mem.
3691 * mem_cgroup_largest_soft_limit_node returns
3692 * NULL if no other cgroup is present on
3693 * the tree
3694 */
3695 next_mz =
3696 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003697 if (next_mz == mz)
Balbir Singh4e416952009-09-23 15:56:39 -07003698 css_put(&next_mz->mem->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003699 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003700 break;
3701 } while (1);
3702 }
Balbir Singh4e416952009-09-23 15:56:39 -07003703 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003704 excess = res_counter_soft_limit_excess(&mz->mem->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003705 /*
3706 * One school of thought says that we should not add
3707 * back the node to the tree if reclaim returns 0.
3708 * But our reclaim could return 0, simply because due
3709 * to priority we are exposing a smaller subset of
3710 * memory to reclaim from. Consider this as a longer
3711 * term TODO.
3712 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003713 /* If excess == 0, no tree ops */
3714 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003715 spin_unlock(&mctz->lock);
3716 css_put(&mz->mem->css);
3717 loop++;
3718 /*
3719 * Could not reclaim anything and there are no more
3720 * mem cgroups to try or we seem to be looping without
3721 * reclaiming anything.
3722 */
3723 if (!nr_reclaimed &&
3724 (next_mz == NULL ||
3725 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3726 break;
3727 } while (!nr_reclaimed);
3728 if (next_mz)
3729 css_put(&next_mz->mem->css);
3730 return nr_reclaimed;
3731}
3732
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003733/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003734 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003735 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3736 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003737static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003738 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003739{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003740 struct zone *zone;
3741 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003742 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003743 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003744 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003745 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003746
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003747 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003748 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07003749 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003750
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003751 loop = MEM_CGROUP_ZSTAT(mz, lru);
3752 /* give some margin against EBUSY etc...*/
3753 loop += 256;
3754 busy = NULL;
3755 while (loop--) {
Johannes Weiner5564e882011-03-23 16:42:29 -07003756 struct page *page;
3757
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003758 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003759 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003760 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003761 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003762 break;
3763 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003764 pc = list_entry(list->prev, struct page_cgroup, lru);
3765 if (busy == pc) {
3766 list_move(&pc->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003767 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003768 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003769 continue;
3770 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003771 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003772
Johannes Weiner6b3ae582011-03-23 16:42:30 -07003773 page = lookup_cgroup_page(pc);
Johannes Weiner5564e882011-03-23 16:42:29 -07003774
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003775 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003776 if (ret == -ENOMEM)
3777 break;
3778
3779 if (ret == -EBUSY || ret == -EINVAL) {
3780 /* found lock contention or "pc" is obsolete. */
3781 busy = pc;
3782 cond_resched();
3783 } else
3784 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003785 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003786
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003787 if (!ret && !list_empty(list))
3788 return -EBUSY;
3789 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003790}
3791
3792/*
3793 * make mem_cgroup's charge to be 0 if there is no task.
3794 * This enables deleting this mem_cgroup.
3795 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003796static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003797{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003798 int ret;
3799 int node, zid, shrink;
3800 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003801 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003802
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003803 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003804
3805 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003806 /* should free all ? */
3807 if (free_all)
3808 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003809move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003810 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003811 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003812 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003813 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003814 ret = -EINTR;
3815 if (signal_pending(current))
3816 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003817 /* This is for making all *used* pages to be on LRU. */
3818 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003819 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003820 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003821 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003822 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003823 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07003824 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003825 for_each_lru(l) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003826 ret = mem_cgroup_force_empty_list(memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003827 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003828 if (ret)
3829 break;
3830 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003831 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003832 if (ret)
3833 break;
3834 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003835 mem_cgroup_end_move(memcg);
3836 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003837 /* it seems parent cgroup doesn't have enough mem */
3838 if (ret == -ENOMEM)
3839 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003840 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003841 /* "ret" should also be checked to ensure all lists are empty. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003842 } while (memcg->res.usage > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003843out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003844 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003845 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003846
3847try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003848 /* returns EBUSY if there is a task or if we come here twice. */
3849 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003850 ret = -EBUSY;
3851 goto out;
3852 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003853 /* we call try-to-free pages for make this cgroup empty */
3854 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003855 /* try to free all pages in this cgroup */
3856 shrink = 1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003857 while (nr_retries && memcg->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003858 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003859
3860 if (signal_pending(current)) {
3861 ret = -EINTR;
3862 goto out;
3863 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003864 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003865 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003866 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003867 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003868 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003869 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003870 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003871
3872 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003873 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003874 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003875 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003876}
3877
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003878int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3879{
3880 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3881}
3882
3883
Balbir Singh18f59ea2009-01-07 18:08:07 -08003884static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3885{
3886 return mem_cgroup_from_cont(cont)->use_hierarchy;
3887}
3888
3889static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3890 u64 val)
3891{
3892 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003893 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003894 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003895 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003896
3897 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003898 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003899
3900 cgroup_lock();
3901 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003902 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003903 * in the child subtrees. If it is unset, then the change can
3904 * occur, provided the current cgroup has no children.
3905 *
3906 * For the root cgroup, parent_mem is NULL, we allow value to be
3907 * set if there are no children.
3908 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003909 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003910 (val == 1 || val == 0)) {
3911 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003912 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003913 else
3914 retval = -EBUSY;
3915 } else
3916 retval = -EINVAL;
3917 cgroup_unlock();
3918
3919 return retval;
3920}
3921
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003922
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003923static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003924 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003925{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003926 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003927 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003928
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003929 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003930 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003931 val += mem_cgroup_read_stat(iter, idx);
3932
3933 if (val < 0) /* race ? */
3934 val = 0;
3935 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003936}
3937
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003938static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003939{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003940 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003941
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003942 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003943 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003944 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003945 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003946 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003947 }
3948
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003949 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3950 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003951
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003952 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003953 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003954
3955 return val << PAGE_SHIFT;
3956}
3957
Paul Menage2c3daa72008-04-29 00:59:58 -07003958static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003959{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003960 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003961 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003962 int type, name;
3963
3964 type = MEMFILE_TYPE(cft->private);
3965 name = MEMFILE_ATTR(cft->private);
3966 switch (type) {
3967 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003968 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003969 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003970 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003971 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003972 break;
3973 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003974 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003975 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003976 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003977 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003978 break;
3979 default:
3980 BUG();
3981 break;
3982 }
3983 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003984}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003985/*
3986 * The user of this function is...
3987 * RES_LIMIT.
3988 */
Paul Menage856c13a2008-07-25 01:47:04 -07003989static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3990 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003991{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003992 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003993 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003994 unsigned long long val;
3995 int ret;
3996
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003997 type = MEMFILE_TYPE(cft->private);
3998 name = MEMFILE_ATTR(cft->private);
3999 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004000 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07004001 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
4002 ret = -EINVAL;
4003 break;
4004 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004005 /* This function does all necessary parse...reuse it */
4006 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004007 if (ret)
4008 break;
4009 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004010 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004011 else
4012 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004013 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07004014 case RES_SOFT_LIMIT:
4015 ret = res_counter_memparse_write_strategy(buffer, &val);
4016 if (ret)
4017 break;
4018 /*
4019 * For memsw, soft limits are hard to implement in terms
4020 * of semantics, for now, we support soft limits for
4021 * control without swap
4022 */
4023 if (type == _MEM)
4024 ret = res_counter_set_soft_limit(&memcg->res, val);
4025 else
4026 ret = -EINVAL;
4027 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004028 default:
4029 ret = -EINVAL; /* should be BUG() ? */
4030 break;
4031 }
4032 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004033}
4034
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004035static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
4036 unsigned long long *mem_limit, unsigned long long *memsw_limit)
4037{
4038 struct cgroup *cgroup;
4039 unsigned long long min_limit, min_memsw_limit, tmp;
4040
4041 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4042 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4043 cgroup = memcg->css.cgroup;
4044 if (!memcg->use_hierarchy)
4045 goto out;
4046
4047 while (cgroup->parent) {
4048 cgroup = cgroup->parent;
4049 memcg = mem_cgroup_from_cont(cgroup);
4050 if (!memcg->use_hierarchy)
4051 break;
4052 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4053 min_limit = min(min_limit, tmp);
4054 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4055 min_memsw_limit = min(min_memsw_limit, tmp);
4056 }
4057out:
4058 *mem_limit = min_limit;
4059 *memsw_limit = min_memsw_limit;
4060 return;
4061}
4062
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004063static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004064{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004065 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004066 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004067
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004068 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004069 type = MEMFILE_TYPE(event);
4070 name = MEMFILE_ATTR(event);
4071 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004072 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004073 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004074 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004075 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004076 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004077 break;
4078 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004079 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004080 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004081 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004082 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004083 break;
4084 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004085
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07004086 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004087}
4088
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004089static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4090 struct cftype *cft)
4091{
4092 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4093}
4094
Daisuke Nishimura02491442010-03-10 15:22:17 -08004095#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004096static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4097 struct cftype *cft, u64 val)
4098{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004099 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004100
4101 if (val >= (1 << NR_MOVE_TYPE))
4102 return -EINVAL;
4103 /*
4104 * We check this value several times in both in can_attach() and
4105 * attach(), so we need cgroup lock to prevent this value from being
4106 * inconsistent.
4107 */
4108 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004109 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004110 cgroup_unlock();
4111
4112 return 0;
4113}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004114#else
4115static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4116 struct cftype *cft, u64 val)
4117{
4118 return -ENOSYS;
4119}
4120#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004121
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004122
4123/* For read statistics */
4124enum {
4125 MCS_CACHE,
4126 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004127 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004128 MCS_PGPGIN,
4129 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004130 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07004131 MCS_PGFAULT,
4132 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004133 MCS_INACTIVE_ANON,
4134 MCS_ACTIVE_ANON,
4135 MCS_INACTIVE_FILE,
4136 MCS_ACTIVE_FILE,
4137 MCS_UNEVICTABLE,
4138 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004139};
4140
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004141struct mcs_total_stat {
4142 s64 stat[NR_MCS_STAT];
4143};
4144
4145struct {
4146 char *local_name;
4147 char *total_name;
4148} memcg_stat_strings[NR_MCS_STAT] = {
4149 {"cache", "total_cache"},
4150 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004151 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004152 {"pgpgin", "total_pgpgin"},
4153 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004154 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004155 {"pgfault", "total_pgfault"},
4156 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004157 {"inactive_anon", "total_inactive_anon"},
4158 {"active_anon", "total_active_anon"},
4159 {"inactive_file", "total_inactive_file"},
4160 {"active_file", "total_active_file"},
4161 {"unevictable", "total_unevictable"}
4162};
4163
4164
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004165static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004166mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004167{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004168 s64 val;
4169
4170 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004171 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004172 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004173 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004174 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004175 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004176 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004177 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004178 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004179 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004180 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004181 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004182 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004183 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4184 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004185 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004186 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004187 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004188 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004189
4190 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004191 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004192 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004193 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004194 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004195 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004196 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004197 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004198 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004199 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004200 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004201}
4202
4203static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004204mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004205{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004206 struct mem_cgroup *iter;
4207
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004208 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004209 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004210}
4211
Ying Han406eb0c2011-05-26 16:25:37 -07004212#ifdef CONFIG_NUMA
4213static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4214{
4215 int nid;
4216 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4217 unsigned long node_nr;
4218 struct cgroup *cont = m->private;
4219 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4220
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004221 total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004222 seq_printf(m, "total=%lu", total_nr);
4223 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004224 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004225 seq_printf(m, " N%d=%lu", nid, node_nr);
4226 }
4227 seq_putc(m, '\n');
4228
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004229 file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004230 seq_printf(m, "file=%lu", file_nr);
4231 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004232 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4233 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004234 seq_printf(m, " N%d=%lu", nid, node_nr);
4235 }
4236 seq_putc(m, '\n');
4237
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004238 anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004239 seq_printf(m, "anon=%lu", anon_nr);
4240 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004241 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4242 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004243 seq_printf(m, " N%d=%lu", nid, node_nr);
4244 }
4245 seq_putc(m, '\n');
4246
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004247 unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004248 seq_printf(m, "unevictable=%lu", unevictable_nr);
4249 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004250 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4251 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004252 seq_printf(m, " N%d=%lu", nid, node_nr);
4253 }
4254 seq_putc(m, '\n');
4255 return 0;
4256}
4257#endif /* CONFIG_NUMA */
4258
Paul Menagec64745c2008-04-29 01:00:02 -07004259static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4260 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004261{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004262 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004263 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004264 int i;
4265
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004266 memset(&mystat, 0, sizeof(mystat));
4267 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004268
Ying Han406eb0c2011-05-26 16:25:37 -07004269
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004270 for (i = 0; i < NR_MCS_STAT; i++) {
4271 if (i == MCS_SWAP && !do_swap_account)
4272 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004273 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004274 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004275
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004276 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004277 {
4278 unsigned long long limit, memsw_limit;
4279 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
4280 cb->fill(cb, "hierarchical_memory_limit", limit);
4281 if (do_swap_account)
4282 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4283 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004284
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004285 memset(&mystat, 0, sizeof(mystat));
4286 mem_cgroup_get_total_stat(mem_cont, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004287 for (i = 0; i < NR_MCS_STAT; i++) {
4288 if (i == MCS_SWAP && !do_swap_account)
4289 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004290 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004291 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004292
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004293#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004294 {
4295 int nid, zid;
4296 struct mem_cgroup_per_zone *mz;
4297 unsigned long recent_rotated[2] = {0, 0};
4298 unsigned long recent_scanned[2] = {0, 0};
4299
4300 for_each_online_node(nid)
4301 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4302 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
4303
4304 recent_rotated[0] +=
4305 mz->reclaim_stat.recent_rotated[0];
4306 recent_rotated[1] +=
4307 mz->reclaim_stat.recent_rotated[1];
4308 recent_scanned[0] +=
4309 mz->reclaim_stat.recent_scanned[0];
4310 recent_scanned[1] +=
4311 mz->reclaim_stat.recent_scanned[1];
4312 }
4313 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4314 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4315 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4316 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4317 }
4318#endif
4319
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004320 return 0;
4321}
4322
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004323static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4324{
4325 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4326
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004327 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004328}
4329
4330static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4331 u64 val)
4332{
4333 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4334 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004335
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004336 if (val > 100)
4337 return -EINVAL;
4338
4339 if (cgrp->parent == NULL)
4340 return -EINVAL;
4341
4342 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004343
4344 cgroup_lock();
4345
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004346 /* If under hierarchy, only empty-root can set this value */
4347 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004348 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4349 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004350 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004351 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004352
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004353 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004354
Li Zefan068b38c2009-01-15 13:51:26 -08004355 cgroup_unlock();
4356
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004357 return 0;
4358}
4359
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004360static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4361{
4362 struct mem_cgroup_threshold_ary *t;
4363 u64 usage;
4364 int i;
4365
4366 rcu_read_lock();
4367 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004368 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004369 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004370 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004371
4372 if (!t)
4373 goto unlock;
4374
4375 usage = mem_cgroup_usage(memcg, swap);
4376
4377 /*
4378 * current_threshold points to threshold just below usage.
4379 * If it's not true, a threshold was crossed after last
4380 * call of __mem_cgroup_threshold().
4381 */
Phil Carmody5407a562010-05-26 14:42:42 -07004382 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004383
4384 /*
4385 * Iterate backward over array of thresholds starting from
4386 * current_threshold and check if a threshold is crossed.
4387 * If none of thresholds below usage is crossed, we read
4388 * only one element of the array here.
4389 */
4390 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4391 eventfd_signal(t->entries[i].eventfd, 1);
4392
4393 /* i = current_threshold + 1 */
4394 i++;
4395
4396 /*
4397 * Iterate forward over array of thresholds starting from
4398 * current_threshold+1 and check if a threshold is crossed.
4399 * If none of thresholds above usage is crossed, we read
4400 * only one element of the array here.
4401 */
4402 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4403 eventfd_signal(t->entries[i].eventfd, 1);
4404
4405 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004406 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004407unlock:
4408 rcu_read_unlock();
4409}
4410
4411static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4412{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004413 while (memcg) {
4414 __mem_cgroup_threshold(memcg, false);
4415 if (do_swap_account)
4416 __mem_cgroup_threshold(memcg, true);
4417
4418 memcg = parent_mem_cgroup(memcg);
4419 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004420}
4421
4422static int compare_thresholds(const void *a, const void *b)
4423{
4424 const struct mem_cgroup_threshold *_a = a;
4425 const struct mem_cgroup_threshold *_b = b;
4426
4427 return _a->threshold - _b->threshold;
4428}
4429
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004430static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004431{
4432 struct mem_cgroup_eventfd_list *ev;
4433
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004434 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004435 eventfd_signal(ev->eventfd, 1);
4436 return 0;
4437}
4438
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004439static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004440{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004441 struct mem_cgroup *iter;
4442
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004443 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004444 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004445}
4446
4447static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4448 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004449{
4450 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004451 struct mem_cgroup_thresholds *thresholds;
4452 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004453 int type = MEMFILE_TYPE(cft->private);
4454 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004455 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004456
4457 ret = res_counter_memparse_write_strategy(args, &threshold);
4458 if (ret)
4459 return ret;
4460
4461 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004462
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004463 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004464 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004465 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004466 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004467 else
4468 BUG();
4469
4470 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4471
4472 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004473 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004474 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4475
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004476 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004477
4478 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004479 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004480 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004481 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004482 ret = -ENOMEM;
4483 goto unlock;
4484 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004485 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004486
4487 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004488 if (thresholds->primary) {
4489 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004490 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004491 }
4492
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004493 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004494 new->entries[size - 1].eventfd = eventfd;
4495 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004496
4497 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004498 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004499 compare_thresholds, NULL);
4500
4501 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004502 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004503 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004504 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004505 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004506 * new->current_threshold will not be used until
4507 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004508 * it here.
4509 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004510 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511 }
4512 }
4513
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004514 /* Free old spare buffer and save old primary buffer as spare */
4515 kfree(thresholds->spare);
4516 thresholds->spare = thresholds->primary;
4517
4518 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004519
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004520 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004521 synchronize_rcu();
4522
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004523unlock:
4524 mutex_unlock(&memcg->thresholds_lock);
4525
4526 return ret;
4527}
4528
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004529static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004530 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004531{
4532 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004533 struct mem_cgroup_thresholds *thresholds;
4534 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004535 int type = MEMFILE_TYPE(cft->private);
4536 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004537 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004538
4539 mutex_lock(&memcg->thresholds_lock);
4540 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004541 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004542 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004543 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004544 else
4545 BUG();
4546
4547 /*
4548 * Something went wrong if we trying to unregister a threshold
4549 * if we don't have thresholds
4550 */
4551 BUG_ON(!thresholds);
4552
4553 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4554
4555 /* Check if a threshold crossed before removing */
4556 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4557
4558 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004559 size = 0;
4560 for (i = 0; i < thresholds->primary->size; i++) {
4561 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004562 size++;
4563 }
4564
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004565 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004566
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004567 /* Set thresholds array to NULL if we don't have thresholds */
4568 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004569 kfree(new);
4570 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004571 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004572 }
4573
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004574 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004575
4576 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004577 new->current_threshold = -1;
4578 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4579 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004580 continue;
4581
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004582 new->entries[j] = thresholds->primary->entries[i];
4583 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004584 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004585 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004586 * until rcu_assign_pointer(), so it's safe to increment
4587 * it here.
4588 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004589 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004590 }
4591 j++;
4592 }
4593
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004594swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004595 /* Swap primary and spare array */
4596 thresholds->spare = thresholds->primary;
4597 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004598
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004599 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004600 synchronize_rcu();
4601
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004602 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004603}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004604
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004605static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4606 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4607{
4608 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4609 struct mem_cgroup_eventfd_list *event;
4610 int type = MEMFILE_TYPE(cft->private);
4611
4612 BUG_ON(type != _OOM_TYPE);
4613 event = kmalloc(sizeof(*event), GFP_KERNEL);
4614 if (!event)
4615 return -ENOMEM;
4616
Michal Hocko1af8efe2011-07-26 16:08:24 -07004617 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004618
4619 event->eventfd = eventfd;
4620 list_add(&event->list, &memcg->oom_notify);
4621
4622 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004623 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004624 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004625 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004626
4627 return 0;
4628}
4629
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004630static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004631 struct cftype *cft, struct eventfd_ctx *eventfd)
4632{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004633 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004634 struct mem_cgroup_eventfd_list *ev, *tmp;
4635 int type = MEMFILE_TYPE(cft->private);
4636
4637 BUG_ON(type != _OOM_TYPE);
4638
Michal Hocko1af8efe2011-07-26 16:08:24 -07004639 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004640
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004641 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004642 if (ev->eventfd == eventfd) {
4643 list_del(&ev->list);
4644 kfree(ev);
4645 }
4646 }
4647
Michal Hocko1af8efe2011-07-26 16:08:24 -07004648 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004649}
4650
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004651static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4652 struct cftype *cft, struct cgroup_map_cb *cb)
4653{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004654 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004655
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004656 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004657
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004658 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004659 cb->fill(cb, "under_oom", 1);
4660 else
4661 cb->fill(cb, "under_oom", 0);
4662 return 0;
4663}
4664
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004665static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4666 struct cftype *cft, u64 val)
4667{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004668 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004669 struct mem_cgroup *parent;
4670
4671 /* cannot set to root cgroup and only 0 and 1 are allowed */
4672 if (!cgrp->parent || !((val == 0) || (val == 1)))
4673 return -EINVAL;
4674
4675 parent = mem_cgroup_from_cont(cgrp->parent);
4676
4677 cgroup_lock();
4678 /* oom-kill-disable is a flag for subhierarchy. */
4679 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004680 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004681 cgroup_unlock();
4682 return -EINVAL;
4683 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004684 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004685 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004686 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004687 cgroup_unlock();
4688 return 0;
4689}
4690
Ying Han406eb0c2011-05-26 16:25:37 -07004691#ifdef CONFIG_NUMA
4692static const struct file_operations mem_control_numa_stat_file_operations = {
4693 .read = seq_read,
4694 .llseek = seq_lseek,
4695 .release = single_release,
4696};
4697
4698static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4699{
4700 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4701
4702 file->f_op = &mem_control_numa_stat_file_operations;
4703 return single_open(file, mem_control_numa_stat_show, cont);
4704}
4705#endif /* CONFIG_NUMA */
4706
Glauber Costae5671df2011-12-11 21:47:01 +00004707#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae5671df2011-12-11 21:47:01 +00004708static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4709{
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004710 /*
4711 * Part of this would be better living in a separate allocation
4712 * function, leaving us with just the cgroup tree population work.
4713 * We, however, depend on state such as network's proto_list that
4714 * is only initialized after cgroup creation. I found the less
4715 * cumbersome way to deal with it to defer it all to populate time
4716 */
Glauber Costa65c64ce2011-12-22 01:02:27 +00004717 return mem_cgroup_sockets_init(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004718};
4719
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004720static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4721 struct cgroup *cont)
4722{
4723 mem_cgroup_sockets_destroy(cont, ss);
4724}
Glauber Costae5671df2011-12-11 21:47:01 +00004725#else
4726static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4727{
4728 return 0;
4729}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004730
4731static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4732 struct cgroup *cont)
4733{
4734}
Glauber Costae5671df2011-12-11 21:47:01 +00004735#endif
4736
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004737static struct cftype mem_cgroup_files[] = {
4738 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004739 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004740 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004741 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004742 .register_event = mem_cgroup_usage_register_event,
4743 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004744 },
4745 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004746 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004747 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004748 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004749 .read_u64 = mem_cgroup_read,
4750 },
4751 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004752 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004753 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004754 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004755 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004756 },
4757 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004758 .name = "soft_limit_in_bytes",
4759 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4760 .write_string = mem_cgroup_write,
4761 .read_u64 = mem_cgroup_read,
4762 },
4763 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004764 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004765 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004766 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004767 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004768 },
Balbir Singh8697d332008-02-07 00:13:59 -08004769 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004770 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004771 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004772 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004773 {
4774 .name = "force_empty",
4775 .trigger = mem_cgroup_force_empty_write,
4776 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004777 {
4778 .name = "use_hierarchy",
4779 .write_u64 = mem_cgroup_hierarchy_write,
4780 .read_u64 = mem_cgroup_hierarchy_read,
4781 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004782 {
4783 .name = "swappiness",
4784 .read_u64 = mem_cgroup_swappiness_read,
4785 .write_u64 = mem_cgroup_swappiness_write,
4786 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004787 {
4788 .name = "move_charge_at_immigrate",
4789 .read_u64 = mem_cgroup_move_charge_read,
4790 .write_u64 = mem_cgroup_move_charge_write,
4791 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004792 {
4793 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004794 .read_map = mem_cgroup_oom_control_read,
4795 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004796 .register_event = mem_cgroup_oom_register_event,
4797 .unregister_event = mem_cgroup_oom_unregister_event,
4798 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4799 },
Ying Han406eb0c2011-05-26 16:25:37 -07004800#ifdef CONFIG_NUMA
4801 {
4802 .name = "numa_stat",
4803 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004804 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004805 },
4806#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004807};
4808
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004809#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4810static struct cftype memsw_cgroup_files[] = {
4811 {
4812 .name = "memsw.usage_in_bytes",
4813 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4814 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004815 .register_event = mem_cgroup_usage_register_event,
4816 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004817 },
4818 {
4819 .name = "memsw.max_usage_in_bytes",
4820 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4821 .trigger = mem_cgroup_reset,
4822 .read_u64 = mem_cgroup_read,
4823 },
4824 {
4825 .name = "memsw.limit_in_bytes",
4826 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4827 .write_string = mem_cgroup_write,
4828 .read_u64 = mem_cgroup_read,
4829 },
4830 {
4831 .name = "memsw.failcnt",
4832 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4833 .trigger = mem_cgroup_reset,
4834 .read_u64 = mem_cgroup_read,
4835 },
4836};
4837
4838static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4839{
4840 if (!do_swap_account)
4841 return 0;
4842 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4843 ARRAY_SIZE(memsw_cgroup_files));
4844};
4845#else
4846static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4847{
4848 return 0;
4849}
4850#endif
4851
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004852static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004853{
4854 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004855 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004856 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004857 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004858 /*
4859 * This routine is called against possible nodes.
4860 * But it's BUG to call kmalloc() against offline node.
4861 *
4862 * TODO: this routine can waste much memory for nodes which will
4863 * never be onlined. It's better to use memory hotplug callback
4864 * function.
4865 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004866 if (!node_state(node, N_NORMAL_MEMORY))
4867 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004868 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004869 if (!pn)
4870 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004871
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004872 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4873 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07004874 for_each_lru(l)
4875 INIT_LIST_HEAD(&mz->lists[l]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004876 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004877 mz->on_tree = false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004878 mz->mem = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004879 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004880 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004881 return 0;
4882}
4883
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004884static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004885{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004886 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004887}
4888
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004889static struct mem_cgroup *mem_cgroup_alloc(void)
4890{
4891 struct mem_cgroup *mem;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004892 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004893
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004894 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004895 if (size < PAGE_SIZE)
Jesper Juhl17295c82011-01-13 15:47:42 -08004896 mem = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004897 else
Jesper Juhl17295c82011-01-13 15:47:42 -08004898 mem = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004899
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004900 if (!mem)
4901 return NULL;
4902
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004903 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004904 if (!mem->stat)
4905 goto out_free;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004906 spin_lock_init(&mem->pcp_counter_lock);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004907 return mem;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004908
4909out_free:
4910 if (size < PAGE_SIZE)
4911 kfree(mem);
4912 else
4913 vfree(mem);
4914 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004915}
4916
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004917/*
4918 * At destroying mem_cgroup, references from swap_cgroup can remain.
4919 * (scanning all at force_empty is too costly...)
4920 *
4921 * Instead of clearing all references at force_empty, we remember
4922 * the number of reference from swap_cgroup and free mem_cgroup when
4923 * it goes down to 0.
4924 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004925 * Removal of cgroup itself succeeds regardless of refs from swap.
4926 */
4927
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004928static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004929{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004930 int node;
4931
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004932 mem_cgroup_remove_from_trees(memcg);
4933 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004934
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004935 for_each_node_state(node, N_POSSIBLE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004936 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004937
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004938 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004939 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004940 kfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004941 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004942 vfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004943}
4944
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004945static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004946{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004947 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004948}
4949
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004950static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004951{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004952 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4953 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4954 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004955 if (parent)
4956 mem_cgroup_put(parent);
4957 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004958}
4959
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004960static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004961{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004962 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004963}
4964
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004965/*
4966 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4967 */
Glauber Costae1aab162011-12-11 21:47:03 +00004968struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004969{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004970 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004971 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004972 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004973}
Glauber Costae1aab162011-12-11 21:47:03 +00004974EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004975
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004976#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4977static void __init enable_swap_cgroup(void)
4978{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08004979 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004980 do_swap_account = 1;
4981}
4982#else
4983static void __init enable_swap_cgroup(void)
4984{
4985}
4986#endif
4987
Balbir Singhf64c3f52009-09-23 15:56:37 -07004988static int mem_cgroup_soft_limit_tree_init(void)
4989{
4990 struct mem_cgroup_tree_per_node *rtpn;
4991 struct mem_cgroup_tree_per_zone *rtpz;
4992 int tmp, node, zone;
4993
4994 for_each_node_state(node, N_POSSIBLE) {
4995 tmp = node;
4996 if (!node_state(node, N_NORMAL_MEMORY))
4997 tmp = -1;
4998 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4999 if (!rtpn)
5000 return 1;
5001
5002 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5003
5004 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5005 rtpz = &rtpn->rb_tree_per_zone[zone];
5006 rtpz->rb_root = RB_ROOT;
5007 spin_lock_init(&rtpz->lock);
5008 }
5009 }
5010 return 0;
5011}
5012
Li Zefan0eb253e2009-01-15 13:51:25 -08005013static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005014mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
5015{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005016 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005017 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005018 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005019
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005020 memcg = mem_cgroup_alloc();
5021 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005022 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005023
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005024 for_each_node_state(node, N_POSSIBLE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005025 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005026 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005027
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005028 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005029 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08005030 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005031 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005032 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005033 if (mem_cgroup_soft_limit_tree_init())
5034 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08005035 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08005036 for_each_possible_cpu(cpu) {
5037 struct memcg_stock_pcp *stock =
5038 &per_cpu(memcg_stock, cpu);
5039 INIT_WORK(&stock->work, drain_local_stock);
5040 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07005041 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005042 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005043 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005044 memcg->use_hierarchy = parent->use_hierarchy;
5045 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005046 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005047
Balbir Singh18f59ea2009-01-07 18:08:07 -08005048 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005049 res_counter_init(&memcg->res, &parent->res);
5050 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08005051 /*
5052 * We increment refcnt of the parent to ensure that we can
5053 * safely access it on res_counter_charge/uncharge.
5054 * This refcnt will be decremented when freeing this
5055 * mem_cgroup(see mem_cgroup_put).
5056 */
5057 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005058 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005059 res_counter_init(&memcg->res, NULL);
5060 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005061 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005062 memcg->last_scanned_node = MAX_NUMNODES;
5063 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08005064
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005065 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005066 memcg->swappiness = mem_cgroup_swappiness(parent);
5067 atomic_set(&memcg->refcnt, 1);
5068 memcg->move_charge_at_immigrate = 0;
5069 mutex_init(&memcg->thresholds_lock);
5070 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005071free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005072 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005073 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005074}
5075
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005076static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005077 struct cgroup *cont)
5078{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005079 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005080
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005081 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005082}
5083
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005084static void mem_cgroup_destroy(struct cgroup_subsys *ss,
5085 struct cgroup *cont)
5086{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005087 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005088
Glauber Costad1a4c0b2011-12-11 21:47:04 +00005089 kmem_cgroup_destroy(ss, cont);
5090
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005091 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005092}
5093
5094static int mem_cgroup_populate(struct cgroup_subsys *ss,
5095 struct cgroup *cont)
5096{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005097 int ret;
5098
5099 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
5100 ARRAY_SIZE(mem_cgroup_files));
5101
5102 if (!ret)
5103 ret = register_memsw_files(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00005104
5105 if (!ret)
5106 ret = register_kmem_files(cont, ss);
5107
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08005108 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005109}
5110
Daisuke Nishimura02491442010-03-10 15:22:17 -08005111#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005112/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005113#define PRECHARGE_COUNT_AT_ONCE 256
5114static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005115{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005116 int ret = 0;
5117 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005118 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005119
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005120 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005121 mc.precharge += count;
5122 /* we don't need css_get for root */
5123 return ret;
5124 }
5125 /* try to charge at once */
5126 if (count > 1) {
5127 struct res_counter *dummy;
5128 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005129 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005130 * by cgroup_lock_live_cgroup() that it is not removed and we
5131 * are still under the same cgroup_mutex. So we can postpone
5132 * css_get().
5133 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005134 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005135 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005136 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005137 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005138 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005139 goto one_by_one;
5140 }
5141 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005142 return ret;
5143 }
5144one_by_one:
5145 /* fall back to one by one charge */
5146 while (count--) {
5147 if (signal_pending(current)) {
5148 ret = -EINTR;
5149 break;
5150 }
5151 if (!batch_count--) {
5152 batch_count = PRECHARGE_COUNT_AT_ONCE;
5153 cond_resched();
5154 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005155 ret = __mem_cgroup_try_charge(NULL,
5156 GFP_KERNEL, 1, &memcg, false);
5157 if (ret || !memcg)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005158 /* mem_cgroup_clear_mc() will do uncharge later */
5159 return -ENOMEM;
5160 mc.precharge++;
5161 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005162 return ret;
5163}
5164
5165/**
5166 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5167 * @vma: the vma the pte to be checked belongs
5168 * @addr: the address corresponding to the pte to be checked
5169 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005170 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005171 *
5172 * Returns
5173 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5174 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5175 * move charge. if @target is not NULL, the page is stored in target->page
5176 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005177 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5178 * target for charge migration. if @target is not NULL, the entry is stored
5179 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005180 *
5181 * Called with pte lock held.
5182 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005183union mc_target {
5184 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005185 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005186};
5187
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005188enum mc_target_type {
5189 MC_TARGET_NONE, /* not used */
5190 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005191 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005192};
5193
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005194static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5195 unsigned long addr, pte_t ptent)
5196{
5197 struct page *page = vm_normal_page(vma, addr, ptent);
5198
5199 if (!page || !page_mapped(page))
5200 return NULL;
5201 if (PageAnon(page)) {
5202 /* we don't move shared anon */
5203 if (!move_anon() || page_mapcount(page) > 2)
5204 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005205 } else if (!move_file())
5206 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005207 return NULL;
5208 if (!get_page_unless_zero(page))
5209 return NULL;
5210
5211 return page;
5212}
5213
5214static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5215 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5216{
5217 int usage_count;
5218 struct page *page = NULL;
5219 swp_entry_t ent = pte_to_swp_entry(ptent);
5220
5221 if (!move_anon() || non_swap_entry(ent))
5222 return NULL;
5223 usage_count = mem_cgroup_count_swap_user(ent, &page);
5224 if (usage_count > 1) { /* we don't move shared anon */
5225 if (page)
5226 put_page(page);
5227 return NULL;
5228 }
5229 if (do_swap_account)
5230 entry->val = ent.val;
5231
5232 return page;
5233}
5234
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005235static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5236 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5237{
5238 struct page *page = NULL;
5239 struct inode *inode;
5240 struct address_space *mapping;
5241 pgoff_t pgoff;
5242
5243 if (!vma->vm_file) /* anonymous vma */
5244 return NULL;
5245 if (!move_file())
5246 return NULL;
5247
5248 inode = vma->vm_file->f_path.dentry->d_inode;
5249 mapping = vma->vm_file->f_mapping;
5250 if (pte_none(ptent))
5251 pgoff = linear_page_index(vma, addr);
5252 else /* pte_file(ptent) is true */
5253 pgoff = pte_to_pgoff(ptent);
5254
5255 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005256 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005257
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005258#ifdef CONFIG_SWAP
5259 /* shmem/tmpfs may report page out on swap: account for that too. */
5260 if (radix_tree_exceptional_entry(page)) {
5261 swp_entry_t swap = radix_to_swp_entry(page);
5262 if (do_swap_account)
5263 *entry = swap;
5264 page = find_get_page(&swapper_space, swap.val);
5265 }
5266#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005267 return page;
5268}
5269
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005270static int is_target_pte_for_mc(struct vm_area_struct *vma,
5271 unsigned long addr, pte_t ptent, union mc_target *target)
5272{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005273 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005274 struct page_cgroup *pc;
5275 int ret = 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005276 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005277
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005278 if (pte_present(ptent))
5279 page = mc_handle_present_pte(vma, addr, ptent);
5280 else if (is_swap_pte(ptent))
5281 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005282 else if (pte_none(ptent) || pte_file(ptent))
5283 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005284
5285 if (!page && !ent.val)
Daisuke Nishimura02491442010-03-10 15:22:17 -08005286 return 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005287 if (page) {
5288 pc = lookup_page_cgroup(page);
5289 /*
5290 * Do only loose check w/o page_cgroup lock.
5291 * mem_cgroup_move_account() checks the pc is valid or not under
5292 * the lock.
5293 */
5294 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5295 ret = MC_TARGET_PAGE;
5296 if (target)
5297 target->page = page;
5298 }
5299 if (!ret || !target)
5300 put_page(page);
5301 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005302 /* There is a swap entry and a page doesn't exist or isn't charged */
5303 if (ent.val && !ret &&
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005304 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
5305 ret = MC_TARGET_SWAP;
5306 if (target)
5307 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005308 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005309 return ret;
5310}
5311
5312static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5313 unsigned long addr, unsigned long end,
5314 struct mm_walk *walk)
5315{
5316 struct vm_area_struct *vma = walk->private;
5317 pte_t *pte;
5318 spinlock_t *ptl;
5319
Dave Hansen03319322011-03-22 16:32:56 -07005320 split_huge_page_pmd(walk->mm, pmd);
5321
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005322 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5323 for (; addr != end; pte++, addr += PAGE_SIZE)
5324 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
5325 mc.precharge++; /* increment precharge temporarily */
5326 pte_unmap_unlock(pte - 1, ptl);
5327 cond_resched();
5328
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005329 return 0;
5330}
5331
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005332static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5333{
5334 unsigned long precharge;
5335 struct vm_area_struct *vma;
5336
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005337 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005338 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5339 struct mm_walk mem_cgroup_count_precharge_walk = {
5340 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5341 .mm = mm,
5342 .private = vma,
5343 };
5344 if (is_vm_hugetlb_page(vma))
5345 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005346 walk_page_range(vma->vm_start, vma->vm_end,
5347 &mem_cgroup_count_precharge_walk);
5348 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005349 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005350
5351 precharge = mc.precharge;
5352 mc.precharge = 0;
5353
5354 return precharge;
5355}
5356
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005357static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5358{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005359 unsigned long precharge = mem_cgroup_count_precharge(mm);
5360
5361 VM_BUG_ON(mc.moving_task);
5362 mc.moving_task = current;
5363 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005364}
5365
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005366/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5367static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005368{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005369 struct mem_cgroup *from = mc.from;
5370 struct mem_cgroup *to = mc.to;
5371
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005372 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005373 if (mc.precharge) {
5374 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5375 mc.precharge = 0;
5376 }
5377 /*
5378 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5379 * we must uncharge here.
5380 */
5381 if (mc.moved_charge) {
5382 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5383 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005384 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005385 /* we must fixup refcnts and charges */
5386 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005387 /* uncharge swap account from the old cgroup */
5388 if (!mem_cgroup_is_root(mc.from))
5389 res_counter_uncharge(&mc.from->memsw,
5390 PAGE_SIZE * mc.moved_swap);
5391 __mem_cgroup_put(mc.from, mc.moved_swap);
5392
5393 if (!mem_cgroup_is_root(mc.to)) {
5394 /*
5395 * we charged both to->res and to->memsw, so we should
5396 * uncharge to->res.
5397 */
5398 res_counter_uncharge(&mc.to->res,
5399 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005400 }
5401 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005402 mc.moved_swap = 0;
5403 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005404 memcg_oom_recover(from);
5405 memcg_oom_recover(to);
5406 wake_up_all(&mc.waitq);
5407}
5408
5409static void mem_cgroup_clear_mc(void)
5410{
5411 struct mem_cgroup *from = mc.from;
5412
5413 /*
5414 * we must clear moving_task before waking up waiters at the end of
5415 * task migration.
5416 */
5417 mc.moving_task = NULL;
5418 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005419 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005420 mc.from = NULL;
5421 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005422 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005423 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005424}
5425
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005426static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5427 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005428 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005429{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005430 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005431 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005432 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005433
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005434 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005435 struct mm_struct *mm;
5436 struct mem_cgroup *from = mem_cgroup_from_task(p);
5437
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005438 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005439
5440 mm = get_task_mm(p);
5441 if (!mm)
5442 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005443 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005444 if (mm->owner == p) {
5445 VM_BUG_ON(mc.from);
5446 VM_BUG_ON(mc.to);
5447 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005448 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005449 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005450 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005451 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005452 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005453 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005454 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005455 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005456
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005457 ret = mem_cgroup_precharge_mc(mm);
5458 if (ret)
5459 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005460 }
5461 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005462 }
5463 return ret;
5464}
5465
5466static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5467 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005468 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005469{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005470 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005471}
5472
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005473static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5474 unsigned long addr, unsigned long end,
5475 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005476{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005477 int ret = 0;
5478 struct vm_area_struct *vma = walk->private;
5479 pte_t *pte;
5480 spinlock_t *ptl;
5481
Dave Hansen03319322011-03-22 16:32:56 -07005482 split_huge_page_pmd(walk->mm, pmd);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005483retry:
5484 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5485 for (; addr != end; addr += PAGE_SIZE) {
5486 pte_t ptent = *(pte++);
5487 union mc_target target;
5488 int type;
5489 struct page *page;
5490 struct page_cgroup *pc;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005491 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005492
5493 if (!mc.precharge)
5494 break;
5495
5496 type = is_target_pte_for_mc(vma, addr, ptent, &target);
5497 switch (type) {
5498 case MC_TARGET_PAGE:
5499 page = target.page;
5500 if (isolate_lru_page(page))
5501 goto put;
5502 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005503 if (!mem_cgroup_move_account(page, 1, pc,
5504 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005505 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005506 /* we uncharge from mc.from later. */
5507 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005508 }
5509 putback_lru_page(page);
5510put: /* is_target_pte_for_mc() gets the page */
5511 put_page(page);
5512 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005513 case MC_TARGET_SWAP:
5514 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005515 if (!mem_cgroup_move_swap_account(ent,
5516 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005517 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005518 /* we fixup refcnts and charges later. */
5519 mc.moved_swap++;
5520 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005521 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005522 default:
5523 break;
5524 }
5525 }
5526 pte_unmap_unlock(pte - 1, ptl);
5527 cond_resched();
5528
5529 if (addr != end) {
5530 /*
5531 * We have consumed all precharges we got in can_attach().
5532 * We try charge one by one, but don't do any additional
5533 * charges to mc.to if we have failed in charge once in attach()
5534 * phase.
5535 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005536 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005537 if (!ret)
5538 goto retry;
5539 }
5540
5541 return ret;
5542}
5543
5544static void mem_cgroup_move_charge(struct mm_struct *mm)
5545{
5546 struct vm_area_struct *vma;
5547
5548 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005549retry:
5550 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5551 /*
5552 * Someone who are holding the mmap_sem might be waiting in
5553 * waitq. So we cancel all extra charges, wake up all waiters,
5554 * and retry. Because we cancel precharges, we might not be able
5555 * to move enough charges, but moving charge is a best-effort
5556 * feature anyway, so it wouldn't be a big problem.
5557 */
5558 __mem_cgroup_clear_mc();
5559 cond_resched();
5560 goto retry;
5561 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005562 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5563 int ret;
5564 struct mm_walk mem_cgroup_move_charge_walk = {
5565 .pmd_entry = mem_cgroup_move_charge_pte_range,
5566 .mm = mm,
5567 .private = vma,
5568 };
5569 if (is_vm_hugetlb_page(vma))
5570 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005571 ret = walk_page_range(vma->vm_start, vma->vm_end,
5572 &mem_cgroup_move_charge_walk);
5573 if (ret)
5574 /*
5575 * means we have consumed all precharges and failed in
5576 * doing additional charge. Just abandon here.
5577 */
5578 break;
5579 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005580 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005581}
5582
Balbir Singh67e465a2008-02-07 00:13:54 -08005583static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5584 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005585 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005586{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005587 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005588 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005589
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005590 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005591 if (mc.to)
5592 mem_cgroup_move_charge(mm);
5593 put_swap_token(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005594 mmput(mm);
5595 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005596 if (mc.to)
5597 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005598}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005599#else /* !CONFIG_MMU */
5600static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5601 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005602 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005603{
5604 return 0;
5605}
5606static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5607 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005608 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005609{
5610}
5611static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5612 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005613 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005614{
5615}
5616#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005617
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005618struct cgroup_subsys mem_cgroup_subsys = {
5619 .name = "memory",
5620 .subsys_id = mem_cgroup_subsys_id,
5621 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005622 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005623 .destroy = mem_cgroup_destroy,
5624 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005625 .can_attach = mem_cgroup_can_attach,
5626 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005627 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005628 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005629 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005630};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005631
5632#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005633static int __init enable_swap_account(char *s)
5634{
5635 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005636 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005637 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005638 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005639 really_do_swap_account = 0;
5640 return 1;
5641}
Michal Hockoa2c89902011-05-24 17:12:50 -07005642__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005643
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005644#endif