blob: 3dbff4dcde35191a62b0245ac7ee185bc8fc7084 [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 Hiroyuki8c7c6e342009-01-07 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 {
Johannes Weiner6290df52012-01-12 17:18:10 -0800137 struct lruvec lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700138 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800139
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800140 struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
141
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800142 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700143 struct rb_node tree_node; /* RB tree node */
144 unsigned long long usage_in_excess;/* Set to the value by which */
145 /* the soft limit is exceeded*/
146 bool on_tree;
Balbir Singh4e416952009-09-23 15:56:39 -0700147 struct mem_cgroup *mem; /* Back pointer, we cannot */
148 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800149};
150/* Macro for accessing counter */
151#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
152
153struct mem_cgroup_per_node {
154 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
155};
156
157struct mem_cgroup_lru_info {
158 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
159};
160
161/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700162 * Cgroups above their limits are maintained in a RB-Tree, independent of
163 * their hierarchy representation
164 */
165
166struct mem_cgroup_tree_per_zone {
167 struct rb_root rb_root;
168 spinlock_t lock;
169};
170
171struct mem_cgroup_tree_per_node {
172 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
173};
174
175struct mem_cgroup_tree {
176 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
177};
178
179static struct mem_cgroup_tree soft_limit_tree __read_mostly;
180
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800181struct mem_cgroup_threshold {
182 struct eventfd_ctx *eventfd;
183 u64 threshold;
184};
185
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700186/* For threshold */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800187struct mem_cgroup_threshold_ary {
188 /* An array index points to threshold just below usage. */
Phil Carmody5407a562010-05-26 14:42:42 -0700189 int current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800190 /* Size of entries[] */
191 unsigned int size;
192 /* Array of thresholds */
193 struct mem_cgroup_threshold entries[0];
194};
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700195
196struct mem_cgroup_thresholds {
197 /* Primary thresholds array */
198 struct mem_cgroup_threshold_ary *primary;
199 /*
200 * Spare threshold array.
201 * This is needed to make mem_cgroup_unregister_event() "never fail".
202 * It must be able to store at least primary->size - 1 entries.
203 */
204 struct mem_cgroup_threshold_ary *spare;
205};
206
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700207/* for OOM */
208struct mem_cgroup_eventfd_list {
209 struct list_head list;
210 struct eventfd_ctx *eventfd;
211};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800212
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700213static void mem_cgroup_threshold(struct mem_cgroup *memcg);
214static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800215
Balbir Singhf64c3f52009-09-23 15:56:37 -0700216/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800217 * The memory controller data structure. The memory controller controls both
218 * page cache and RSS per cgroup. We would eventually like to provide
219 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
220 * to help the administrator determine what knobs to tune.
221 *
222 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800223 * we hit the water mark. May be even add a low water mark, such that
224 * no reclaim occurs from a cgroup at it's low water mark, this is
225 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800226 */
227struct mem_cgroup {
228 struct cgroup_subsys_state css;
229 /*
230 * the counter to account for memory usage
231 */
232 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800233 /*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800234 * the counter to account for mem+swap usage.
235 */
236 struct res_counter memsw;
237 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800238 * Per cgroup active and inactive list, similar to the
239 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800240 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800241 struct mem_cgroup_lru_info info;
Ying Han889976d2011-05-26 16:25:33 -0700242 int last_scanned_node;
243#if MAX_NUMNODES > 1
244 nodemask_t scan_nodes;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700245 atomic_t numainfo_events;
246 atomic_t numainfo_updating;
Ying Han889976d2011-05-26 16:25:33 -0700247#endif
Balbir Singh18f59ea2009-01-07 18:08:07 -0800248 /*
249 * Should the accounting and control be hierarchical, per subtree?
250 */
251 bool use_hierarchy;
Michal Hocko79dfdac2011-07-26 16:08:23 -0700252
253 bool oom_lock;
254 atomic_t under_oom;
255
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800256 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800257
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -0700258 int swappiness;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -0700259 /* OOM-Killer disable */
260 int oom_kill_disable;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800261
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700262 /* set when res.limit == memsw.limit */
263 bool memsw_is_minimum;
264
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800265 /* protect arrays of thresholds */
266 struct mutex thresholds_lock;
267
268 /* thresholds for memory usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700269 struct mem_cgroup_thresholds thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700270
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800271 /* thresholds for mem+swap usage. RCU-protected */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -0700272 struct mem_cgroup_thresholds memsw_thresholds;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -0700273
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700274 /* For oom notifier event fd */
275 struct list_head oom_notify;
Johannes Weiner185efc02011-09-14 16:21:58 -0700276
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800277 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800278 * Should we move charges of a task when a task is moved into this
279 * mem_cgroup ? And what type of charges should we move ?
280 */
281 unsigned long move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800282 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800283 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800284 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800285 struct mem_cgroup_stat_cpu *stat;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700286 /*
287 * used when a cpu is offlined or other synchronizations
288 * See mem_cgroup_read_stat().
289 */
290 struct mem_cgroup_stat_cpu nocpu_base;
291 spinlock_t pcp_counter_lock;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000292
293#ifdef CONFIG_INET
294 struct tcp_memcontrol tcp_mem;
295#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800296};
297
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800298/* Stuffs for move charges at task migration. */
299/*
300 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
301 * left-shifted bitmap of these types.
302 */
303enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800304 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700305 MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800306 NR_MOVE_TYPE,
307};
308
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800309/* "mc" and its members are protected by cgroup_mutex */
310static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800311 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800312 struct mem_cgroup *from;
313 struct mem_cgroup *to;
314 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800315 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800316 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800317 struct task_struct *moving_task; /* a task moving charges */
318 wait_queue_head_t waitq; /* a waitq for other context */
319} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700320 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800321 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
322};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800323
Daisuke Nishimura90254a62010-05-26 14:42:38 -0700324static bool move_anon(void)
325{
326 return test_bit(MOVE_CHARGE_TYPE_ANON,
327 &mc.to->move_charge_at_immigrate);
328}
329
Daisuke Nishimura87946a72010-05-26 14:42:39 -0700330static bool move_file(void)
331{
332 return test_bit(MOVE_CHARGE_TYPE_FILE,
333 &mc.to->move_charge_at_immigrate);
334}
335
Balbir Singh4e416952009-09-23 15:56:39 -0700336/*
337 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
338 * limit reclaim to prevent infinite loops, if they ever occur.
339 */
340#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
341#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
342
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800343enum charge_type {
344 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
345 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700346 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700347 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800348 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700349 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700350 NR_CHARGE_TYPE,
351};
352
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800353/* for encoding cft->private value on file */
Glauber Costa65c64ce2011-12-22 01:02:27 +0000354#define _MEM (0)
355#define _MEMSWAP (1)
356#define _OOM_TYPE (2)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800357#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
358#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
359#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700360/* Used for OOM nofiier */
361#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800362
Balbir Singh75822b42009-09-23 15:56:38 -0700363/*
364 * Reclaim flags for mem_cgroup_hierarchical_reclaim
365 */
366#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
367#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
368#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
369#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
370
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700371static void mem_cgroup_get(struct mem_cgroup *memcg);
372static void mem_cgroup_put(struct mem_cgroup *memcg);
Glauber Costae1aab162011-12-11 21:47:03 +0000373
374/* Writing them here to avoid exposing memcg's inner layout */
375#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae1aab162011-12-11 21:47:03 +0000376#include <net/sock.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000377#include <net/ip.h>
Glauber Costae1aab162011-12-11 21:47:03 +0000378
379static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
380void sock_update_memcg(struct sock *sk)
381{
Glauber Costae1aab162011-12-11 21:47:03 +0000382 if (static_branch(&memcg_socket_limit_enabled)) {
383 struct mem_cgroup *memcg;
384
385 BUG_ON(!sk->sk_prot->proto_cgroup);
386
Glauber Costaf3f511e2012-01-05 20:16:39 +0000387 /* Socket cloning can throw us here with sk_cgrp already
388 * filled. It won't however, necessarily happen from
389 * process context. So the test for root memcg given
390 * the current task's memcg won't help us in this case.
391 *
392 * Respecting the original socket's memcg is a better
393 * decision in this case.
394 */
395 if (sk->sk_cgrp) {
396 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
397 mem_cgroup_get(sk->sk_cgrp->memcg);
398 return;
399 }
400
Glauber Costae1aab162011-12-11 21:47:03 +0000401 rcu_read_lock();
402 memcg = mem_cgroup_from_task(current);
403 if (!mem_cgroup_is_root(memcg)) {
404 mem_cgroup_get(memcg);
405 sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg);
406 }
407 rcu_read_unlock();
408 }
409}
410EXPORT_SYMBOL(sock_update_memcg);
411
412void sock_release_memcg(struct sock *sk)
413{
414 if (static_branch(&memcg_socket_limit_enabled) && sk->sk_cgrp) {
415 struct mem_cgroup *memcg;
416 WARN_ON(!sk->sk_cgrp->memcg);
417 memcg = sk->sk_cgrp->memcg;
418 mem_cgroup_put(memcg);
419 }
420}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000421
Glauber Costa319d3b92012-01-15 22:04:39 +0000422#ifdef CONFIG_INET
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000423struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
424{
425 if (!memcg || mem_cgroup_is_root(memcg))
426 return NULL;
427
428 return &memcg->tcp_mem.cg_proto;
429}
430EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000431#endif /* CONFIG_INET */
432#endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
433
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700434static void drain_all_stock_async(struct mem_cgroup *memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800435
Balbir Singhf64c3f52009-09-23 15:56:37 -0700436static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700437mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700438{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700439 return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700440}
441
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700442struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
Wu Fengguangd3242362009-12-16 12:19:59 +0100443{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700444 return &memcg->css;
Wu Fengguangd3242362009-12-16 12:19:59 +0100445}
446
Balbir Singhf64c3f52009-09-23 15:56:37 -0700447static struct mem_cgroup_per_zone *
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700448page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700449{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700450 int nid = page_to_nid(page);
451 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700452
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700453 return mem_cgroup_zoneinfo(memcg, nid, zid);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700454}
455
456static struct mem_cgroup_tree_per_zone *
457soft_limit_tree_node_zone(int nid, int zid)
458{
459 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
460}
461
462static struct mem_cgroup_tree_per_zone *
463soft_limit_tree_from_page(struct page *page)
464{
465 int nid = page_to_nid(page);
466 int zid = page_zonenum(page);
467
468 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
469}
470
471static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700472__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700473 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700474 struct mem_cgroup_tree_per_zone *mctz,
475 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700476{
477 struct rb_node **p = &mctz->rb_root.rb_node;
478 struct rb_node *parent = NULL;
479 struct mem_cgroup_per_zone *mz_node;
480
481 if (mz->on_tree)
482 return;
483
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700484 mz->usage_in_excess = new_usage_in_excess;
485 if (!mz->usage_in_excess)
486 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700487 while (*p) {
488 parent = *p;
489 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
490 tree_node);
491 if (mz->usage_in_excess < mz_node->usage_in_excess)
492 p = &(*p)->rb_left;
493 /*
494 * We can't avoid mem cgroups that are over their soft
495 * limit by the same amount
496 */
497 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
498 p = &(*p)->rb_right;
499 }
500 rb_link_node(&mz->tree_node, parent, p);
501 rb_insert_color(&mz->tree_node, &mctz->rb_root);
502 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700503}
504
505static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700506__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singh4e416952009-09-23 15:56:39 -0700507 struct mem_cgroup_per_zone *mz,
508 struct mem_cgroup_tree_per_zone *mctz)
509{
510 if (!mz->on_tree)
511 return;
512 rb_erase(&mz->tree_node, &mctz->rb_root);
513 mz->on_tree = false;
514}
515
516static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700517mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700518 struct mem_cgroup_per_zone *mz,
519 struct mem_cgroup_tree_per_zone *mctz)
520{
521 spin_lock(&mctz->lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700522 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700523 spin_unlock(&mctz->lock);
524}
525
Balbir Singhf64c3f52009-09-23 15:56:37 -0700526
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700527static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700528{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700529 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700530 struct mem_cgroup_per_zone *mz;
531 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700532 int nid = page_to_nid(page);
533 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700534 mctz = soft_limit_tree_from_page(page);
535
536 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700537 * Necessary to update all ancestors when hierarchy is used.
538 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700539 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700540 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
541 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
542 excess = res_counter_soft_limit_excess(&memcg->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700543 /*
544 * We have to update the tree if mz is on RB-tree or
545 * mem is over its softlimit.
546 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700547 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700548 spin_lock(&mctz->lock);
549 /* if on-tree, remove it */
550 if (mz->on_tree)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700551 __mem_cgroup_remove_exceeded(memcg, mz, mctz);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700552 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700553 * Insert again. mz->usage_in_excess will be updated.
554 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700555 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700556 __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700557 spin_unlock(&mctz->lock);
558 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700559 }
560}
561
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700562static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700563{
564 int node, zone;
565 struct mem_cgroup_per_zone *mz;
566 struct mem_cgroup_tree_per_zone *mctz;
567
Bob Liu3ed28fa2012-01-12 17:19:04 -0800568 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -0700569 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700570 mz = mem_cgroup_zoneinfo(memcg, node, zone);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700571 mctz = soft_limit_tree_node_zone(node, zone);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700572 mem_cgroup_remove_exceeded(memcg, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700573 }
574 }
575}
576
Balbir Singh4e416952009-09-23 15:56:39 -0700577static struct mem_cgroup_per_zone *
578__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
579{
580 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700581 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700582
583retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700584 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700585 rightmost = rb_last(&mctz->rb_root);
586 if (!rightmost)
587 goto done; /* Nothing to reclaim from */
588
589 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
590 /*
591 * Remove the node now but someone else can add it back,
592 * we will to add it back at the end of reclaim to its correct
593 * position in the tree.
594 */
595 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
596 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
597 !css_tryget(&mz->mem->css))
598 goto retry;
599done:
600 return mz;
601}
602
603static struct mem_cgroup_per_zone *
604mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
605{
606 struct mem_cgroup_per_zone *mz;
607
608 spin_lock(&mctz->lock);
609 mz = __mem_cgroup_largest_soft_limit_node(mctz);
610 spin_unlock(&mctz->lock);
611 return mz;
612}
613
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700614/*
615 * Implementation Note: reading percpu statistics for memcg.
616 *
617 * Both of vmstat[] and percpu_counter has threshold and do periodic
618 * synchronization to implement "quick" read. There are trade-off between
619 * reading cost and precision of value. Then, we may have a chance to implement
620 * a periodic synchronizion of counter in memcg's counter.
621 *
622 * But this _read() function is used for user interface now. The user accounts
623 * memory usage by memory cgroup and he _always_ requires exact value because
624 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
625 * have to visit all online cpus and make sum. So, for now, unnecessary
626 * synchronization is not implemented. (just implemented for cpu hotplug)
627 *
628 * If there are kernel internal actions which can make use of some not-exact
629 * value, and reading all cpu value can be performance bottleneck in some
630 * common workload, threashold and synchonization as vmstat[] should be
631 * implemented.
632 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700633static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700634 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800635{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700636 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800637 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800638
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700639 get_online_cpus();
640 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700641 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700642#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700643 spin_lock(&memcg->pcp_counter_lock);
644 val += memcg->nocpu_base.count[idx];
645 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700646#endif
647 put_online_cpus();
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800648 return val;
649}
650
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700651static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700652 bool charge)
653{
654 int val = (charge) ? 1 : -1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700655 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700656}
657
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700658static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700659 enum mem_cgroup_events_index idx)
660{
661 unsigned long val = 0;
662 int cpu;
663
664 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700665 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700666#ifdef CONFIG_HOTPLUG_CPU
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700667 spin_lock(&memcg->pcp_counter_lock);
668 val += memcg->nocpu_base.events[idx];
669 spin_unlock(&memcg->pcp_counter_lock);
Johannes Weinere9f89742011-03-23 16:42:37 -0700670#endif
671 return val;
672}
673
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700674static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800675 bool file, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800676{
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800677 preempt_disable();
678
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800679 if (file)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
681 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800682 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700683 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
684 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700685
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800686 /* pagein of a big page is an event. So, ignore page size */
687 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800689 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700690 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800691 nr_pages = -nr_pages; /* for event */
692 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800693
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700694 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800695
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800696 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800697}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800698
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700699unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700700mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700701 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700702{
703 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700704 enum lru_list l;
705 unsigned long ret = 0;
706
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700707 mz = mem_cgroup_zoneinfo(memcg, nid, zid);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700708
709 for_each_lru(l) {
710 if (BIT(l) & lru_mask)
711 ret += MEM_CGROUP_ZSTAT(mz, l);
712 }
713 return ret;
714}
715
716static unsigned long
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700717mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700718 int nid, unsigned int lru_mask)
719{
Ying Han889976d2011-05-26 16:25:33 -0700720 u64 total = 0;
721 int zid;
722
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700723 for (zid = 0; zid < MAX_NR_ZONES; zid++)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700724 total += mem_cgroup_zone_nr_lru_pages(memcg,
725 nid, zid, lru_mask);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700726
Ying Han889976d2011-05-26 16:25:33 -0700727 return total;
728}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700729
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700730static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700731 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800732{
Ying Han889976d2011-05-26 16:25:33 -0700733 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800734 u64 total = 0;
735
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700736 for_each_node_state(nid, N_HIGH_MEMORY)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700737 total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800738 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800739}
740
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800741static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
742 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800743{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700744 unsigned long val, next;
745
Steven Rostedt47994012011-11-02 13:38:33 -0700746 val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]);
747 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700748 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800749 if ((long)next - (long)val < 0) {
750 switch (target) {
751 case MEM_CGROUP_TARGET_THRESH:
752 next = val + THRESHOLDS_EVENTS_TARGET;
753 break;
754 case MEM_CGROUP_TARGET_SOFTLIMIT:
755 next = val + SOFTLIMIT_EVENTS_TARGET;
756 break;
757 case MEM_CGROUP_TARGET_NUMAINFO:
758 next = val + NUMAINFO_EVENTS_TARGET;
759 break;
760 default:
761 break;
762 }
763 __this_cpu_write(memcg->stat->targets[target], next);
764 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700765 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800766 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800767}
768
769/*
770 * Check events in order.
771 *
772 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700773static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800774{
Steven Rostedt47994012011-11-02 13:38:33 -0700775 preempt_disable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800776 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800777 if (unlikely(mem_cgroup_event_ratelimit(memcg,
778 MEM_CGROUP_TARGET_THRESH))) {
779 bool do_softlimit, do_numainfo;
780
781 do_softlimit = mem_cgroup_event_ratelimit(memcg,
782 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700783#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800784 do_numainfo = mem_cgroup_event_ratelimit(memcg,
785 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700786#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800787 preempt_enable();
788
789 mem_cgroup_threshold(memcg);
790 if (unlikely(do_softlimit))
791 mem_cgroup_update_tree(memcg, page);
792#if MAX_NUMNODES > 1
793 if (unlikely(do_numainfo))
794 atomic_inc(&memcg->numainfo_events);
795#endif
796 } else
797 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800798}
799
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000800struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800801{
802 return container_of(cgroup_subsys_state(cont,
803 mem_cgroup_subsys_id), struct mem_cgroup,
804 css);
805}
806
Balbir Singhcf475ad2008-04-29 01:00:16 -0700807struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800808{
Balbir Singh31a78f22008-09-28 23:09:31 +0100809 /*
810 * mm_update_next_owner() may clear mm->owner to NULL
811 * if it races with swapoff, page migration, etc.
812 * So this can be called with p == NULL.
813 */
814 if (unlikely(!p))
815 return NULL;
816
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800817 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
818 struct mem_cgroup, css);
819}
820
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700821struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800822{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700823 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700824
825 if (!mm)
826 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800827 /*
828 * Because we have no locks, mm->owner's may be being moved to other
829 * cgroup. We use css_tryget() here even if this looks
830 * pessimistic (rather than adding locks here).
831 */
832 rcu_read_lock();
833 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700834 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
835 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800836 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700837 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800838 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700839 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800840}
841
Johannes Weiner56600482012-01-12 17:17:59 -0800842/**
843 * mem_cgroup_iter - iterate over memory cgroup hierarchy
844 * @root: hierarchy root
845 * @prev: previously returned memcg, NULL on first invocation
846 * @reclaim: cookie for shared reclaim walks, NULL for full walks
847 *
848 * Returns references to children of the hierarchy below @root, or
849 * @root itself, or %NULL after a full round-trip.
850 *
851 * Caller must pass the return value in @prev on subsequent
852 * invocations for reference counting, or use mem_cgroup_iter_break()
853 * to cancel a hierarchy walk before the round-trip is complete.
854 *
855 * Reclaimers can specify a zone and a priority level in @reclaim to
856 * divide up the memcgs in the hierarchy among all concurrent
857 * reclaimers operating on the same zone and priority.
858 */
859struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
860 struct mem_cgroup *prev,
861 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700862{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800863 struct mem_cgroup *memcg = NULL;
864 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700865
Johannes Weiner56600482012-01-12 17:17:59 -0800866 if (mem_cgroup_disabled())
867 return NULL;
868
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700869 if (!root)
870 root = root_mem_cgroup;
871
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800872 if (prev && !reclaim)
873 id = css_id(&prev->css);
874
875 if (prev && prev != root)
876 css_put(&prev->css);
877
878 if (!root->use_hierarchy && root != root_mem_cgroup) {
879 if (prev)
880 return NULL;
881 return root;
882 }
883
884 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800885 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800886 struct cgroup_subsys_state *css;
887
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800888 if (reclaim) {
889 int nid = zone_to_nid(reclaim->zone);
890 int zid = zone_idx(reclaim->zone);
891 struct mem_cgroup_per_zone *mz;
892
893 mz = mem_cgroup_zoneinfo(root, nid, zid);
894 iter = &mz->reclaim_iter[reclaim->priority];
895 if (prev && reclaim->generation != iter->generation)
896 return NULL;
897 id = iter->position;
898 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800899
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700900 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800901 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
902 if (css) {
903 if (css == &root->css || css_tryget(css))
904 memcg = container_of(css,
905 struct mem_cgroup, css);
906 } else
907 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700908 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700909
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800910 if (reclaim) {
911 iter->position = id;
912 if (!css)
913 iter->generation++;
914 else if (!prev && memcg)
915 reclaim->generation = iter->generation;
916 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800917
918 if (prev && !css)
919 return NULL;
920 }
921 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700922}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800923
Johannes Weiner56600482012-01-12 17:17:59 -0800924/**
925 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
926 * @root: hierarchy root
927 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
928 */
929void mem_cgroup_iter_break(struct mem_cgroup *root,
930 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800931{
932 if (!root)
933 root = root_mem_cgroup;
934 if (prev && prev != root)
935 css_put(&prev->css);
936}
937
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700938/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800939 * Iteration constructs for visiting all cgroups (under a tree). If
940 * loops are exited prematurely (break), mem_cgroup_iter_break() must
941 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700942 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800943#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800944 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800945 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800946 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700947
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800948#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800949 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800950 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800951 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700952
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700953static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700954{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700955 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700956}
957
Ying Han456f9982011-05-26 16:25:38 -0700958void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
959{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700960 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700961
962 if (!mm)
963 return;
964
965 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700966 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
967 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700968 goto out;
969
970 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -0700971 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -0800972 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
973 break;
974 case PGMAJFAULT:
975 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -0700976 break;
977 default:
978 BUG();
979 }
980out:
981 rcu_read_unlock();
982}
983EXPORT_SYMBOL(mem_cgroup_count_vm_event);
984
Johannes Weiner925b7672012-01-12 17:18:15 -0800985/**
986 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
987 * @zone: zone of the wanted lruvec
988 * @mem: memcg of the wanted lruvec
989 *
990 * Returns the lru list vector holding pages for the given @zone and
991 * @mem. This can be the global zone lruvec, if the memory controller
992 * is disabled.
993 */
994struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
995 struct mem_cgroup *memcg)
996{
997 struct mem_cgroup_per_zone *mz;
998
999 if (mem_cgroup_disabled())
1000 return &zone->lruvec;
1001
1002 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1003 return &mz->lruvec;
1004}
1005
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001006/*
1007 * Following LRU functions are allowed to be used without PCG_LOCK.
1008 * Operations are called by routine of global LRU independently from memcg.
1009 * What we have to take care of here is validness of pc->mem_cgroup.
1010 *
1011 * Changes to pc->mem_cgroup happens when
1012 * 1. charge
1013 * 2. moving account
1014 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1015 * It is added to LRU before charge.
1016 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1017 * When moving account, the page is not on LRU. It's isolated.
1018 */
1019
Johannes Weiner925b7672012-01-12 17:18:15 -08001020/**
1021 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1022 * @zone: zone of the page
1023 * @page: the page
1024 * @lru: current lru
1025 *
1026 * This function accounts for @page being added to @lru, and returns
1027 * the lruvec for the given @zone and the memcg @page is charged to.
1028 *
1029 * The callsite is then responsible for physically linking the page to
1030 * the returned lruvec->lists[@lru].
Minchan Kim3f58a822011-03-22 16:32:53 -07001031 */
Johannes Weiner925b7672012-01-12 17:18:15 -08001032struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page,
1033 enum lru_list lru)
Minchan Kim3f58a822011-03-22 16:32:53 -07001034{
1035 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001036 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001037 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001038
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001039 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001040 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001041
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001042 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001043 memcg = pc->mem_cgroup;
Johannes Weiner925b7672012-01-12 17:18:15 -08001044 mz = page_cgroup_zoneinfo(memcg, page);
1045 /* compound_order() is stabilized through lru_lock */
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08001046 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001047 return &mz->lruvec;
1048}
1049
1050/**
1051 * mem_cgroup_lru_del_list - account for removing an lru page
1052 * @page: the page
1053 * @lru: target lru
1054 *
1055 * This function accounts for @page being removed from @lru.
1056 *
1057 * The callsite is then responsible for physically unlinking
1058 * @page->lru.
1059 */
1060void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru)
1061{
1062 struct mem_cgroup_per_zone *mz;
1063 struct mem_cgroup *memcg;
1064 struct page_cgroup *pc;
1065
1066 if (mem_cgroup_disabled())
1067 return;
1068
1069 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001070 memcg = pc->mem_cgroup;
1071 VM_BUG_ON(!memcg);
Johannes Weiner925b7672012-01-12 17:18:15 -08001072 mz = page_cgroup_zoneinfo(memcg, page);
1073 /* huge page split is done under lru_lock. so, we have no races. */
Hugh Dickins12d27102012-01-12 17:19:52 -08001074 VM_BUG_ON(MEM_CGROUP_ZSTAT(mz, lru) < (1 << compound_order(page)));
Johannes Weiner925b7672012-01-12 17:18:15 -08001075 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
1076}
1077
1078void mem_cgroup_lru_del(struct page *page)
1079{
1080 mem_cgroup_lru_del_list(page, page_lru(page));
1081}
1082
1083/**
1084 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1085 * @zone: zone of the page
1086 * @page: the page
1087 * @from: current lru
1088 * @to: target lru
1089 *
1090 * This function accounts for @page being moved between the lrus @from
1091 * and @to, and returns the lruvec for the given @zone and the memcg
1092 * @page is charged to.
1093 *
1094 * The callsite is then responsible for physically relinking
1095 * @page->lru to the returned lruvec->lists[@to].
1096 */
1097struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone,
1098 struct page *page,
1099 enum lru_list from,
1100 enum lru_list to)
1101{
1102 /* XXX: Optimize this, especially for @from == @to */
1103 mem_cgroup_lru_del_list(page, from);
1104 return mem_cgroup_lru_add_list(zone, page, to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001105}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001106
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001107/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001108 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001109 * hierarchy subtree
1110 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001111static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1112 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001113{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001114 if (root_memcg != memcg) {
1115 return (root_memcg->use_hierarchy &&
1116 css_is_ancestor(&memcg->css, &root_memcg->css));
Michal Hocko3e920412011-07-26 16:08:29 -07001117 }
1118
1119 return true;
1120}
1121
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001122int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001123{
1124 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001125 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001126 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001127
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001128 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001129 if (p) {
1130 curr = try_get_mem_cgroup_from_mm(p->mm);
1131 task_unlock(p);
1132 } else {
1133 /*
1134 * All threads may have already detached their mm's, but the oom
1135 * killer still needs to detect if they have already been oom
1136 * killed to prevent needlessly killing additional tasks.
1137 */
1138 task_lock(task);
1139 curr = mem_cgroup_from_task(task);
1140 if (curr)
1141 css_get(&curr->css);
1142 task_unlock(task);
1143 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001144 if (!curr)
1145 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001146 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001147 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001148 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001149 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1150 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001151 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001152 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001153 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001154 return ret;
1155}
1156
Johannes Weiner9b272972011-11-02 13:38:23 -07001157int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001158{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001159 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001160 int nid = zone_to_nid(zone);
1161 int zid = zone_idx(zone);
1162 unsigned long inactive;
1163 unsigned long active;
1164 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001165
Johannes Weiner9b272972011-11-02 13:38:23 -07001166 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1167 BIT(LRU_INACTIVE_ANON));
1168 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1169 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001170
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001171 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1172 if (gb)
1173 inactive_ratio = int_sqrt(10 * gb);
1174 else
1175 inactive_ratio = 1;
1176
Johannes Weiner9b272972011-11-02 13:38:23 -07001177 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001178}
1179
Johannes Weiner9b272972011-11-02 13:38:23 -07001180int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001181{
1182 unsigned long active;
1183 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001184 int zid = zone_idx(zone);
1185 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001186
Johannes Weiner9b272972011-11-02 13:38:23 -07001187 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1188 BIT(LRU_INACTIVE_FILE));
1189 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1190 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001191
1192 return (active > inactive);
1193}
1194
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001195struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1196 struct zone *zone)
1197{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001198 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001199 int zid = zone_idx(zone);
1200 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1201
1202 return &mz->reclaim_stat;
1203}
1204
1205struct zone_reclaim_stat *
1206mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1207{
1208 struct page_cgroup *pc;
1209 struct mem_cgroup_per_zone *mz;
1210
1211 if (mem_cgroup_disabled())
1212 return NULL;
1213
1214 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001215 if (!PageCgroupUsed(pc))
1216 return NULL;
Johannes Weiner713735b42011-01-20 14:44:31 -08001217 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1218 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001219 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001220 return &mz->reclaim_stat;
1221}
1222
Balbir Singh6d61ef42009-01-07 18:08:06 -08001223#define mem_cgroup_from_res_counter(counter, member) \
1224 container_of(counter, struct mem_cgroup, member)
1225
Johannes Weiner19942822011-02-01 15:52:43 -08001226/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001227 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1228 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001229 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001230 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001231 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001232 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001233static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001234{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001235 unsigned long long margin;
1236
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001237 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001238 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001239 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001240 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001241}
1242
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001243int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001244{
1245 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001246
1247 /* root ? */
1248 if (cgrp->parent == NULL)
1249 return vm_swappiness;
1250
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001251 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001252}
1253
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001254static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001255{
1256 int cpu;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001257
1258 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001259 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001260 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001261 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1262 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1263 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001264 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001265
1266 synchronize_rcu();
1267}
1268
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001269static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001270{
1271 int cpu;
1272
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001273 if (!memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001274 return;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001275 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001276 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001277 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001278 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1279 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1280 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001281 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001282}
1283/*
1284 * 2 routines for checking "mem" is under move_account() or not.
1285 *
1286 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1287 * for avoiding race in accounting. If true,
1288 * pc->mem_cgroup may be overwritten.
1289 *
1290 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1291 * under hierarchy of moving cgroups. This is for
1292 * waiting at hith-memory prressure caused by "move".
1293 */
1294
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001295static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001296{
1297 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001298 return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001299}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001300
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001301static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001302{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001303 struct mem_cgroup *from;
1304 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001305 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001306 /*
1307 * Unlike task_move routines, we access mc.to, mc.from not under
1308 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1309 */
1310 spin_lock(&mc.lock);
1311 from = mc.from;
1312 to = mc.to;
1313 if (!from)
1314 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001315
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001316 ret = mem_cgroup_same_or_subtree(memcg, from)
1317 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001318unlock:
1319 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001320 return ret;
1321}
1322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001323static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001324{
1325 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001326 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001327 DEFINE_WAIT(wait);
1328 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1329 /* moving charge context might have finished. */
1330 if (mc.moving_task)
1331 schedule();
1332 finish_wait(&mc.waitq, &wait);
1333 return true;
1334 }
1335 }
1336 return false;
1337}
1338
Balbir Singhe2224322009-04-02 16:57:39 -07001339/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001340 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001341 * @memcg: The memory cgroup that went over limit
1342 * @p: Task that is going to be killed
1343 *
1344 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1345 * enabled
1346 */
1347void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1348{
1349 struct cgroup *task_cgrp;
1350 struct cgroup *mem_cgrp;
1351 /*
1352 * Need a buffer in BSS, can't rely on allocations. The code relies
1353 * on the assumption that OOM is serialized for memory controller.
1354 * If this assumption is broken, revisit this code.
1355 */
1356 static char memcg_name[PATH_MAX];
1357 int ret;
1358
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001359 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001360 return;
1361
1362
1363 rcu_read_lock();
1364
1365 mem_cgrp = memcg->css.cgroup;
1366 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1367
1368 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1369 if (ret < 0) {
1370 /*
1371 * Unfortunately, we are unable to convert to a useful name
1372 * But we'll still print out the usage information
1373 */
1374 rcu_read_unlock();
1375 goto done;
1376 }
1377 rcu_read_unlock();
1378
1379 printk(KERN_INFO "Task in %s killed", memcg_name);
1380
1381 rcu_read_lock();
1382 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1383 if (ret < 0) {
1384 rcu_read_unlock();
1385 goto done;
1386 }
1387 rcu_read_unlock();
1388
1389 /*
1390 * Continues from above, so we don't need an KERN_ level
1391 */
1392 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1393done:
1394
1395 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1396 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1397 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1398 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1399 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1400 "failcnt %llu\n",
1401 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1402 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1403 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1404}
1405
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001406/*
1407 * This function returns the number of memcg under hierarchy tree. Returns
1408 * 1(self count) if no children.
1409 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001410static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001411{
1412 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001413 struct mem_cgroup *iter;
1414
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001415 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001416 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001417 return num;
1418}
1419
Balbir Singh6d61ef42009-01-07 18:08:06 -08001420/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001421 * Return the memory (and swap, if configured) limit for a memcg.
1422 */
1423u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1424{
1425 u64 limit;
1426 u64 memsw;
1427
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001428 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1429 limit += total_swap_pages << PAGE_SHIFT;
1430
David Rientjesa63d83f2010-08-09 17:19:46 -07001431 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1432 /*
1433 * If memsw is finite and limits the amount of swap space available
1434 * to this memcg, return that limit.
1435 */
1436 return min(limit, memsw);
1437}
1438
Johannes Weiner56600482012-01-12 17:17:59 -08001439static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1440 gfp_t gfp_mask,
1441 unsigned long flags)
1442{
1443 unsigned long total = 0;
1444 bool noswap = false;
1445 int loop;
1446
1447 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1448 noswap = true;
1449 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1450 noswap = true;
1451
1452 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1453 if (loop)
1454 drain_all_stock_async(memcg);
1455 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1456 /*
1457 * Allow limit shrinkers, which are triggered directly
1458 * by userspace, to catch signals and stop reclaim
1459 * after minimal progress, regardless of the margin.
1460 */
1461 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1462 break;
1463 if (mem_cgroup_margin(memcg))
1464 break;
1465 /*
1466 * If nothing was reclaimed after two attempts, there
1467 * may be no reclaimable pages in this hierarchy.
1468 */
1469 if (loop && !total)
1470 break;
1471 }
1472 return total;
1473}
1474
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001475/**
1476 * test_mem_cgroup_node_reclaimable
1477 * @mem: the target memcg
1478 * @nid: the node ID to be checked.
1479 * @noswap : specify true here if the user wants flle only information.
1480 *
1481 * This function returns whether the specified memcg contains any
1482 * reclaimable pages on a node. Returns true if there are any reclaimable
1483 * pages in the node.
1484 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001485static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001486 int nid, bool noswap)
1487{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001488 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001489 return true;
1490 if (noswap || !total_swap_pages)
1491 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001492 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001493 return true;
1494 return false;
1495
1496}
Ying Han889976d2011-05-26 16:25:33 -07001497#if MAX_NUMNODES > 1
1498
1499/*
1500 * Always updating the nodemask is not very good - even if we have an empty
1501 * list or the wrong list here, we can start from some node and traverse all
1502 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1503 *
1504 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001505static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001506{
1507 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001508 /*
1509 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1510 * pagein/pageout changes since the last update.
1511 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001512 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001513 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001514 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001515 return;
1516
Ying Han889976d2011-05-26 16:25:33 -07001517 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001518 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001519
1520 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1521
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001522 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1523 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001524 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001525
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001526 atomic_set(&memcg->numainfo_events, 0);
1527 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001528}
1529
1530/*
1531 * Selecting a node where we start reclaim from. Because what we need is just
1532 * reducing usage counter, start from anywhere is O,K. Considering
1533 * memory reclaim from current node, there are pros. and cons.
1534 *
1535 * Freeing memory from current node means freeing memory from a node which
1536 * we'll use or we've used. So, it may make LRU bad. And if several threads
1537 * hit limits, it will see a contention on a node. But freeing from remote
1538 * node means more costs for memory reclaim because of memory latency.
1539 *
1540 * Now, we use round-robin. Better algorithm is welcomed.
1541 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001542int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001543{
1544 int node;
1545
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001546 mem_cgroup_may_update_nodemask(memcg);
1547 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001548
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001549 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001550 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001551 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001552 /*
1553 * We call this when we hit limit, not when pages are added to LRU.
1554 * No LRU may hold pages because all pages are UNEVICTABLE or
1555 * memcg is too small and all pages are not on LRU. In that case,
1556 * we use curret node.
1557 */
1558 if (unlikely(node == MAX_NUMNODES))
1559 node = numa_node_id();
1560
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001561 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001562 return node;
1563}
1564
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001565/*
1566 * Check all nodes whether it contains reclaimable pages or not.
1567 * For quick scan, we make use of scan_nodes. This will allow us to skip
1568 * unused nodes. But scan_nodes is lazily updated and may not cotain
1569 * enough new information. We need to do double check.
1570 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001571bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001572{
1573 int nid;
1574
1575 /*
1576 * quick check...making use of scan_node.
1577 * We can skip unused nodes.
1578 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001579 if (!nodes_empty(memcg->scan_nodes)) {
1580 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001581 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001582 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001583
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001584 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001585 return true;
1586 }
1587 }
1588 /*
1589 * Check rest of nodes.
1590 */
1591 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001592 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001593 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001594 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001595 return true;
1596 }
1597 return false;
1598}
1599
Ying Han889976d2011-05-26 16:25:33 -07001600#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001601int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001602{
1603 return 0;
1604}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001605
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001606bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001607{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001608 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001609}
Ying Han889976d2011-05-26 16:25:33 -07001610#endif
1611
Johannes Weiner56600482012-01-12 17:17:59 -08001612static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1613 struct zone *zone,
1614 gfp_t gfp_mask,
1615 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001616{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001617 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001618 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001619 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001620 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001621 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001622 struct mem_cgroup_reclaim_cookie reclaim = {
1623 .zone = zone,
1624 .priority = 0,
1625 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001626
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001627 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001628
Balbir Singh4e416952009-09-23 15:56:39 -07001629 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001630 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001631 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001632 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001633 if (loop >= 2) {
1634 /*
1635 * If we have not been able to reclaim
1636 * anything, it might because there are
1637 * no reclaimable pages under this hierarchy
1638 */
Johannes Weiner56600482012-01-12 17:17:59 -08001639 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001640 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001641 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001642 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001643 * excess >> 2 is not to excessive so as to
1644 * reclaim too much, nor too less that we keep
1645 * coming back to reclaim from this cgroup
1646 */
1647 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001648 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001649 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001650 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001651 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001652 }
Johannes Weiner56600482012-01-12 17:17:59 -08001653 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001654 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001655 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1656 zone, &nr_scanned);
1657 *total_scanned += nr_scanned;
1658 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001659 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001660 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001661 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001662 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001663}
1664
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001665/*
1666 * Check OOM-Killer is already running under our hierarchy.
1667 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001668 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001669 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001670static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001671{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001672 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001673
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001674 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001675 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001676 /*
1677 * this subtree of our hierarchy is already locked
1678 * so we cannot give a lock.
1679 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001680 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001681 mem_cgroup_iter_break(memcg, iter);
1682 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001683 } else
1684 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001685 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001686
Michal Hocko79dfdac2011-07-26 16:08:23 -07001687 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001688 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001689
1690 /*
1691 * OK, we failed to lock the whole subtree so we have to clean up
1692 * what we set up to the failing subtree
1693 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001694 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001695 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001696 mem_cgroup_iter_break(memcg, iter);
1697 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001698 }
1699 iter->oom_lock = false;
1700 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001701 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001702}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001703
Michal Hocko79dfdac2011-07-26 16:08:23 -07001704/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001705 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001706 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001707static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001708{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001709 struct mem_cgroup *iter;
1710
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001711 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001712 iter->oom_lock = false;
1713 return 0;
1714}
1715
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001716static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001717{
1718 struct mem_cgroup *iter;
1719
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001720 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001721 atomic_inc(&iter->under_oom);
1722}
1723
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001724static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001725{
1726 struct mem_cgroup *iter;
1727
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001728 /*
1729 * When a new child is created while the hierarchy is under oom,
1730 * mem_cgroup_oom_lock() may not be called. We have to use
1731 * atomic_add_unless() here.
1732 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001733 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001734 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001735}
1736
Michal Hocko1af8efe2011-07-26 16:08:24 -07001737static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001738static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1739
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001740struct oom_wait_info {
1741 struct mem_cgroup *mem;
1742 wait_queue_t wait;
1743};
1744
1745static int memcg_oom_wake_function(wait_queue_t *wait,
1746 unsigned mode, int sync, void *arg)
1747{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001748 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg,
1749 *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001750 struct oom_wait_info *oom_wait_info;
1751
1752 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001753 oom_wait_memcg = oom_wait_info->mem;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001754
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001755 /*
1756 * Both of oom_wait_info->mem and wake_mem are stable under us.
1757 * Then we can use css_is_ancestor without taking care of RCU.
1758 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001759 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1760 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001761 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001762 return autoremove_wake_function(wait, mode, sync, arg);
1763}
1764
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001765static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001766{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001767 /* for filtering, pass "memcg" as argument. */
1768 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001769}
1770
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001771static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001772{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001773 if (memcg && atomic_read(&memcg->under_oom))
1774 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001775}
1776
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001777/*
1778 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1779 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001780bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001781{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001782 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001783 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001784
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001785 owait.mem = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001786 owait.wait.flags = 0;
1787 owait.wait.func = memcg_oom_wake_function;
1788 owait.wait.private = current;
1789 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001790 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001791 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001792
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001793 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001794 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001795 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001796 /*
1797 * Even if signal_pending(), we can't quit charge() loop without
1798 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1799 * under OOM is always welcomed, use TASK_KILLABLE here.
1800 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001801 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001803 need_to_kill = false;
1804 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001805 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001806 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001807
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001808 if (need_to_kill) {
1809 finish_wait(&memcg_oom_waitq, &owait.wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001810 mem_cgroup_out_of_memory(memcg, mask);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001811 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001812 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001813 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001814 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001815 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001816 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001817 mem_cgroup_oom_unlock(memcg);
1818 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001819 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001820
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001821 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001822
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001823 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1824 return false;
1825 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001826 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001827 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001828}
1829
Balbir Singhd69b0422009-06-17 16:26:34 -07001830/*
1831 * Currently used to update mapped file statistics, but the routine can be
1832 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001833 *
1834 * Notes: Race condition
1835 *
1836 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1837 * it tends to be costly. But considering some conditions, we doesn't need
1838 * to do so _always_.
1839 *
1840 * Considering "charge", lock_page_cgroup() is not required because all
1841 * file-stat operations happen after a page is attached to radix-tree. There
1842 * are no race with "charge".
1843 *
1844 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1845 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1846 * if there are race with "uncharge". Statistics itself is properly handled
1847 * by flags.
1848 *
1849 * Considering "move", this is an only case we see a race. To make the race
1850 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1851 * possibility of race condition. If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001852 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001853
Greg Thelen2a7106f2011-01-13 15:47:37 -08001854void mem_cgroup_update_page_stat(struct page *page,
1855 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001856{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001857 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001858 struct page_cgroup *pc = lookup_page_cgroup(page);
1859 bool need_unlock = false;
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001860 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001861
Johannes Weinercfa44942012-01-12 17:18:38 -08001862 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001863 return;
1864
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001865 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001866 memcg = pc->mem_cgroup;
1867 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001868 goto out;
1869 /* pc->mem_cgroup is unstable ? */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001870 if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) {
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001871 /* take a lock against to access pc->mem_cgroup */
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001872 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001873 need_unlock = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001874 memcg = pc->mem_cgroup;
1875 if (!memcg || !PageCgroupUsed(pc))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001876 goto out;
1877 }
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001878
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001879 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001880 case MEMCG_NR_FILE_MAPPED:
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001881 if (val > 0)
1882 SetPageCgroupFileMapped(pc);
1883 else if (!page_mapped(page))
KAMEZAWA Hiroyuki0c270f82010-10-27 15:33:39 -07001884 ClearPageCgroupFileMapped(pc);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001885 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001886 break;
1887 default:
1888 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001889 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001890
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001891 this_cpu_add(memcg->stat->count[idx], val);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001892
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001893out:
1894 if (unlikely(need_unlock))
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001895 move_unlock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001896 rcu_read_unlock();
1897 return;
Balbir Singhd69b0422009-06-17 16:26:34 -07001898}
Greg Thelen2a7106f2011-01-13 15:47:37 -08001899EXPORT_SYMBOL(mem_cgroup_update_page_stat);
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001900
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001901/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001902 * size of first charge trial. "32" comes from vmscan.c's magic value.
1903 * TODO: maybe necessary to use big numbers in big irons.
1904 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001905#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001906struct memcg_stock_pcp {
1907 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001908 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001909 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001910 unsigned long flags;
1911#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001912};
1913static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001914static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001915
1916/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001917 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001918 * from local stock and true is returned. If the stock is 0 or charges from a
1919 * cgroup which is not current target, returns false. This stock will be
1920 * refilled.
1921 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001922static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001923{
1924 struct memcg_stock_pcp *stock;
1925 bool ret = true;
1926
1927 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001928 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001929 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001930 else /* need to call res_counter_charge */
1931 ret = false;
1932 put_cpu_var(memcg_stock);
1933 return ret;
1934}
1935
1936/*
1937 * Returns stocks cached in percpu to res_counter and reset cached information.
1938 */
1939static void drain_stock(struct memcg_stock_pcp *stock)
1940{
1941 struct mem_cgroup *old = stock->cached;
1942
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001943 if (stock->nr_pages) {
1944 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
1945
1946 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001947 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001948 res_counter_uncharge(&old->memsw, bytes);
1949 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001950 }
1951 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001952}
1953
1954/*
1955 * This must be called under preempt disabled or must be called by
1956 * a thread which is pinned to local cpu.
1957 */
1958static void drain_local_stock(struct work_struct *dummy)
1959{
1960 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1961 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001962 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001963}
1964
1965/*
1966 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001967 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001968 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001969static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001970{
1971 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1972
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001973 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001974 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001975 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001976 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001977 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001978 put_cpu_var(memcg_stock);
1979}
1980
1981/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001982 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07001983 * of the hierarchy under it. sync flag says whether we should block
1984 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001985 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001986static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001987{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001988 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001989
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001990 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001991 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001992 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001993 for_each_online_cpu(cpu) {
1994 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001995 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001996
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001997 memcg = stock->cached;
1998 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001999 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002000 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002001 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002002 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2003 if (cpu == curcpu)
2004 drain_local_stock(&stock->work);
2005 else
2006 schedule_work_on(cpu, &stock->work);
2007 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002008 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002009 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002010
2011 if (!sync)
2012 goto out;
2013
2014 for_each_online_cpu(cpu) {
2015 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002016 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002017 flush_work(&stock->work);
2018 }
2019out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002020 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002021}
2022
2023/*
2024 * Tries to drain stocked charges in other cpus. This function is asynchronous
2025 * and just put a work per cpu for draining localy on each cpu. Caller can
2026 * expects some charges will be back to res_counter later but cannot wait for
2027 * it.
2028 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002029static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002030{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002031 /*
2032 * If someone calls draining, avoid adding more kworker runs.
2033 */
2034 if (!mutex_trylock(&percpu_charge_mutex))
2035 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002036 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002037 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002038}
2039
2040/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002041static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002042{
2043 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002044 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002045 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002046 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002047}
2048
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002049/*
2050 * This function drains percpu counter value from DEAD cpu and
2051 * move it to local cpu. Note that this function can be preempted.
2052 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002053static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002054{
2055 int i;
2056
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002057 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002058 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002059 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002060
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002061 per_cpu(memcg->stat->count[i], cpu) = 0;
2062 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002063 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002064 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002065 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002066
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002067 per_cpu(memcg->stat->events[i], cpu) = 0;
2068 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002069 }
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002070 /* need to clear ON_MOVE value, works as a kind of lock. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002071 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
2072 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002073}
2074
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002075static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002076{
2077 int idx = MEM_CGROUP_ON_MOVE;
2078
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002079 spin_lock(&memcg->pcp_counter_lock);
2080 per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx];
2081 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002082}
2083
2084static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002085 unsigned long action,
2086 void *hcpu)
2087{
2088 int cpu = (unsigned long)hcpu;
2089 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002090 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002091
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002092 if ((action == CPU_ONLINE)) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002093 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002094 synchronize_mem_cgroup_on_move(iter, cpu);
2095 return NOTIFY_OK;
2096 }
2097
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002098 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002099 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002100
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002101 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002102 mem_cgroup_drain_pcp_counter(iter, cpu);
2103
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002104 stock = &per_cpu(memcg_stock, cpu);
2105 drain_stock(stock);
2106 return NOTIFY_OK;
2107}
2108
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002109
2110/* See __mem_cgroup_try_charge() for details */
2111enum {
2112 CHARGE_OK, /* success */
2113 CHARGE_RETRY, /* need to retry but retry is not bad */
2114 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2115 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2116 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2117};
2118
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002119static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002120 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002121{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002122 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002123 struct mem_cgroup *mem_over_limit;
2124 struct res_counter *fail_res;
2125 unsigned long flags = 0;
2126 int ret;
2127
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002128 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002129
2130 if (likely(!ret)) {
2131 if (!do_swap_account)
2132 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002133 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002134 if (likely(!ret))
2135 return CHARGE_OK;
2136
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002137 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002138 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2139 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2140 } else
2141 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002142 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002143 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2144 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002145 *
2146 * Never reclaim on behalf of optional batching, retry with a
2147 * single page instead.
2148 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002149 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002150 return CHARGE_RETRY;
2151
2152 if (!(gfp_mask & __GFP_WAIT))
2153 return CHARGE_WOULDBLOCK;
2154
Johannes Weiner56600482012-01-12 17:17:59 -08002155 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002156 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002157 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002158 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002159 * Even though the limit is exceeded at this point, reclaim
2160 * may have been able to free some pages. Retry the charge
2161 * before killing the task.
2162 *
2163 * Only for regular pages, though: huge pages are rather
2164 * unlikely to succeed so close to the limit, and we fall back
2165 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002166 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002167 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002168 return CHARGE_RETRY;
2169
2170 /*
2171 * At task move, charge accounts can be doubly counted. So, it's
2172 * better to wait until the end of task_move if something is going on.
2173 */
2174 if (mem_cgroup_wait_acct_move(mem_over_limit))
2175 return CHARGE_RETRY;
2176
2177 /* If we don't need to call oom-killer at el, return immediately */
2178 if (!oom_check)
2179 return CHARGE_NOMEM;
2180 /* check OOM */
2181 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
2182 return CHARGE_OOM_DIE;
2183
2184 return CHARGE_RETRY;
2185}
2186
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002187/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002188 * __mem_cgroup_try_charge() does
2189 * 1. detect memcg to be charged against from passed *mm and *ptr,
2190 * 2. update res_counter
2191 * 3. call memory reclaim if necessary.
2192 *
2193 * In some special case, if the task is fatal, fatal_signal_pending() or
2194 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2195 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2196 * as possible without any hazards. 2: all pages should have a valid
2197 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2198 * pointer, that is treated as a charge to root_mem_cgroup.
2199 *
2200 * So __mem_cgroup_try_charge() will return
2201 * 0 ... on success, filling *ptr with a valid memcg pointer.
2202 * -ENOMEM ... charge failure because of resource limits.
2203 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2204 *
2205 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2206 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002207 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002208static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002209 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002210 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002211 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002212 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002213{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002214 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002215 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002216 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002217 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002218
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002219 /*
2220 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2221 * in system level. So, allow to go ahead dying process in addition to
2222 * MEMDIE process.
2223 */
2224 if (unlikely(test_thread_flag(TIF_MEMDIE)
2225 || fatal_signal_pending(current)))
2226 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002227
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002228 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002229 * We always charge the cgroup the mm_struct belongs to.
2230 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002231 * thread group leader migrates. It's possible that mm is not
2232 * set, if so charge the init_mm (happens for pagecache usage).
2233 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002234 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002235 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002236again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002237 if (*ptr) { /* css should be a valid one */
2238 memcg = *ptr;
2239 VM_BUG_ON(css_is_removed(&memcg->css));
2240 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002241 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002242 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002243 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002244 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002245 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002246 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002247
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002248 rcu_read_lock();
2249 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002250 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002251 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002252 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002253 * race with swapoff. Then, we have small risk of mis-accouning.
2254 * But such kind of mis-account by race always happens because
2255 * we don't have cgroup_mutex(). It's overkill and we allo that
2256 * small race, here.
2257 * (*) swapoff at el will charge against mm-struct not against
2258 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002259 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002260 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002261 if (!memcg)
2262 memcg = root_mem_cgroup;
2263 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002264 rcu_read_unlock();
2265 goto done;
2266 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002267 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002268 /*
2269 * It seems dagerous to access memcg without css_get().
2270 * But considering how consume_stok works, it's not
2271 * necessary. If consume_stock success, some charges
2272 * from this memcg are cached on this cpu. So, we
2273 * don't need to call css_get()/css_tryget() before
2274 * calling consume_stock().
2275 */
2276 rcu_read_unlock();
2277 goto done;
2278 }
2279 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002280 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002281 rcu_read_unlock();
2282 goto again;
2283 }
2284 rcu_read_unlock();
2285 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002286
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002287 do {
2288 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002289
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002290 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002291 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002292 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002293 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002294 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002295
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002296 oom_check = false;
2297 if (oom && !nr_oom_retries) {
2298 oom_check = true;
2299 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2300 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002301
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002302 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002303 switch (ret) {
2304 case CHARGE_OK:
2305 break;
2306 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002307 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002308 css_put(&memcg->css);
2309 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002310 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002311 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002312 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002313 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002314 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002315 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002316 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002317 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002318 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002319 /* If oom, we never return -ENOMEM */
2320 nr_oom_retries--;
2321 break;
2322 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002323 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002324 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002325 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002326 } while (ret != CHARGE_OK);
2327
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002328 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002329 refill_stock(memcg, batch - nr_pages);
2330 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002331done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002332 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002333 return 0;
2334nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002335 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002336 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002337bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002338 *ptr = root_mem_cgroup;
2339 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002340}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002341
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002342/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002343 * Somemtimes we have to undo a charge we got by try_charge().
2344 * This function is for that and do uncharge, put css's refcnt.
2345 * gotten by try_charge().
2346 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002347static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002348 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002349{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002350 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002351 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002352
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002353 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002354 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002355 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002356 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002357}
2358
2359/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002360 * A helper function to get mem_cgroup from ID. must be called under
2361 * rcu_read_lock(). The caller must check css_is_removed() or some if
2362 * it's concern. (dropping refcnt from swap can be called against removed
2363 * memcg.)
2364 */
2365static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2366{
2367 struct cgroup_subsys_state *css;
2368
2369 /* ID 0 is unused ID */
2370 if (!id)
2371 return NULL;
2372 css = css_lookup(&mem_cgroup_subsys, id);
2373 if (!css)
2374 return NULL;
2375 return container_of(css, struct mem_cgroup, css);
2376}
2377
Wu Fengguange42d9d52009-12-16 12:19:59 +01002378struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002379{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002380 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002381 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002382 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002383 swp_entry_t ent;
2384
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002385 VM_BUG_ON(!PageLocked(page));
2386
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002387 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002388 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002389 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002390 memcg = pc->mem_cgroup;
2391 if (memcg && !css_tryget(&memcg->css))
2392 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002393 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002394 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002395 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002396 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002397 memcg = mem_cgroup_lookup(id);
2398 if (memcg && !css_tryget(&memcg->css))
2399 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002400 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002401 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002402 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002403 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002404}
2405
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002406static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002407 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002408 unsigned int nr_pages,
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002409 struct page_cgroup *pc,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002410 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002411{
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002412 lock_page_cgroup(pc);
2413 if (unlikely(PageCgroupUsed(pc))) {
2414 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002415 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002416 return;
2417 }
2418 /*
2419 * we don't need page_cgroup_lock about tail pages, becase they are not
2420 * accessed by any other context at this point.
2421 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002422 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002423 /*
2424 * We access a page_cgroup asynchronously without lock_page_cgroup().
2425 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2426 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2427 * before USED bit, we need memory barrier here.
2428 * See mem_cgroup_add_lru_list(), etc.
2429 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002430 smp_wmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -07002431 switch (ctype) {
2432 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2433 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2434 SetPageCgroupCache(pc);
2435 SetPageCgroupUsed(pc);
2436 break;
2437 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2438 ClearPageCgroupCache(pc);
2439 SetPageCgroupUsed(pc);
2440 break;
2441 default:
2442 break;
2443 }
Hugh Dickins3be91272008-02-07 00:14:19 -08002444
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002445 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002446 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002447 WARN_ON_ONCE(PageLRU(page));
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002448 /*
2449 * "charge_statistics" updated event counter. Then, check it.
2450 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2451 * if they exceeds softlimit.
2452 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002453 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002454}
2455
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002456#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2457
2458#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002459 (1 << PCG_MIGRATION))
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002460/*
2461 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002462 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2463 * charge/uncharge will be never happen and move_account() is done under
2464 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002465 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002466void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002467{
2468 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002469 struct page_cgroup *pc;
2470 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002471
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002472 if (mem_cgroup_disabled())
2473 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002474 for (i = 1; i < HPAGE_PMD_NR; i++) {
2475 pc = head_pc + i;
2476 pc->mem_cgroup = head_pc->mem_cgroup;
2477 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002478 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2479 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002480}
Hugh Dickins12d27102012-01-12 17:19:52 -08002481#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002482
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002483/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002484 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002485 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002486 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002487 * @pc: page_cgroup of the page.
2488 * @from: mem_cgroup which the page is moved from.
2489 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002490 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002491 *
2492 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002493 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002494 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002495 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002496 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002497 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002498 * true, this function does "uncharge" from old cgroup, but it doesn't if
2499 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002500 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002501static int mem_cgroup_move_account(struct page *page,
2502 unsigned int nr_pages,
2503 struct page_cgroup *pc,
2504 struct mem_cgroup *from,
2505 struct mem_cgroup *to,
2506 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002507{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002508 unsigned long flags;
2509 int ret;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002510
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002511 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002512 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002513 /*
2514 * The page is isolated from LRU. So, collapse function
2515 * will not handle this page. But page splitting can happen.
2516 * Do this check under compound_page_lock(). The caller should
2517 * hold it.
2518 */
2519 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002520 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002521 goto out;
2522
2523 lock_page_cgroup(pc);
2524
2525 ret = -EINVAL;
2526 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2527 goto unlock;
2528
2529 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002530
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002531 if (PageCgroupFileMapped(pc)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002532 /* Update mapped_file data for mem_cgroup */
2533 preempt_disable();
2534 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2535 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2536 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002537 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002538 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002539 if (uncharge)
2540 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b8d2011-03-23 16:42:33 -07002541 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002542
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002543 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002544 pc->mem_cgroup = to;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002545 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002546 /*
2547 * We charges against "to" which may not have any tasks. Then, "to"
2548 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002549 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002550 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002551 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002552 */
Johannes Weinerde3638d2011-03-23 16:42:28 -07002553 move_unlock_page_cgroup(pc, &flags);
2554 ret = 0;
2555unlock:
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002556 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002557 /*
2558 * check events
2559 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002560 memcg_check_events(to, page);
2561 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002562out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002563 return ret;
2564}
2565
2566/*
2567 * move charges to its parent.
2568 */
2569
Johannes Weiner5564e882011-03-23 16:42:29 -07002570static int mem_cgroup_move_parent(struct page *page,
2571 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002572 struct mem_cgroup *child,
2573 gfp_t gfp_mask)
2574{
2575 struct cgroup *cg = child->css.cgroup;
2576 struct cgroup *pcg = cg->parent;
2577 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002578 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002579 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002580 int ret;
2581
2582 /* Is ROOT ? */
2583 if (!pcg)
2584 return -EINVAL;
2585
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002586 ret = -EBUSY;
2587 if (!get_page_unless_zero(page))
2588 goto out;
2589 if (isolate_lru_page(page))
2590 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002591
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002592 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002593
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002594 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002595 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002596 if (ret)
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002597 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002598
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002599 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002600 flags = compound_lock_irqsave(page);
2601
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002602 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002603 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002604 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002605
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002606 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002607 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002608put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002609 putback_lru_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002610put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002611 put_page(page);
Daisuke Nishimura57f9fd7d2009-12-15 16:47:11 -08002612out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002613 return ret;
2614}
2615
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002616/*
2617 * Charge the memory controller for page usage.
2618 * Return
2619 * 0 if the charge was successful
2620 * < 0 if the cgroup is over its limit
2621 */
2622static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002623 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002624{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002625 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002626 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002627 struct page_cgroup *pc;
2628 bool oom = true;
2629 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002630
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002631 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002632 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002633 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002634 /*
2635 * Never OOM-kill a process for a huge page. The
2636 * fault handler will fall back to regular pages.
2637 */
2638 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002639 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002640
2641 pc = lookup_page_cgroup(page);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002642 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002643 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002644 return ret;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002645 __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002646 return 0;
2647}
2648
2649int mem_cgroup_newpage_charge(struct page *page,
2650 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002651{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002652 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002653 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002654 VM_BUG_ON(page_mapped(page));
2655 VM_BUG_ON(page->mapping && !PageAnon(page));
2656 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002657 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002658 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002659}
2660
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002661static void
2662__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2663 enum charge_type ctype);
2664
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002665static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002666__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002667 enum charge_type ctype)
2668{
2669 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002670 struct zone *zone = page_zone(page);
2671 unsigned long flags;
2672 bool removed = false;
2673
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002674 /*
2675 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2676 * is already on LRU. It means the page may on some other page_cgroup's
2677 * LRU. Take care of it.
2678 */
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002679 spin_lock_irqsave(&zone->lru_lock, flags);
2680 if (PageLRU(page)) {
2681 del_page_from_lru_list(zone, page, page_lru(page));
2682 ClearPageLRU(page);
2683 removed = true;
2684 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002685 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002686 if (removed) {
2687 add_page_to_lru_list(zone, page, page_lru(page));
2688 SetPageLRU(page);
2689 }
2690 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002691 return;
2692}
2693
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002694int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2695 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002696{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002697 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002698 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002699 int ret;
2700
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002701 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002702 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002703 if (PageCompound(page))
2704 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002705
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002706 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002707 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002708 if (!page_is_file_cache(page))
2709 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002710
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002711 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002712 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002713 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002715 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002716 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2717 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002718 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002719}
2720
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002721/*
2722 * While swap-in, try_charge -> commit or cancel, the page is locked.
2723 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002724 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002725 * "commit()" or removed by "cancel()"
2726 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002727int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2728 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002729 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002730{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002731 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002732 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002733
Johannes Weiner72835c82012-01-12 17:18:32 -08002734 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002735
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002736 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002737 return 0;
2738
2739 if (!do_swap_account)
2740 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002741 /*
2742 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002743 * the pte, and even removed page from swap cache: in those cases
2744 * do_swap_page()'s pte_same() test will fail; but there's also a
2745 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002746 */
2747 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002748 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002749 memcg = try_get_mem_cgroup_from_page(page);
2750 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002751 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002752 *memcgp = memcg;
2753 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002754 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002755 if (ret == -EINTR)
2756 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002757 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002758charge_cur_mm:
2759 if (unlikely(!mm))
2760 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002761 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2762 if (ret == -EINTR)
2763 ret = 0;
2764 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002765}
2766
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002767static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002768__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002769 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002770{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002771 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002772 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002773 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002774 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002775 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002776
Johannes Weiner72835c82012-01-12 17:18:32 -08002777 __mem_cgroup_commit_charge_lrucare(page, memcg, ctype);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002778 /*
2779 * Now swap is on-memory. This means this page may be
2780 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002781 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2782 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2783 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002784 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002785 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002786 swp_entry_t ent = {.val = page_private(page)};
Johannes Weiner72835c82012-01-12 17:18:32 -08002787 struct mem_cgroup *swap_memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002788 unsigned short id;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002789
2790 id = swap_cgroup_record(ent, 0);
2791 rcu_read_lock();
Johannes Weiner72835c82012-01-12 17:18:32 -08002792 swap_memcg = mem_cgroup_lookup(id);
2793 if (swap_memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002794 /*
2795 * This recorded memcg can be obsolete one. So, avoid
2796 * calling css_tryget
2797 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002798 if (!mem_cgroup_is_root(swap_memcg))
2799 res_counter_uncharge(&swap_memcg->memsw,
2800 PAGE_SIZE);
2801 mem_cgroup_swap_statistics(swap_memcg, false);
2802 mem_cgroup_put(swap_memcg);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002803 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002804 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002805 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002806 /*
2807 * At swapin, we may charge account against cgroup which has no tasks.
2808 * So, rmdir()->pre_destroy() can be called while we do this charge.
2809 * In that case, we need to call pre_destroy() again. check it here.
2810 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002811 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002812}
2813
Johannes Weiner72835c82012-01-12 17:18:32 -08002814void mem_cgroup_commit_charge_swapin(struct page *page,
2815 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002816{
Johannes Weiner72835c82012-01-12 17:18:32 -08002817 __mem_cgroup_commit_charge_swapin(page, memcg,
2818 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002819}
2820
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002821void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002822{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002823 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002824 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002825 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002826 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002827 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002828}
2829
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002830static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002831 unsigned int nr_pages,
2832 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002833{
2834 struct memcg_batch_info *batch = NULL;
2835 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002836
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002837 /* If swapout, usage of swap doesn't decrease */
2838 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2839 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002840
2841 batch = &current->memcg_batch;
2842 /*
2843 * In usual, we do css_get() when we remember memcg pointer.
2844 * But in this case, we keep res->usage until end of a series of
2845 * uncharges. Then, it's ok to ignore memcg's refcnt.
2846 */
2847 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002848 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002849 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002850 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002851 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002852 * the same cgroup and we have chance to coalesce uncharges.
2853 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2854 * because we want to do uncharge as soon as possible.
2855 */
2856
2857 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2858 goto direct_uncharge;
2859
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002860 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002861 goto direct_uncharge;
2862
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002863 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002864 * In typical case, batch->memcg == mem. This means we can
2865 * merge a series of uncharges to an uncharge of res_counter.
2866 * If not, we uncharge res_counter ony by one.
2867 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002868 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002869 goto direct_uncharge;
2870 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002871 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002872 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002873 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002874 return;
2875direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002876 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002877 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002878 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2879 if (unlikely(batch->memcg != memcg))
2880 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002881 return;
2882}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002883
Balbir Singh8697d332008-02-07 00:13:59 -08002884/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002885 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002886 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002887static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002888__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002889{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002890 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002891 unsigned int nr_pages = 1;
2892 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002893
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002894 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002895 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002896
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002897 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002898 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002899
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002900 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002901 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002902 VM_BUG_ON(!PageTransHuge(page));
2903 }
Balbir Singh8697d332008-02-07 00:13:59 -08002904 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002905 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002906 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002907 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002908 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002909 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002910
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002911 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002912
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002913 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002914
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002915 if (!PageCgroupUsed(pc))
2916 goto unlock_out;
2917
2918 switch (ctype) {
2919 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002920 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07002921 /* See mem_cgroup_prepare_migration() */
2922 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002923 goto unlock_out;
2924 break;
2925 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2926 if (!PageAnon(page)) { /* Shared memory */
2927 if (page->mapping && !page_is_file_cache(page))
2928 goto unlock_out;
2929 } else if (page_mapped(page)) /* Anon */
2930 goto unlock_out;
2931 break;
2932 default:
2933 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002934 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002935
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002936 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002937
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002938 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002939 /*
2940 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2941 * freed from LRU. This is safe because uncharged page is expected not
2942 * to be reused (freed soon). Exception is SwapCache, it's handled by
2943 * special functions.
2944 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08002945
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002946 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002947 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002948 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002949 * will never be freed.
2950 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002951 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002952 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002953 mem_cgroup_swap_statistics(memcg, true);
2954 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002955 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002956 if (!mem_cgroup_is_root(memcg))
2957 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002958
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002959 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002960
2961unlock_out:
2962 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002963 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08002964}
2965
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002966void mem_cgroup_uncharge_page(struct page *page)
2967{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002968 /* early check. */
2969 if (page_mapped(page))
2970 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08002971 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002972 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2973}
2974
2975void mem_cgroup_uncharge_cache_page(struct page *page)
2976{
2977 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07002978 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002979 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2980}
2981
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002982/*
2983 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2984 * In that cases, pages are freed continuously and we can expect pages
2985 * are in the same memcg. All these calls itself limits the number of
2986 * pages freed at once, then uncharge_start/end() is called properly.
2987 * This may be called prural(2) times in a context,
2988 */
2989
2990void mem_cgroup_uncharge_start(void)
2991{
2992 current->memcg_batch.do_batch++;
2993 /* We can do nest. */
2994 if (current->memcg_batch.do_batch == 1) {
2995 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002996 current->memcg_batch.nr_pages = 0;
2997 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002998 }
2999}
3000
3001void mem_cgroup_uncharge_end(void)
3002{
3003 struct memcg_batch_info *batch = &current->memcg_batch;
3004
3005 if (!batch->do_batch)
3006 return;
3007
3008 batch->do_batch--;
3009 if (batch->do_batch) /* If stacked, do nothing. */
3010 return;
3011
3012 if (!batch->memcg)
3013 return;
3014 /*
3015 * This "batch->memcg" is valid without any css_get/put etc...
3016 * bacause we hide charges behind us.
3017 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003018 if (batch->nr_pages)
3019 res_counter_uncharge(&batch->memcg->res,
3020 batch->nr_pages * PAGE_SIZE);
3021 if (batch->memsw_nr_pages)
3022 res_counter_uncharge(&batch->memcg->memsw,
3023 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003024 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003025 /* forget this pointer (for sanity check) */
3026 batch->memcg = NULL;
3027}
3028
KAMEZAWA Hiroyuki4e5f01c2012-01-12 17:18:58 -08003029/*
3030 * A function for resetting pc->mem_cgroup for newly allocated pages.
3031 * This function should be called if the newpage will be added to LRU
3032 * before start accounting.
3033 */
3034void mem_cgroup_reset_owner(struct page *newpage)
3035{
3036 struct page_cgroup *pc;
3037
3038 if (mem_cgroup_disabled())
3039 return;
3040
3041 pc = lookup_page_cgroup(newpage);
3042 VM_BUG_ON(PageCgroupUsed(pc));
3043 pc->mem_cgroup = root_mem_cgroup;
3044}
3045
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003046#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003047/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003048 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003049 * memcg information is recorded to swap_cgroup of "ent"
3050 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003051void
3052mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003053{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003054 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003055 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003056
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003057 if (!swapout) /* this was a swap cache but the swap is unused ! */
3058 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3059
3060 memcg = __mem_cgroup_uncharge_common(page, ctype);
3061
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003062 /*
3063 * record memcg information, if swapout && memcg != NULL,
3064 * mem_cgroup_get() was called in uncharge().
3065 */
3066 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003067 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003068}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003069#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003070
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003071#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3072/*
3073 * called from swap_entry_free(). remove record in swap_cgroup and
3074 * uncharge "memsw" account.
3075 */
3076void mem_cgroup_uncharge_swap(swp_entry_t ent)
3077{
3078 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003079 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003080
3081 if (!do_swap_account)
3082 return;
3083
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003084 id = swap_cgroup_record(ent, 0);
3085 rcu_read_lock();
3086 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003087 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003088 /*
3089 * We uncharge this because swap is freed.
3090 * This memcg can be obsolete one. We avoid calling css_tryget
3091 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003092 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003093 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003094 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003095 mem_cgroup_put(memcg);
3096 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003097 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003098}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003099
3100/**
3101 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3102 * @entry: swap entry to be moved
3103 * @from: mem_cgroup which the entry is moved from
3104 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003105 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003106 *
3107 * It succeeds only when the swap_cgroup's record for this entry is the same
3108 * as the mem_cgroup's id of @from.
3109 *
3110 * Returns 0 on success, -EINVAL on failure.
3111 *
3112 * The caller must have charged to @to, IOW, called res_counter_charge() about
3113 * both res and memsw, and called css_get().
3114 */
3115static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003116 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003117{
3118 unsigned short old_id, new_id;
3119
3120 old_id = css_id(&from->css);
3121 new_id = css_id(&to->css);
3122
3123 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003124 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003125 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003126 /*
3127 * This function is only called from task migration context now.
3128 * It postpones res_counter and refcount handling till the end
3129 * of task migration(mem_cgroup_clear_mc()) for performance
3130 * improvement. But we cannot postpone mem_cgroup_get(to)
3131 * because if the process that has been moved to @to does
3132 * swap-in, the refcount of @to might be decreased to 0.
3133 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003134 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003135 if (need_fixup) {
3136 if (!mem_cgroup_is_root(from))
3137 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3138 mem_cgroup_put(from);
3139 /*
3140 * we charged both to->res and to->memsw, so we should
3141 * uncharge to->res.
3142 */
3143 if (!mem_cgroup_is_root(to))
3144 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003145 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003146 return 0;
3147 }
3148 return -EINVAL;
3149}
3150#else
3151static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003152 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003153{
3154 return -EINVAL;
3155}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003156#endif
3157
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003158/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003159 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3160 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003161 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003162int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003163 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003164{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003165 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003166 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003167 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003168 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003169
Johannes Weiner72835c82012-01-12 17:18:32 -08003170 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003171
Andrea Arcangeliec168512011-01-13 15:46:56 -08003172 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08003173 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003174 return 0;
3175
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003176 pc = lookup_page_cgroup(page);
3177 lock_page_cgroup(pc);
3178 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003179 memcg = pc->mem_cgroup;
3180 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003181 /*
3182 * At migrating an anonymous page, its mapcount goes down
3183 * to 0 and uncharge() will be called. But, even if it's fully
3184 * unmapped, migration may fail and this page has to be
3185 * charged again. We set MIGRATION flag here and delay uncharge
3186 * until end_migration() is called
3187 *
3188 * Corner Case Thinking
3189 * A)
3190 * When the old page was mapped as Anon and it's unmap-and-freed
3191 * while migration was ongoing.
3192 * If unmap finds the old page, uncharge() of it will be delayed
3193 * until end_migration(). If unmap finds a new page, it's
3194 * uncharged when it make mapcount to be 1->0. If unmap code
3195 * finds swap_migration_entry, the new page will not be mapped
3196 * and end_migration() will find it(mapcount==0).
3197 *
3198 * B)
3199 * When the old page was mapped but migraion fails, the kernel
3200 * remaps it. A charge for it is kept by MIGRATION flag even
3201 * if mapcount goes down to 0. We can do remap successfully
3202 * without charging it again.
3203 *
3204 * C)
3205 * The "old" page is under lock_page() until the end of
3206 * migration, so, the old page itself will not be swapped-out.
3207 * If the new page is swapped out before end_migraton, our
3208 * hook to usual swap-out path will catch the event.
3209 */
3210 if (PageAnon(page))
3211 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003212 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003213 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003214 /*
3215 * If the page is not charged at this point,
3216 * we return here.
3217 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003218 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003219 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003220
Johannes Weiner72835c82012-01-12 17:18:32 -08003221 *memcgp = memcg;
3222 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003223 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003224 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003225 if (PageAnon(page)) {
3226 lock_page_cgroup(pc);
3227 ClearPageCgroupMigration(pc);
3228 unlock_page_cgroup(pc);
3229 /*
3230 * The old page may be fully unmapped while we kept it.
3231 */
3232 mem_cgroup_uncharge_page(page);
3233 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003234 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003235 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003236 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003237 /*
3238 * We charge new page before it's used/mapped. So, even if unlock_page()
3239 * is called before end_migration, we can catch all events on this new
3240 * page. In the case new page is migrated but not remapped, new page's
3241 * mapcount will be finally 0 and we call uncharge in end_migration().
3242 */
3243 pc = lookup_page_cgroup(newpage);
3244 if (PageAnon(page))
3245 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3246 else if (page_is_file_cache(page))
3247 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3248 else
3249 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003250 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003251 return ret;
3252}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003253
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003254/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003255void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003256 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003257{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003258 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003259 struct page_cgroup *pc;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003260
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003261 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003262 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003263 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003264 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003265 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003266 used = oldpage;
3267 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003268 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003269 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003270 unused = oldpage;
3271 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003272 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003273 * We disallowed uncharge of pages under migration because mapcount
3274 * of the page goes down to zero, temporarly.
3275 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003276 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003277 pc = lookup_page_cgroup(oldpage);
3278 lock_page_cgroup(pc);
3279 ClearPageCgroupMigration(pc);
3280 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003281
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003282 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3283
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003284 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003285 * If a page is a file cache, radix-tree replacement is very atomic
3286 * and we can skip this check. When it was an Anon page, its mapcount
3287 * goes down to 0. But because we added MIGRATION flage, it's not
3288 * uncharged yet. There are several case but page->mapcount check
3289 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3290 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003291 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003292 if (PageAnon(used))
3293 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003294 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003295 * At migration, we may charge account against cgroup which has no
3296 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003297 * So, rmdir()->pre_destroy() can be called while we do this charge.
3298 * In that case, we need to call pre_destroy() again. check it here.
3299 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003300 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003301}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003302
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003303/*
3304 * At replace page cache, newpage is not under any memcg but it's on
3305 * LRU. So, this function doesn't touch res_counter but handles LRU
3306 * in correct way. Both pages are locked so we cannot race with uncharge.
3307 */
3308void mem_cgroup_replace_page_cache(struct page *oldpage,
3309 struct page *newpage)
3310{
3311 struct mem_cgroup *memcg;
3312 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003313 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003314
3315 if (mem_cgroup_disabled())
3316 return;
3317
3318 pc = lookup_page_cgroup(oldpage);
3319 /* fix accounting on old pages */
3320 lock_page_cgroup(pc);
3321 memcg = pc->mem_cgroup;
3322 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -1);
3323 ClearPageCgroupUsed(pc);
3324 unlock_page_cgroup(pc);
3325
3326 if (PageSwapBacked(oldpage))
3327 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3328
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003329 /*
3330 * Even if newpage->mapping was NULL before starting replacement,
3331 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3332 * LRU while we overwrite pc->mem_cgroup.
3333 */
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08003334 __mem_cgroup_commit_charge_lrucare(newpage, memcg, type);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003335}
3336
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003337#ifdef CONFIG_DEBUG_VM
3338static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3339{
3340 struct page_cgroup *pc;
3341
3342 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003343 /*
3344 * Can be NULL while feeding pages into the page allocator for
3345 * the first time, i.e. during boot or memory hotplug;
3346 * or when mem_cgroup_disabled().
3347 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003348 if (likely(pc) && PageCgroupUsed(pc))
3349 return pc;
3350 return NULL;
3351}
3352
3353bool mem_cgroup_bad_page_check(struct page *page)
3354{
3355 if (mem_cgroup_disabled())
3356 return false;
3357
3358 return lookup_page_cgroup_used(page) != NULL;
3359}
3360
3361void mem_cgroup_print_bad_page(struct page *page)
3362{
3363 struct page_cgroup *pc;
3364
3365 pc = lookup_page_cgroup_used(page);
3366 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003367 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003368 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003369 }
3370}
3371#endif
3372
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003373static DEFINE_MUTEX(set_limit_mutex);
3374
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003375static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003376 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003377{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003378 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003379 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003380 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003381 int children = mem_cgroup_count_children(memcg);
3382 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003383 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003384
3385 /*
3386 * For keeping hierarchical_reclaim simple, how long we should retry
3387 * is depends on callers. We set our retry-count to be function
3388 * of # of children which we should visit in this loop.
3389 */
3390 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3391
3392 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003393
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003394 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003395 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003396 if (signal_pending(current)) {
3397 ret = -EINTR;
3398 break;
3399 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003400 /*
3401 * Rather than hide all in some function, I do this in
3402 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003403 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003404 */
3405 mutex_lock(&set_limit_mutex);
3406 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3407 if (memswlimit < val) {
3408 ret = -EINVAL;
3409 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003410 break;
3411 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003412
3413 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3414 if (memlimit < val)
3415 enlarge = 1;
3416
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003417 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003418 if (!ret) {
3419 if (memswlimit == val)
3420 memcg->memsw_is_minimum = true;
3421 else
3422 memcg->memsw_is_minimum = false;
3423 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003424 mutex_unlock(&set_limit_mutex);
3425
3426 if (!ret)
3427 break;
3428
Johannes Weiner56600482012-01-12 17:17:59 -08003429 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3430 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003431 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3432 /* Usage is reduced ? */
3433 if (curusage >= oldusage)
3434 retry_count--;
3435 else
3436 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003437 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003438 if (!ret && enlarge)
3439 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003440
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003441 return ret;
3442}
3443
Li Zefan338c8432009-06-17 16:27:15 -07003444static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3445 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003446{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003447 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003448 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003449 int children = mem_cgroup_count_children(memcg);
3450 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003451 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003452
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003453 /* see mem_cgroup_resize_res_limit */
3454 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3455 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003456 while (retry_count) {
3457 if (signal_pending(current)) {
3458 ret = -EINTR;
3459 break;
3460 }
3461 /*
3462 * Rather than hide all in some function, I do this in
3463 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003464 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003465 */
3466 mutex_lock(&set_limit_mutex);
3467 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3468 if (memlimit > val) {
3469 ret = -EINVAL;
3470 mutex_unlock(&set_limit_mutex);
3471 break;
3472 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003473 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3474 if (memswlimit < val)
3475 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003476 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003477 if (!ret) {
3478 if (memlimit == val)
3479 memcg->memsw_is_minimum = true;
3480 else
3481 memcg->memsw_is_minimum = false;
3482 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003483 mutex_unlock(&set_limit_mutex);
3484
3485 if (!ret)
3486 break;
3487
Johannes Weiner56600482012-01-12 17:17:59 -08003488 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3489 MEM_CGROUP_RECLAIM_NOSWAP |
3490 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003491 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003492 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003493 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003494 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003495 else
3496 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003497 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003498 if (!ret && enlarge)
3499 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003500 return ret;
3501}
3502
Balbir Singh4e416952009-09-23 15:56:39 -07003503unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003504 gfp_t gfp_mask,
3505 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003506{
3507 unsigned long nr_reclaimed = 0;
3508 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3509 unsigned long reclaimed;
3510 int loop = 0;
3511 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003512 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003513 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003514
3515 if (order > 0)
3516 return 0;
3517
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003518 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003519 /*
3520 * This loop can run a while, specially if mem_cgroup's continuously
3521 * keep exceeding their soft limit and putting the system under
3522 * pressure
3523 */
3524 do {
3525 if (next_mz)
3526 mz = next_mz;
3527 else
3528 mz = mem_cgroup_largest_soft_limit_node(mctz);
3529 if (!mz)
3530 break;
3531
Ying Han0ae5e892011-05-26 16:25:25 -07003532 nr_scanned = 0;
Johannes Weiner56600482012-01-12 17:17:59 -08003533 reclaimed = mem_cgroup_soft_reclaim(mz->mem, zone,
3534 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003535 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003536 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003537 spin_lock(&mctz->lock);
3538
3539 /*
3540 * If we failed to reclaim anything from this memory cgroup
3541 * it is time to move on to the next cgroup
3542 */
3543 next_mz = NULL;
3544 if (!reclaimed) {
3545 do {
3546 /*
3547 * Loop until we find yet another one.
3548 *
3549 * By the time we get the soft_limit lock
3550 * again, someone might have aded the
3551 * group back on the RB tree. Iterate to
3552 * make sure we get a different mem.
3553 * mem_cgroup_largest_soft_limit_node returns
3554 * NULL if no other cgroup is present on
3555 * the tree
3556 */
3557 next_mz =
3558 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003559 if (next_mz == mz)
Balbir Singh4e416952009-09-23 15:56:39 -07003560 css_put(&next_mz->mem->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003561 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003562 break;
3563 } while (1);
3564 }
Balbir Singh4e416952009-09-23 15:56:39 -07003565 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003566 excess = res_counter_soft_limit_excess(&mz->mem->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003567 /*
3568 * One school of thought says that we should not add
3569 * back the node to the tree if reclaim returns 0.
3570 * But our reclaim could return 0, simply because due
3571 * to priority we are exposing a smaller subset of
3572 * memory to reclaim from. Consider this as a longer
3573 * term TODO.
3574 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003575 /* If excess == 0, no tree ops */
3576 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003577 spin_unlock(&mctz->lock);
3578 css_put(&mz->mem->css);
3579 loop++;
3580 /*
3581 * Could not reclaim anything and there are no more
3582 * mem cgroups to try or we seem to be looping without
3583 * reclaiming anything.
3584 */
3585 if (!nr_reclaimed &&
3586 (next_mz == NULL ||
3587 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3588 break;
3589 } while (!nr_reclaimed);
3590 if (next_mz)
3591 css_put(&next_mz->mem->css);
3592 return nr_reclaimed;
3593}
3594
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003595/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003596 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003597 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3598 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003599static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003600 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003601{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003602 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003603 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003604 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003605 struct page *busy;
3606 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003607 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08003608
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003609 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003610 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003611 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003612
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003613 loop = MEM_CGROUP_ZSTAT(mz, lru);
3614 /* give some margin against EBUSY etc...*/
3615 loop += 256;
3616 busy = NULL;
3617 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003618 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003619 struct page *page;
3620
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003621 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003622 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003623 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003624 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003625 break;
3626 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003627 page = list_entry(list->prev, struct page, lru);
3628 if (busy == page) {
3629 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003630 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003631 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003632 continue;
3633 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003634 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003635
Johannes Weiner925b7672012-01-12 17:18:15 -08003636 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003637
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003638 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003639 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003640 break;
3641
3642 if (ret == -EBUSY || ret == -EINVAL) {
3643 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003644 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003645 cond_resched();
3646 } else
3647 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003648 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003649
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003650 if (!ret && !list_empty(list))
3651 return -EBUSY;
3652 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003653}
3654
3655/*
3656 * make mem_cgroup's charge to be 0 if there is no task.
3657 * This enables deleting this mem_cgroup.
3658 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003659static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003660{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003661 int ret;
3662 int node, zid, shrink;
3663 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003664 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003665
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003666 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003667
3668 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003669 /* should free all ? */
3670 if (free_all)
3671 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003672move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003673 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003674 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003675 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003676 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003677 ret = -EINTR;
3678 if (signal_pending(current))
3679 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003680 /* This is for making all *used* pages to be on LRU. */
3681 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003682 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003683 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003684 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003685 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003686 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07003687 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003688 for_each_lru(l) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003689 ret = mem_cgroup_force_empty_list(memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003690 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003691 if (ret)
3692 break;
3693 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003694 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003695 if (ret)
3696 break;
3697 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003698 mem_cgroup_end_move(memcg);
3699 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003700 /* it seems parent cgroup doesn't have enough mem */
3701 if (ret == -ENOMEM)
3702 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003703 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003704 /* "ret" should also be checked to ensure all lists are empty. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003705 } while (memcg->res.usage > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003706out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003707 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003708 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003709
3710try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003711 /* returns EBUSY if there is a task or if we come here twice. */
3712 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003713 ret = -EBUSY;
3714 goto out;
3715 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003716 /* we call try-to-free pages for make this cgroup empty */
3717 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003718 /* try to free all pages in this cgroup */
3719 shrink = 1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003720 while (nr_retries && memcg->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003721 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003722
3723 if (signal_pending(current)) {
3724 ret = -EINTR;
3725 goto out;
3726 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003727 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003728 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003729 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003730 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003731 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003732 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003733 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003734
3735 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003736 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003737 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003738 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003739}
3740
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003741int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3742{
3743 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3744}
3745
3746
Balbir Singh18f59ea2009-01-07 18:08:07 -08003747static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3748{
3749 return mem_cgroup_from_cont(cont)->use_hierarchy;
3750}
3751
3752static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3753 u64 val)
3754{
3755 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003756 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003757 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003758 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003759
3760 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003761 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003762
3763 cgroup_lock();
3764 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003765 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003766 * in the child subtrees. If it is unset, then the change can
3767 * occur, provided the current cgroup has no children.
3768 *
3769 * For the root cgroup, parent_mem is NULL, we allow value to be
3770 * set if there are no children.
3771 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003772 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003773 (val == 1 || val == 0)) {
3774 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003775 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003776 else
3777 retval = -EBUSY;
3778 } else
3779 retval = -EINVAL;
3780 cgroup_unlock();
3781
3782 return retval;
3783}
3784
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003785
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003786static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003787 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003788{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003789 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003790 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003791
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003792 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003793 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003794 val += mem_cgroup_read_stat(iter, idx);
3795
3796 if (val < 0) /* race ? */
3797 val = 0;
3798 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003799}
3800
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003801static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003802{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003803 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003804
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003805 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003806 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003807 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003808 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003809 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003810 }
3811
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003812 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3813 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003814
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003815 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003816 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003817
3818 return val << PAGE_SHIFT;
3819}
3820
Paul Menage2c3daa72008-04-29 00:59:58 -07003821static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003822{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003823 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003824 u64 val;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003825 int type, name;
3826
3827 type = MEMFILE_TYPE(cft->private);
3828 name = MEMFILE_ATTR(cft->private);
3829 switch (type) {
3830 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003831 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003832 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003833 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003834 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003835 break;
3836 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003837 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003838 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003839 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003840 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003841 break;
3842 default:
3843 BUG();
3844 break;
3845 }
3846 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003847}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003848/*
3849 * The user of this function is...
3850 * RES_LIMIT.
3851 */
Paul Menage856c13a2008-07-25 01:47:04 -07003852static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3853 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003854{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003855 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003856 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003857 unsigned long long val;
3858 int ret;
3859
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003860 type = MEMFILE_TYPE(cft->private);
3861 name = MEMFILE_ATTR(cft->private);
3862 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003863 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003864 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3865 ret = -EINVAL;
3866 break;
3867 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003868 /* This function does all necessary parse...reuse it */
3869 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003870 if (ret)
3871 break;
3872 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003873 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003874 else
3875 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003876 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003877 case RES_SOFT_LIMIT:
3878 ret = res_counter_memparse_write_strategy(buffer, &val);
3879 if (ret)
3880 break;
3881 /*
3882 * For memsw, soft limits are hard to implement in terms
3883 * of semantics, for now, we support soft limits for
3884 * control without swap
3885 */
3886 if (type == _MEM)
3887 ret = res_counter_set_soft_limit(&memcg->res, val);
3888 else
3889 ret = -EINVAL;
3890 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003891 default:
3892 ret = -EINVAL; /* should be BUG() ? */
3893 break;
3894 }
3895 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003896}
3897
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003898static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3899 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3900{
3901 struct cgroup *cgroup;
3902 unsigned long long min_limit, min_memsw_limit, tmp;
3903
3904 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3905 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3906 cgroup = memcg->css.cgroup;
3907 if (!memcg->use_hierarchy)
3908 goto out;
3909
3910 while (cgroup->parent) {
3911 cgroup = cgroup->parent;
3912 memcg = mem_cgroup_from_cont(cgroup);
3913 if (!memcg->use_hierarchy)
3914 break;
3915 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3916 min_limit = min(min_limit, tmp);
3917 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3918 min_memsw_limit = min(min_memsw_limit, tmp);
3919 }
3920out:
3921 *mem_limit = min_limit;
3922 *memsw_limit = min_memsw_limit;
3923 return;
3924}
3925
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003926static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003927{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003928 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003929 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003930
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003931 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003932 type = MEMFILE_TYPE(event);
3933 name = MEMFILE_ATTR(event);
3934 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003935 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003936 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003937 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003938 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003939 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003940 break;
3941 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003942 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003943 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003944 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003945 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003946 break;
3947 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003948
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07003949 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003950}
3951
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003952static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3953 struct cftype *cft)
3954{
3955 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3956}
3957
Daisuke Nishimura02491442010-03-10 15:22:17 -08003958#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003959static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3960 struct cftype *cft, u64 val)
3961{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003962 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003963
3964 if (val >= (1 << NR_MOVE_TYPE))
3965 return -EINVAL;
3966 /*
3967 * We check this value several times in both in can_attach() and
3968 * attach(), so we need cgroup lock to prevent this value from being
3969 * inconsistent.
3970 */
3971 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003972 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003973 cgroup_unlock();
3974
3975 return 0;
3976}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003977#else
3978static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3979 struct cftype *cft, u64 val)
3980{
3981 return -ENOSYS;
3982}
3983#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003984
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003985
3986/* For read statistics */
3987enum {
3988 MCS_CACHE,
3989 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08003990 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003991 MCS_PGPGIN,
3992 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003993 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07003994 MCS_PGFAULT,
3995 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003996 MCS_INACTIVE_ANON,
3997 MCS_ACTIVE_ANON,
3998 MCS_INACTIVE_FILE,
3999 MCS_ACTIVE_FILE,
4000 MCS_UNEVICTABLE,
4001 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004002};
4003
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004004struct mcs_total_stat {
4005 s64 stat[NR_MCS_STAT];
4006};
4007
4008struct {
4009 char *local_name;
4010 char *total_name;
4011} memcg_stat_strings[NR_MCS_STAT] = {
4012 {"cache", "total_cache"},
4013 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004014 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004015 {"pgpgin", "total_pgpgin"},
4016 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004017 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004018 {"pgfault", "total_pgfault"},
4019 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004020 {"inactive_anon", "total_inactive_anon"},
4021 {"active_anon", "total_active_anon"},
4022 {"inactive_file", "total_inactive_file"},
4023 {"active_file", "total_active_file"},
4024 {"unevictable", "total_unevictable"}
4025};
4026
4027
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004028static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004029mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004030{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004031 s64 val;
4032
4033 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004034 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004035 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004036 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004037 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004038 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004039 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004040 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004041 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004042 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004043 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004044 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004045 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004046 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4047 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004048 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004049 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004050 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004051 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004052
4053 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004054 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004055 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004056 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004057 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004058 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004059 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004060 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004061 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004062 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004063 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004064}
4065
4066static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004067mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004068{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004069 struct mem_cgroup *iter;
4070
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004071 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004072 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004073}
4074
Ying Han406eb0c2011-05-26 16:25:37 -07004075#ifdef CONFIG_NUMA
4076static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4077{
4078 int nid;
4079 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4080 unsigned long node_nr;
4081 struct cgroup *cont = m->private;
4082 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4083
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004084 total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004085 seq_printf(m, "total=%lu", total_nr);
4086 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004087 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004088 seq_printf(m, " N%d=%lu", nid, node_nr);
4089 }
4090 seq_putc(m, '\n');
4091
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004092 file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004093 seq_printf(m, "file=%lu", file_nr);
4094 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004095 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4096 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004097 seq_printf(m, " N%d=%lu", nid, node_nr);
4098 }
4099 seq_putc(m, '\n');
4100
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004101 anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004102 seq_printf(m, "anon=%lu", anon_nr);
4103 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004104 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4105 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004106 seq_printf(m, " N%d=%lu", nid, node_nr);
4107 }
4108 seq_putc(m, '\n');
4109
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004110 unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004111 seq_printf(m, "unevictable=%lu", unevictable_nr);
4112 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004113 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4114 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004115 seq_printf(m, " N%d=%lu", nid, node_nr);
4116 }
4117 seq_putc(m, '\n');
4118 return 0;
4119}
4120#endif /* CONFIG_NUMA */
4121
Paul Menagec64745c2008-04-29 01:00:02 -07004122static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4123 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004124{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004125 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004126 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004127 int i;
4128
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004129 memset(&mystat, 0, sizeof(mystat));
4130 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004131
Ying Han406eb0c2011-05-26 16:25:37 -07004132
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004133 for (i = 0; i < NR_MCS_STAT; i++) {
4134 if (i == MCS_SWAP && !do_swap_account)
4135 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004136 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004137 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004138
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004139 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004140 {
4141 unsigned long long limit, memsw_limit;
4142 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
4143 cb->fill(cb, "hierarchical_memory_limit", limit);
4144 if (do_swap_account)
4145 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4146 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004147
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004148 memset(&mystat, 0, sizeof(mystat));
4149 mem_cgroup_get_total_stat(mem_cont, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004150 for (i = 0; i < NR_MCS_STAT; i++) {
4151 if (i == MCS_SWAP && !do_swap_account)
4152 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004153 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004154 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004155
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004156#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004157 {
4158 int nid, zid;
4159 struct mem_cgroup_per_zone *mz;
4160 unsigned long recent_rotated[2] = {0, 0};
4161 unsigned long recent_scanned[2] = {0, 0};
4162
4163 for_each_online_node(nid)
4164 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4165 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
4166
4167 recent_rotated[0] +=
4168 mz->reclaim_stat.recent_rotated[0];
4169 recent_rotated[1] +=
4170 mz->reclaim_stat.recent_rotated[1];
4171 recent_scanned[0] +=
4172 mz->reclaim_stat.recent_scanned[0];
4173 recent_scanned[1] +=
4174 mz->reclaim_stat.recent_scanned[1];
4175 }
4176 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4177 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4178 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4179 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4180 }
4181#endif
4182
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004183 return 0;
4184}
4185
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004186static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4187{
4188 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4189
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004190 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004191}
4192
4193static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4194 u64 val)
4195{
4196 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4197 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004198
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004199 if (val > 100)
4200 return -EINVAL;
4201
4202 if (cgrp->parent == NULL)
4203 return -EINVAL;
4204
4205 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004206
4207 cgroup_lock();
4208
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004209 /* If under hierarchy, only empty-root can set this value */
4210 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004211 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4212 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004213 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004214 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004215
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004216 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004217
Li Zefan068b38c2009-01-15 13:51:26 -08004218 cgroup_unlock();
4219
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004220 return 0;
4221}
4222
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004223static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4224{
4225 struct mem_cgroup_threshold_ary *t;
4226 u64 usage;
4227 int i;
4228
4229 rcu_read_lock();
4230 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004231 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004232 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004233 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004234
4235 if (!t)
4236 goto unlock;
4237
4238 usage = mem_cgroup_usage(memcg, swap);
4239
4240 /*
4241 * current_threshold points to threshold just below usage.
4242 * If it's not true, a threshold was crossed after last
4243 * call of __mem_cgroup_threshold().
4244 */
Phil Carmody5407a562010-05-26 14:42:42 -07004245 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004246
4247 /*
4248 * Iterate backward over array of thresholds starting from
4249 * current_threshold and check if a threshold is crossed.
4250 * If none of thresholds below usage is crossed, we read
4251 * only one element of the array here.
4252 */
4253 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4254 eventfd_signal(t->entries[i].eventfd, 1);
4255
4256 /* i = current_threshold + 1 */
4257 i++;
4258
4259 /*
4260 * Iterate forward over array of thresholds starting from
4261 * current_threshold+1 and check if a threshold is crossed.
4262 * If none of thresholds above usage is crossed, we read
4263 * only one element of the array here.
4264 */
4265 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4266 eventfd_signal(t->entries[i].eventfd, 1);
4267
4268 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004269 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004270unlock:
4271 rcu_read_unlock();
4272}
4273
4274static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4275{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004276 while (memcg) {
4277 __mem_cgroup_threshold(memcg, false);
4278 if (do_swap_account)
4279 __mem_cgroup_threshold(memcg, true);
4280
4281 memcg = parent_mem_cgroup(memcg);
4282 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004283}
4284
4285static int compare_thresholds(const void *a, const void *b)
4286{
4287 const struct mem_cgroup_threshold *_a = a;
4288 const struct mem_cgroup_threshold *_b = b;
4289
4290 return _a->threshold - _b->threshold;
4291}
4292
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004293static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004294{
4295 struct mem_cgroup_eventfd_list *ev;
4296
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004297 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004298 eventfd_signal(ev->eventfd, 1);
4299 return 0;
4300}
4301
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004302static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004303{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004304 struct mem_cgroup *iter;
4305
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004306 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004307 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004308}
4309
4310static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4311 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004312{
4313 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004314 struct mem_cgroup_thresholds *thresholds;
4315 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004316 int type = MEMFILE_TYPE(cft->private);
4317 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004318 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004319
4320 ret = res_counter_memparse_write_strategy(args, &threshold);
4321 if (ret)
4322 return ret;
4323
4324 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004325
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004326 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004327 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004328 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004329 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004330 else
4331 BUG();
4332
4333 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4334
4335 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004336 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004337 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4338
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004339 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004340
4341 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004342 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004343 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004344 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004345 ret = -ENOMEM;
4346 goto unlock;
4347 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004348 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004349
4350 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004351 if (thresholds->primary) {
4352 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004353 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004354 }
4355
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004356 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004357 new->entries[size - 1].eventfd = eventfd;
4358 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004359
4360 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004361 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004362 compare_thresholds, NULL);
4363
4364 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004365 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004366 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004367 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004368 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004369 * new->current_threshold will not be used until
4370 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004371 * it here.
4372 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004373 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004374 }
4375 }
4376
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004377 /* Free old spare buffer and save old primary buffer as spare */
4378 kfree(thresholds->spare);
4379 thresholds->spare = thresholds->primary;
4380
4381 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004382
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004383 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004384 synchronize_rcu();
4385
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004386unlock:
4387 mutex_unlock(&memcg->thresholds_lock);
4388
4389 return ret;
4390}
4391
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004392static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004393 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004394{
4395 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004396 struct mem_cgroup_thresholds *thresholds;
4397 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398 int type = MEMFILE_TYPE(cft->private);
4399 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004400 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004401
4402 mutex_lock(&memcg->thresholds_lock);
4403 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004404 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004405 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004406 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004407 else
4408 BUG();
4409
4410 /*
4411 * Something went wrong if we trying to unregister a threshold
4412 * if we don't have thresholds
4413 */
4414 BUG_ON(!thresholds);
4415
4416 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4417
4418 /* Check if a threshold crossed before removing */
4419 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4420
4421 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004422 size = 0;
4423 for (i = 0; i < thresholds->primary->size; i++) {
4424 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004425 size++;
4426 }
4427
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004428 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004429
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004430 /* Set thresholds array to NULL if we don't have thresholds */
4431 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004432 kfree(new);
4433 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004434 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004435 }
4436
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004437 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004438
4439 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004440 new->current_threshold = -1;
4441 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4442 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004443 continue;
4444
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004445 new->entries[j] = thresholds->primary->entries[i];
4446 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004447 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004448 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004449 * until rcu_assign_pointer(), so it's safe to increment
4450 * it here.
4451 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004452 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004453 }
4454 j++;
4455 }
4456
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004457swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004458 /* Swap primary and spare array */
4459 thresholds->spare = thresholds->primary;
4460 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004461
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004462 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004463 synchronize_rcu();
4464
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004465 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004466}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004467
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004468static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4469 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4470{
4471 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4472 struct mem_cgroup_eventfd_list *event;
4473 int type = MEMFILE_TYPE(cft->private);
4474
4475 BUG_ON(type != _OOM_TYPE);
4476 event = kmalloc(sizeof(*event), GFP_KERNEL);
4477 if (!event)
4478 return -ENOMEM;
4479
Michal Hocko1af8efe2011-07-26 16:08:24 -07004480 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004481
4482 event->eventfd = eventfd;
4483 list_add(&event->list, &memcg->oom_notify);
4484
4485 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004486 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004487 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004488 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004489
4490 return 0;
4491}
4492
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004493static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004494 struct cftype *cft, struct eventfd_ctx *eventfd)
4495{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004496 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004497 struct mem_cgroup_eventfd_list *ev, *tmp;
4498 int type = MEMFILE_TYPE(cft->private);
4499
4500 BUG_ON(type != _OOM_TYPE);
4501
Michal Hocko1af8efe2011-07-26 16:08:24 -07004502 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004503
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004504 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004505 if (ev->eventfd == eventfd) {
4506 list_del(&ev->list);
4507 kfree(ev);
4508 }
4509 }
4510
Michal Hocko1af8efe2011-07-26 16:08:24 -07004511 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004512}
4513
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004514static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4515 struct cftype *cft, struct cgroup_map_cb *cb)
4516{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004517 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004519 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004520
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004521 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004522 cb->fill(cb, "under_oom", 1);
4523 else
4524 cb->fill(cb, "under_oom", 0);
4525 return 0;
4526}
4527
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004528static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4529 struct cftype *cft, u64 val)
4530{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004531 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004532 struct mem_cgroup *parent;
4533
4534 /* cannot set to root cgroup and only 0 and 1 are allowed */
4535 if (!cgrp->parent || !((val == 0) || (val == 1)))
4536 return -EINVAL;
4537
4538 parent = mem_cgroup_from_cont(cgrp->parent);
4539
4540 cgroup_lock();
4541 /* oom-kill-disable is a flag for subhierarchy. */
4542 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004543 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004544 cgroup_unlock();
4545 return -EINVAL;
4546 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004547 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004548 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004549 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004550 cgroup_unlock();
4551 return 0;
4552}
4553
Ying Han406eb0c2011-05-26 16:25:37 -07004554#ifdef CONFIG_NUMA
4555static const struct file_operations mem_control_numa_stat_file_operations = {
4556 .read = seq_read,
4557 .llseek = seq_lseek,
4558 .release = single_release,
4559};
4560
4561static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4562{
4563 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4564
4565 file->f_op = &mem_control_numa_stat_file_operations;
4566 return single_open(file, mem_control_numa_stat_show, cont);
4567}
4568#endif /* CONFIG_NUMA */
4569
Glauber Costae5671df2011-12-11 21:47:01 +00004570#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae5671df2011-12-11 21:47:01 +00004571static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4572{
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004573 /*
4574 * Part of this would be better living in a separate allocation
4575 * function, leaving us with just the cgroup tree population work.
4576 * We, however, depend on state such as network's proto_list that
4577 * is only initialized after cgroup creation. I found the less
4578 * cumbersome way to deal with it to defer it all to populate time
4579 */
Glauber Costa65c64ce2011-12-22 01:02:27 +00004580 return mem_cgroup_sockets_init(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004581};
4582
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004583static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4584 struct cgroup *cont)
4585{
4586 mem_cgroup_sockets_destroy(cont, ss);
4587}
Glauber Costae5671df2011-12-11 21:47:01 +00004588#else
4589static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4590{
4591 return 0;
4592}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004593
4594static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4595 struct cgroup *cont)
4596{
4597}
Glauber Costae5671df2011-12-11 21:47:01 +00004598#endif
4599
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004600static struct cftype mem_cgroup_files[] = {
4601 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004602 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004603 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004604 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004605 .register_event = mem_cgroup_usage_register_event,
4606 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004607 },
4608 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004609 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004610 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004611 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004612 .read_u64 = mem_cgroup_read,
4613 },
4614 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004615 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004616 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004617 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004618 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004619 },
4620 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004621 .name = "soft_limit_in_bytes",
4622 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4623 .write_string = mem_cgroup_write,
4624 .read_u64 = mem_cgroup_read,
4625 },
4626 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004627 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004628 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004629 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004630 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004631 },
Balbir Singh8697d332008-02-07 00:13:59 -08004632 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004633 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004634 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004635 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004636 {
4637 .name = "force_empty",
4638 .trigger = mem_cgroup_force_empty_write,
4639 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004640 {
4641 .name = "use_hierarchy",
4642 .write_u64 = mem_cgroup_hierarchy_write,
4643 .read_u64 = mem_cgroup_hierarchy_read,
4644 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004645 {
4646 .name = "swappiness",
4647 .read_u64 = mem_cgroup_swappiness_read,
4648 .write_u64 = mem_cgroup_swappiness_write,
4649 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004650 {
4651 .name = "move_charge_at_immigrate",
4652 .read_u64 = mem_cgroup_move_charge_read,
4653 .write_u64 = mem_cgroup_move_charge_write,
4654 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004655 {
4656 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004657 .read_map = mem_cgroup_oom_control_read,
4658 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004659 .register_event = mem_cgroup_oom_register_event,
4660 .unregister_event = mem_cgroup_oom_unregister_event,
4661 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4662 },
Ying Han406eb0c2011-05-26 16:25:37 -07004663#ifdef CONFIG_NUMA
4664 {
4665 .name = "numa_stat",
4666 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004667 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004668 },
4669#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004670};
4671
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004672#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4673static struct cftype memsw_cgroup_files[] = {
4674 {
4675 .name = "memsw.usage_in_bytes",
4676 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4677 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004678 .register_event = mem_cgroup_usage_register_event,
4679 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004680 },
4681 {
4682 .name = "memsw.max_usage_in_bytes",
4683 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4684 .trigger = mem_cgroup_reset,
4685 .read_u64 = mem_cgroup_read,
4686 },
4687 {
4688 .name = "memsw.limit_in_bytes",
4689 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4690 .write_string = mem_cgroup_write,
4691 .read_u64 = mem_cgroup_read,
4692 },
4693 {
4694 .name = "memsw.failcnt",
4695 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4696 .trigger = mem_cgroup_reset,
4697 .read_u64 = mem_cgroup_read,
4698 },
4699};
4700
4701static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4702{
4703 if (!do_swap_account)
4704 return 0;
4705 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4706 ARRAY_SIZE(memsw_cgroup_files));
4707};
4708#else
4709static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4710{
4711 return 0;
4712}
4713#endif
4714
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004715static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004716{
4717 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004718 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004719 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004720 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004721 /*
4722 * This routine is called against possible nodes.
4723 * But it's BUG to call kmalloc() against offline node.
4724 *
4725 * TODO: this routine can waste much memory for nodes which will
4726 * never be onlined. It's better to use memory hotplug callback
4727 * function.
4728 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004729 if (!node_state(node, N_NORMAL_MEMORY))
4730 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004731 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004732 if (!pn)
4733 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004734
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004735 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4736 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07004737 for_each_lru(l)
Johannes Weiner6290df52012-01-12 17:18:10 -08004738 INIT_LIST_HEAD(&mz->lruvec.lists[l]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004739 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004740 mz->on_tree = false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004741 mz->mem = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004742 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004743 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004744 return 0;
4745}
4746
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004747static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004748{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004749 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004750}
4751
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004752static struct mem_cgroup *mem_cgroup_alloc(void)
4753{
4754 struct mem_cgroup *mem;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004755 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004756
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004757 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004758 if (size < PAGE_SIZE)
Jesper Juhl17295c82011-01-13 15:47:42 -08004759 mem = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004760 else
Jesper Juhl17295c82011-01-13 15:47:42 -08004761 mem = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004762
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004763 if (!mem)
4764 return NULL;
4765
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004766 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004767 if (!mem->stat)
4768 goto out_free;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004769 spin_lock_init(&mem->pcp_counter_lock);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004770 return mem;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004771
4772out_free:
4773 if (size < PAGE_SIZE)
4774 kfree(mem);
4775 else
4776 vfree(mem);
4777 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004778}
4779
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004780/*
4781 * At destroying mem_cgroup, references from swap_cgroup can remain.
4782 * (scanning all at force_empty is too costly...)
4783 *
4784 * Instead of clearing all references at force_empty, we remember
4785 * the number of reference from swap_cgroup and free mem_cgroup when
4786 * it goes down to 0.
4787 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004788 * Removal of cgroup itself succeeds regardless of refs from swap.
4789 */
4790
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004791static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004792{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004793 int node;
4794
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004795 mem_cgroup_remove_from_trees(memcg);
4796 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004797
Bob Liu3ed28fa2012-01-12 17:19:04 -08004798 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004799 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004800
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004801 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004802 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004803 kfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004804 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004805 vfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004806}
4807
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004808static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004809{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004810 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004811}
4812
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004813static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004814{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004815 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4816 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4817 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004818 if (parent)
4819 mem_cgroup_put(parent);
4820 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004821}
4822
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004823static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004824{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004825 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004826}
4827
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004828/*
4829 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4830 */
Glauber Costae1aab162011-12-11 21:47:03 +00004831struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004832{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004833 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004834 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004835 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004836}
Glauber Costae1aab162011-12-11 21:47:03 +00004837EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004838
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004839#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4840static void __init enable_swap_cgroup(void)
4841{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08004842 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004843 do_swap_account = 1;
4844}
4845#else
4846static void __init enable_swap_cgroup(void)
4847{
4848}
4849#endif
4850
Balbir Singhf64c3f52009-09-23 15:56:37 -07004851static int mem_cgroup_soft_limit_tree_init(void)
4852{
4853 struct mem_cgroup_tree_per_node *rtpn;
4854 struct mem_cgroup_tree_per_zone *rtpz;
4855 int tmp, node, zone;
4856
Bob Liu3ed28fa2012-01-12 17:19:04 -08004857 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004858 tmp = node;
4859 if (!node_state(node, N_NORMAL_MEMORY))
4860 tmp = -1;
4861 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4862 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004863 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004864
4865 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4866
4867 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4868 rtpz = &rtpn->rb_tree_per_zone[zone];
4869 rtpz->rb_root = RB_ROOT;
4870 spin_lock_init(&rtpz->lock);
4871 }
4872 }
4873 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004874
4875err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004876 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004877 if (!soft_limit_tree.rb_tree_per_node[node])
4878 break;
4879 kfree(soft_limit_tree.rb_tree_per_node[node]);
4880 soft_limit_tree.rb_tree_per_node[node] = NULL;
4881 }
4882 return 1;
4883
Balbir Singhf64c3f52009-09-23 15:56:37 -07004884}
4885
Li Zefan0eb253e2009-01-15 13:51:25 -08004886static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004887mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4888{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004889 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004890 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004891 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004892
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004893 memcg = mem_cgroup_alloc();
4894 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004895 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004896
Bob Liu3ed28fa2012-01-12 17:19:04 -08004897 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004898 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004899 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004900
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004901 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004902 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004903 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004904 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004905 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004906 if (mem_cgroup_soft_limit_tree_init())
4907 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004908 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004909 for_each_possible_cpu(cpu) {
4910 struct memcg_stock_pcp *stock =
4911 &per_cpu(memcg_stock, cpu);
4912 INIT_WORK(&stock->work, drain_local_stock);
4913 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004914 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004915 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004916 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004917 memcg->use_hierarchy = parent->use_hierarchy;
4918 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004919 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004920
Balbir Singh18f59ea2009-01-07 18:08:07 -08004921 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004922 res_counter_init(&memcg->res, &parent->res);
4923 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004924 /*
4925 * We increment refcnt of the parent to ensure that we can
4926 * safely access it on res_counter_charge/uncharge.
4927 * This refcnt will be decremented when freeing this
4928 * mem_cgroup(see mem_cgroup_put).
4929 */
4930 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004931 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004932 res_counter_init(&memcg->res, NULL);
4933 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004934 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004935 memcg->last_scanned_node = MAX_NUMNODES;
4936 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08004937
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004938 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004939 memcg->swappiness = mem_cgroup_swappiness(parent);
4940 atomic_set(&memcg->refcnt, 1);
4941 memcg->move_charge_at_immigrate = 0;
4942 mutex_init(&memcg->thresholds_lock);
4943 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004944free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004945 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004946 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004947}
4948
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004949static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004950 struct cgroup *cont)
4951{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004952 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004953
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004954 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004955}
4956
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004957static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4958 struct cgroup *cont)
4959{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004960 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004961
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004962 kmem_cgroup_destroy(ss, cont);
4963
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004964 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004965}
4966
4967static int mem_cgroup_populate(struct cgroup_subsys *ss,
4968 struct cgroup *cont)
4969{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004970 int ret;
4971
4972 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4973 ARRAY_SIZE(mem_cgroup_files));
4974
4975 if (!ret)
4976 ret = register_memsw_files(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004977
4978 if (!ret)
4979 ret = register_kmem_files(cont, ss);
4980
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004981 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004982}
4983
Daisuke Nishimura02491442010-03-10 15:22:17 -08004984#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004985/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004986#define PRECHARGE_COUNT_AT_ONCE 256
4987static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004988{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004989 int ret = 0;
4990 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004991 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004992
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004993 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004994 mc.precharge += count;
4995 /* we don't need css_get for root */
4996 return ret;
4997 }
4998 /* try to charge at once */
4999 if (count > 1) {
5000 struct res_counter *dummy;
5001 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005002 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005003 * by cgroup_lock_live_cgroup() that it is not removed and we
5004 * are still under the same cgroup_mutex. So we can postpone
5005 * css_get().
5006 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005007 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005008 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005009 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005010 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005011 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005012 goto one_by_one;
5013 }
5014 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005015 return ret;
5016 }
5017one_by_one:
5018 /* fall back to one by one charge */
5019 while (count--) {
5020 if (signal_pending(current)) {
5021 ret = -EINTR;
5022 break;
5023 }
5024 if (!batch_count--) {
5025 batch_count = PRECHARGE_COUNT_AT_ONCE;
5026 cond_resched();
5027 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005028 ret = __mem_cgroup_try_charge(NULL,
5029 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005030 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005031 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005032 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005033 mc.precharge++;
5034 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005035 return ret;
5036}
5037
5038/**
5039 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5040 * @vma: the vma the pte to be checked belongs
5041 * @addr: the address corresponding to the pte to be checked
5042 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005043 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005044 *
5045 * Returns
5046 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5047 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5048 * move charge. if @target is not NULL, the page is stored in target->page
5049 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005050 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5051 * target for charge migration. if @target is not NULL, the entry is stored
5052 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005053 *
5054 * Called with pte lock held.
5055 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005056union mc_target {
5057 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005058 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005059};
5060
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005061enum mc_target_type {
5062 MC_TARGET_NONE, /* not used */
5063 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005064 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005065};
5066
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005067static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5068 unsigned long addr, pte_t ptent)
5069{
5070 struct page *page = vm_normal_page(vma, addr, ptent);
5071
5072 if (!page || !page_mapped(page))
5073 return NULL;
5074 if (PageAnon(page)) {
5075 /* we don't move shared anon */
5076 if (!move_anon() || page_mapcount(page) > 2)
5077 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005078 } else if (!move_file())
5079 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005080 return NULL;
5081 if (!get_page_unless_zero(page))
5082 return NULL;
5083
5084 return page;
5085}
5086
5087static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5088 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5089{
5090 int usage_count;
5091 struct page *page = NULL;
5092 swp_entry_t ent = pte_to_swp_entry(ptent);
5093
5094 if (!move_anon() || non_swap_entry(ent))
5095 return NULL;
5096 usage_count = mem_cgroup_count_swap_user(ent, &page);
5097 if (usage_count > 1) { /* we don't move shared anon */
5098 if (page)
5099 put_page(page);
5100 return NULL;
5101 }
5102 if (do_swap_account)
5103 entry->val = ent.val;
5104
5105 return page;
5106}
5107
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005108static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5109 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5110{
5111 struct page *page = NULL;
5112 struct inode *inode;
5113 struct address_space *mapping;
5114 pgoff_t pgoff;
5115
5116 if (!vma->vm_file) /* anonymous vma */
5117 return NULL;
5118 if (!move_file())
5119 return NULL;
5120
5121 inode = vma->vm_file->f_path.dentry->d_inode;
5122 mapping = vma->vm_file->f_mapping;
5123 if (pte_none(ptent))
5124 pgoff = linear_page_index(vma, addr);
5125 else /* pte_file(ptent) is true */
5126 pgoff = pte_to_pgoff(ptent);
5127
5128 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005129 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005130
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005131#ifdef CONFIG_SWAP
5132 /* shmem/tmpfs may report page out on swap: account for that too. */
5133 if (radix_tree_exceptional_entry(page)) {
5134 swp_entry_t swap = radix_to_swp_entry(page);
5135 if (do_swap_account)
5136 *entry = swap;
5137 page = find_get_page(&swapper_space, swap.val);
5138 }
5139#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005140 return page;
5141}
5142
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005143static int is_target_pte_for_mc(struct vm_area_struct *vma,
5144 unsigned long addr, pte_t ptent, union mc_target *target)
5145{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005146 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005147 struct page_cgroup *pc;
5148 int ret = 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005149 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005150
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005151 if (pte_present(ptent))
5152 page = mc_handle_present_pte(vma, addr, ptent);
5153 else if (is_swap_pte(ptent))
5154 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005155 else if (pte_none(ptent) || pte_file(ptent))
5156 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005157
5158 if (!page && !ent.val)
Daisuke Nishimura02491442010-03-10 15:22:17 -08005159 return 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005160 if (page) {
5161 pc = lookup_page_cgroup(page);
5162 /*
5163 * Do only loose check w/o page_cgroup lock.
5164 * mem_cgroup_move_account() checks the pc is valid or not under
5165 * the lock.
5166 */
5167 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5168 ret = MC_TARGET_PAGE;
5169 if (target)
5170 target->page = page;
5171 }
5172 if (!ret || !target)
5173 put_page(page);
5174 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005175 /* There is a swap entry and a page doesn't exist or isn't charged */
5176 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005177 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005178 ret = MC_TARGET_SWAP;
5179 if (target)
5180 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005181 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005182 return ret;
5183}
5184
5185static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5186 unsigned long addr, unsigned long end,
5187 struct mm_walk *walk)
5188{
5189 struct vm_area_struct *vma = walk->private;
5190 pte_t *pte;
5191 spinlock_t *ptl;
5192
Dave Hansen03319322011-03-22 16:32:56 -07005193 split_huge_page_pmd(walk->mm, pmd);
5194
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005195 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5196 for (; addr != end; pte++, addr += PAGE_SIZE)
5197 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
5198 mc.precharge++; /* increment precharge temporarily */
5199 pte_unmap_unlock(pte - 1, ptl);
5200 cond_resched();
5201
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005202 return 0;
5203}
5204
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005205static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5206{
5207 unsigned long precharge;
5208 struct vm_area_struct *vma;
5209
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005210 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005211 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5212 struct mm_walk mem_cgroup_count_precharge_walk = {
5213 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5214 .mm = mm,
5215 .private = vma,
5216 };
5217 if (is_vm_hugetlb_page(vma))
5218 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005219 walk_page_range(vma->vm_start, vma->vm_end,
5220 &mem_cgroup_count_precharge_walk);
5221 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005222 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005223
5224 precharge = mc.precharge;
5225 mc.precharge = 0;
5226
5227 return precharge;
5228}
5229
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005230static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5231{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005232 unsigned long precharge = mem_cgroup_count_precharge(mm);
5233
5234 VM_BUG_ON(mc.moving_task);
5235 mc.moving_task = current;
5236 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005237}
5238
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005239/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5240static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005241{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005242 struct mem_cgroup *from = mc.from;
5243 struct mem_cgroup *to = mc.to;
5244
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005245 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005246 if (mc.precharge) {
5247 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5248 mc.precharge = 0;
5249 }
5250 /*
5251 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5252 * we must uncharge here.
5253 */
5254 if (mc.moved_charge) {
5255 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5256 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005257 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005258 /* we must fixup refcnts and charges */
5259 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005260 /* uncharge swap account from the old cgroup */
5261 if (!mem_cgroup_is_root(mc.from))
5262 res_counter_uncharge(&mc.from->memsw,
5263 PAGE_SIZE * mc.moved_swap);
5264 __mem_cgroup_put(mc.from, mc.moved_swap);
5265
5266 if (!mem_cgroup_is_root(mc.to)) {
5267 /*
5268 * we charged both to->res and to->memsw, so we should
5269 * uncharge to->res.
5270 */
5271 res_counter_uncharge(&mc.to->res,
5272 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005273 }
5274 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005275 mc.moved_swap = 0;
5276 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005277 memcg_oom_recover(from);
5278 memcg_oom_recover(to);
5279 wake_up_all(&mc.waitq);
5280}
5281
5282static void mem_cgroup_clear_mc(void)
5283{
5284 struct mem_cgroup *from = mc.from;
5285
5286 /*
5287 * we must clear moving_task before waking up waiters at the end of
5288 * task migration.
5289 */
5290 mc.moving_task = NULL;
5291 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005292 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005293 mc.from = NULL;
5294 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005295 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005296 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005297}
5298
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005299static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5300 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005301 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005302{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005303 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005304 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005305 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005306
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005307 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005308 struct mm_struct *mm;
5309 struct mem_cgroup *from = mem_cgroup_from_task(p);
5310
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005311 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005312
5313 mm = get_task_mm(p);
5314 if (!mm)
5315 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005316 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005317 if (mm->owner == p) {
5318 VM_BUG_ON(mc.from);
5319 VM_BUG_ON(mc.to);
5320 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005321 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005322 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005323 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005324 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005325 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005326 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005327 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005328 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005329
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005330 ret = mem_cgroup_precharge_mc(mm);
5331 if (ret)
5332 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005333 }
5334 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005335 }
5336 return ret;
5337}
5338
5339static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5340 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005341 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005342{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005343 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005344}
5345
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005346static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5347 unsigned long addr, unsigned long end,
5348 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005349{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005350 int ret = 0;
5351 struct vm_area_struct *vma = walk->private;
5352 pte_t *pte;
5353 spinlock_t *ptl;
5354
Dave Hansen03319322011-03-22 16:32:56 -07005355 split_huge_page_pmd(walk->mm, pmd);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005356retry:
5357 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5358 for (; addr != end; addr += PAGE_SIZE) {
5359 pte_t ptent = *(pte++);
5360 union mc_target target;
5361 int type;
5362 struct page *page;
5363 struct page_cgroup *pc;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005364 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005365
5366 if (!mc.precharge)
5367 break;
5368
5369 type = is_target_pte_for_mc(vma, addr, ptent, &target);
5370 switch (type) {
5371 case MC_TARGET_PAGE:
5372 page = target.page;
5373 if (isolate_lru_page(page))
5374 goto put;
5375 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005376 if (!mem_cgroup_move_account(page, 1, pc,
5377 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005378 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005379 /* we uncharge from mc.from later. */
5380 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005381 }
5382 putback_lru_page(page);
5383put: /* is_target_pte_for_mc() gets the page */
5384 put_page(page);
5385 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005386 case MC_TARGET_SWAP:
5387 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005388 if (!mem_cgroup_move_swap_account(ent,
5389 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005390 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005391 /* we fixup refcnts and charges later. */
5392 mc.moved_swap++;
5393 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005394 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005395 default:
5396 break;
5397 }
5398 }
5399 pte_unmap_unlock(pte - 1, ptl);
5400 cond_resched();
5401
5402 if (addr != end) {
5403 /*
5404 * We have consumed all precharges we got in can_attach().
5405 * We try charge one by one, but don't do any additional
5406 * charges to mc.to if we have failed in charge once in attach()
5407 * phase.
5408 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005409 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005410 if (!ret)
5411 goto retry;
5412 }
5413
5414 return ret;
5415}
5416
5417static void mem_cgroup_move_charge(struct mm_struct *mm)
5418{
5419 struct vm_area_struct *vma;
5420
5421 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005422retry:
5423 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5424 /*
5425 * Someone who are holding the mmap_sem might be waiting in
5426 * waitq. So we cancel all extra charges, wake up all waiters,
5427 * and retry. Because we cancel precharges, we might not be able
5428 * to move enough charges, but moving charge is a best-effort
5429 * feature anyway, so it wouldn't be a big problem.
5430 */
5431 __mem_cgroup_clear_mc();
5432 cond_resched();
5433 goto retry;
5434 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005435 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5436 int ret;
5437 struct mm_walk mem_cgroup_move_charge_walk = {
5438 .pmd_entry = mem_cgroup_move_charge_pte_range,
5439 .mm = mm,
5440 .private = vma,
5441 };
5442 if (is_vm_hugetlb_page(vma))
5443 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005444 ret = walk_page_range(vma->vm_start, vma->vm_end,
5445 &mem_cgroup_move_charge_walk);
5446 if (ret)
5447 /*
5448 * means we have consumed all precharges and failed in
5449 * doing additional charge. Just abandon here.
5450 */
5451 break;
5452 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005453 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005454}
5455
Balbir Singh67e465a2008-02-07 00:13:54 -08005456static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5457 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005458 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005459{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005460 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005461 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005462
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005463 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005464 if (mc.to)
5465 mem_cgroup_move_charge(mm);
5466 put_swap_token(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005467 mmput(mm);
5468 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005469 if (mc.to)
5470 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005471}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005472#else /* !CONFIG_MMU */
5473static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5474 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005475 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005476{
5477 return 0;
5478}
5479static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5480 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005481 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005482{
5483}
5484static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5485 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005486 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005487{
5488}
5489#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005490
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005491struct cgroup_subsys mem_cgroup_subsys = {
5492 .name = "memory",
5493 .subsys_id = mem_cgroup_subsys_id,
5494 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005495 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005496 .destroy = mem_cgroup_destroy,
5497 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005498 .can_attach = mem_cgroup_can_attach,
5499 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005500 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005501 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005502 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005503};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005504
5505#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005506static int __init enable_swap_account(char *s)
5507{
5508 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005509 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005510 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005511 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005512 really_do_swap_account = 0;
5513 return 1;
5514}
Michal Hockoa2c89902011-05-24 17:12:50 -07005515__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005516
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005517#endif