blob: 228d6461c12ade8941b37416ca867d3df3277acd [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080013 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 */
23
24#include <linux/res_counter.h>
25#include <linux/memcontrol.h>
26#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080027#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080028#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080029#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080030#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080031#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070035#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040036#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080037#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070038#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080041#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080043#include <linux/eventfd.h>
44#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080045#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080046#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070047#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070048#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070049#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080050#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070051#include <linux/oom.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080052#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000053#include <net/sock.h>
54#include <net/tcp_memcontrol.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080055
Balbir Singh8697d332008-02-07 00:13:59 -080056#include <asm/uaccess.h>
57
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070058#include <trace/events/vmscan.h>
59
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070060struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070061#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070062struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080063
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080064#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070065/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080066int do_swap_account __read_mostly;
Michal Hockoa42c3902010-11-24 12:57:08 -080067
68/* for remember boot option*/
69#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70static int really_do_swap_account __initdata = 1;
71#else
72static int really_do_swap_account __initdata = 0;
73#endif
74
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075#else
76#define do_swap_account (0)
77#endif
78
79
Balbir Singh8cdea7c2008-02-07 00:13:50 -080080/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080081 * Statistics for memory cgroup.
82 */
83enum mem_cgroup_stat_index {
84 /*
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 */
87 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070088 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080089 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070090 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -070091 MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -070092 MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080093 MEM_CGROUP_STAT_NSTATS,
94};
95
Johannes Weinere9f89742011-03-23 16:42:37 -070096enum mem_cgroup_events_index {
97 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
98 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
99 MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
Ying Han456f9982011-05-26 16:25:38 -0700100 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
Johannes Weinere9f89742011-03-23 16:42:37 -0700102 MEM_CGROUP_EVENTS_NSTATS,
103};
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700104/*
105 * Per memcg event counter is incremented at every pagein/pageout. With THP,
106 * it will be incremated by the number of pages. This counter is used for
107 * for trigger some periodic events. This is straightforward and better
108 * than using jiffies etc. to handle periodic memcg event.
109 */
110enum mem_cgroup_events_target {
111 MEM_CGROUP_TARGET_THRESH,
112 MEM_CGROUP_TARGET_SOFTLIMIT,
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700113 MEM_CGROUP_TARGET_NUMAINFO,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700114 MEM_CGROUP_NTARGETS,
115};
116#define THRESHOLDS_EVENTS_TARGET (128)
117#define SOFTLIMIT_EVENTS_TARGET (1024)
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700118#define NUMAINFO_EVENTS_TARGET (1024)
Johannes Weinere9f89742011-03-23 16:42:37 -0700119
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800120struct mem_cgroup_stat_cpu {
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700121 long count[MEM_CGROUP_STAT_NSTATS];
Johannes Weinere9f89742011-03-23 16:42:37 -0700122 unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700123 unsigned long targets[MEM_CGROUP_NTARGETS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800124};
125
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800126struct mem_cgroup_reclaim_iter {
127 /* css_id of the last scanned hierarchy member */
128 int position;
129 /* scan generation, increased every round-trip */
130 unsigned int generation;
131};
132
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800133/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800134 * per-zone information in memory controller.
135 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800136struct mem_cgroup_per_zone {
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 Hiroyuki8c7c6e32009-01-07 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 Hiroyuki8c7c6e32009-01-07 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 Hiroyuki8c7c6e32009-01-07 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 Hiroyuki8c7c6e32009-01-07 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 Hiroyuki8c7c6e32009-01-07 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 Costa376be5f2012-01-20 04:57:14 +0000382 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000383 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{
Glauber Costa376be5f2012-01-20 04:57:14 +0000414 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000415 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 Hiroyuki8c7c6e32009-01-07 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))) {
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800779 bool do_softlimit;
780 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800781
782 do_softlimit = mem_cgroup_event_ratelimit(memcg,
783 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700784#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800785 do_numainfo = mem_cgroup_event_ratelimit(memcg,
786 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700787#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800788 preempt_enable();
789
790 mem_cgroup_threshold(memcg);
791 if (unlikely(do_softlimit))
792 mem_cgroup_update_tree(memcg, page);
793#if MAX_NUMNODES > 1
794 if (unlikely(do_numainfo))
795 atomic_inc(&memcg->numainfo_events);
796#endif
797 } else
798 preempt_enable();
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800799}
800
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000801struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800802{
803 return container_of(cgroup_subsys_state(cont,
804 mem_cgroup_subsys_id), struct mem_cgroup,
805 css);
806}
807
Balbir Singhcf475ad2008-04-29 01:00:16 -0700808struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800809{
Balbir Singh31a78f22008-09-28 23:09:31 +0100810 /*
811 * mm_update_next_owner() may clear mm->owner to NULL
812 * if it races with swapoff, page migration, etc.
813 * So this can be called with p == NULL.
814 */
815 if (unlikely(!p))
816 return NULL;
817
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800818 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
819 struct mem_cgroup, css);
820}
821
KOSAKI Motohiroa4336582011-06-15 15:08:13 -0700822struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800823{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700824 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700825
826 if (!mm)
827 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800828 /*
829 * Because we have no locks, mm->owner's may be being moved to other
830 * cgroup. We use css_tryget() here even if this looks
831 * pessimistic (rather than adding locks here).
832 */
833 rcu_read_lock();
834 do {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700835 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
836 if (unlikely(!memcg))
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800837 break;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700838 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800839 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700840 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800841}
842
Johannes Weiner56600482012-01-12 17:17:59 -0800843/**
844 * mem_cgroup_iter - iterate over memory cgroup hierarchy
845 * @root: hierarchy root
846 * @prev: previously returned memcg, NULL on first invocation
847 * @reclaim: cookie for shared reclaim walks, NULL for full walks
848 *
849 * Returns references to children of the hierarchy below @root, or
850 * @root itself, or %NULL after a full round-trip.
851 *
852 * Caller must pass the return value in @prev on subsequent
853 * invocations for reference counting, or use mem_cgroup_iter_break()
854 * to cancel a hierarchy walk before the round-trip is complete.
855 *
856 * Reclaimers can specify a zone and a priority level in @reclaim to
857 * divide up the memcgs in the hierarchy among all concurrent
858 * reclaimers operating on the same zone and priority.
859 */
860struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
861 struct mem_cgroup *prev,
862 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700863{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800864 struct mem_cgroup *memcg = NULL;
865 int id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700866
Johannes Weiner56600482012-01-12 17:17:59 -0800867 if (mem_cgroup_disabled())
868 return NULL;
869
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700870 if (!root)
871 root = root_mem_cgroup;
872
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800873 if (prev && !reclaim)
874 id = css_id(&prev->css);
875
876 if (prev && prev != root)
877 css_put(&prev->css);
878
879 if (!root->use_hierarchy && root != root_mem_cgroup) {
880 if (prev)
881 return NULL;
882 return root;
883 }
884
885 while (!memcg) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800886 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800887 struct cgroup_subsys_state *css;
888
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800889 if (reclaim) {
890 int nid = zone_to_nid(reclaim->zone);
891 int zid = zone_idx(reclaim->zone);
892 struct mem_cgroup_per_zone *mz;
893
894 mz = mem_cgroup_zoneinfo(root, nid, zid);
895 iter = &mz->reclaim_iter[reclaim->priority];
896 if (prev && reclaim->generation != iter->generation)
897 return NULL;
898 id = iter->position;
899 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800900
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700901 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800902 css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
903 if (css) {
904 if (css == &root->css || css_tryget(css))
905 memcg = container_of(css,
906 struct mem_cgroup, css);
907 } else
908 id = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700909 rcu_read_unlock();
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700910
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800911 if (reclaim) {
912 iter->position = id;
913 if (!css)
914 iter->generation++;
915 else if (!prev && memcg)
916 reclaim->generation = iter->generation;
917 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800918
919 if (prev && !css)
920 return NULL;
921 }
922 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700923}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800924
Johannes Weiner56600482012-01-12 17:17:59 -0800925/**
926 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
927 * @root: hierarchy root
928 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
929 */
930void mem_cgroup_iter_break(struct mem_cgroup *root,
931 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800932{
933 if (!root)
934 root = root_mem_cgroup;
935 if (prev && prev != root)
936 css_put(&prev->css);
937}
938
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700939/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800940 * Iteration constructs for visiting all cgroups (under a tree). If
941 * loops are exited prematurely (break), mem_cgroup_iter_break() must
942 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700943 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800944#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800945 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800946 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800947 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700948
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800949#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800950 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800951 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800952 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700953
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700954static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700955{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700956 return (memcg == root_mem_cgroup);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700957}
958
Ying Han456f9982011-05-26 16:25:38 -0700959void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
960{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700961 struct mem_cgroup *memcg;
Ying Han456f9982011-05-26 16:25:38 -0700962
963 if (!mm)
964 return;
965
966 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700967 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
968 if (unlikely(!memcg))
Ying Han456f9982011-05-26 16:25:38 -0700969 goto out;
970
971 switch (idx) {
Ying Han456f9982011-05-26 16:25:38 -0700972 case PGFAULT:
Johannes Weiner0e574a92012-01-12 17:18:35 -0800973 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
974 break;
975 case PGMAJFAULT:
976 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Ying Han456f9982011-05-26 16:25:38 -0700977 break;
978 default:
979 BUG();
980 }
981out:
982 rcu_read_unlock();
983}
984EXPORT_SYMBOL(mem_cgroup_count_vm_event);
985
Johannes Weiner925b7672012-01-12 17:18:15 -0800986/**
987 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
988 * @zone: zone of the wanted lruvec
989 * @mem: memcg of the wanted lruvec
990 *
991 * Returns the lru list vector holding pages for the given @zone and
992 * @mem. This can be the global zone lruvec, if the memory controller
993 * is disabled.
994 */
995struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
996 struct mem_cgroup *memcg)
997{
998 struct mem_cgroup_per_zone *mz;
999
1000 if (mem_cgroup_disabled())
1001 return &zone->lruvec;
1002
1003 mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1004 return &mz->lruvec;
1005}
1006
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001007/*
1008 * Following LRU functions are allowed to be used without PCG_LOCK.
1009 * Operations are called by routine of global LRU independently from memcg.
1010 * What we have to take care of here is validness of pc->mem_cgroup.
1011 *
1012 * Changes to pc->mem_cgroup happens when
1013 * 1. charge
1014 * 2. moving account
1015 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1016 * It is added to LRU before charge.
1017 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1018 * When moving account, the page is not on LRU. It's isolated.
1019 */
1020
Johannes Weiner925b7672012-01-12 17:18:15 -08001021/**
1022 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1023 * @zone: zone of the page
1024 * @page: the page
1025 * @lru: current lru
1026 *
1027 * This function accounts for @page being added to @lru, and returns
1028 * the lruvec for the given @zone and the memcg @page is charged to.
1029 *
1030 * The callsite is then responsible for physically linking the page to
1031 * the returned lruvec->lists[@lru].
Minchan Kim3f58a822011-03-22 16:32:53 -07001032 */
Johannes Weiner925b7672012-01-12 17:18:15 -08001033struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page,
1034 enum lru_list lru)
Minchan Kim3f58a822011-03-22 16:32:53 -07001035{
1036 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001037 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001038 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001039
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001040 if (mem_cgroup_disabled())
Johannes Weiner925b7672012-01-12 17:18:15 -08001041 return &zone->lruvec;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001042
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001043 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001044 memcg = pc->mem_cgroup;
Johannes Weiner925b7672012-01-12 17:18:15 -08001045 mz = page_cgroup_zoneinfo(memcg, page);
1046 /* compound_order() is stabilized through lru_lock */
KAMEZAWA Hiroyukiece35ca2011-01-20 14:44:24 -08001047 MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
Johannes Weiner925b7672012-01-12 17:18:15 -08001048 return &mz->lruvec;
1049}
1050
1051/**
1052 * mem_cgroup_lru_del_list - account for removing an lru page
1053 * @page: the page
1054 * @lru: target lru
1055 *
1056 * This function accounts for @page being removed from @lru.
1057 *
1058 * The callsite is then responsible for physically unlinking
1059 * @page->lru.
1060 */
1061void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru)
1062{
1063 struct mem_cgroup_per_zone *mz;
1064 struct mem_cgroup *memcg;
1065 struct page_cgroup *pc;
1066
1067 if (mem_cgroup_disabled())
1068 return;
1069
1070 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08001071 memcg = pc->mem_cgroup;
1072 VM_BUG_ON(!memcg);
Johannes Weiner925b7672012-01-12 17:18:15 -08001073 mz = page_cgroup_zoneinfo(memcg, page);
1074 /* huge page split is done under lru_lock. so, we have no races. */
Hugh Dickins12d27102012-01-12 17:19:52 -08001075 VM_BUG_ON(MEM_CGROUP_ZSTAT(mz, lru) < (1 << compound_order(page)));
Johannes Weiner925b7672012-01-12 17:18:15 -08001076 MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
1077}
1078
1079void mem_cgroup_lru_del(struct page *page)
1080{
1081 mem_cgroup_lru_del_list(page, page_lru(page));
1082}
1083
1084/**
1085 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1086 * @zone: zone of the page
1087 * @page: the page
1088 * @from: current lru
1089 * @to: target lru
1090 *
1091 * This function accounts for @page being moved between the lrus @from
1092 * and @to, and returns the lruvec for the given @zone and the memcg
1093 * @page is charged to.
1094 *
1095 * The callsite is then responsible for physically relinking
1096 * @page->lru to the returned lruvec->lists[@to].
1097 */
1098struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone,
1099 struct page *page,
1100 enum lru_list from,
1101 enum lru_list to)
1102{
1103 /* XXX: Optimize this, especially for @from == @to */
1104 mem_cgroup_lru_del_list(page, from);
1105 return mem_cgroup_lru_add_list(zone, page, to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001106}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001107
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001108/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001109 * Checks whether given mem is same or in the root_mem_cgroup's
Michal Hocko3e920412011-07-26 16:08:29 -07001110 * hierarchy subtree
1111 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001112static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1113 struct mem_cgroup *memcg)
Michal Hocko3e920412011-07-26 16:08:29 -07001114{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001115 if (root_memcg != memcg) {
1116 return (root_memcg->use_hierarchy &&
1117 css_is_ancestor(&memcg->css, &root_memcg->css));
Michal Hocko3e920412011-07-26 16:08:29 -07001118 }
1119
1120 return true;
1121}
1122
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001123int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
David Rientjes4c4a2212008-02-07 00:14:06 -08001124{
1125 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001126 struct mem_cgroup *curr = NULL;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001127 struct task_struct *p;
David Rientjes4c4a2212008-02-07 00:14:06 -08001128
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001129 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001130 if (p) {
1131 curr = try_get_mem_cgroup_from_mm(p->mm);
1132 task_unlock(p);
1133 } else {
1134 /*
1135 * All threads may have already detached their mm's, but the oom
1136 * killer still needs to detect if they have already been oom
1137 * killed to prevent needlessly killing additional tasks.
1138 */
1139 task_lock(task);
1140 curr = mem_cgroup_from_task(task);
1141 if (curr)
1142 css_get(&curr->css);
1143 task_unlock(task);
1144 }
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001145 if (!curr)
1146 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001147 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001148 * We should check use_hierarchy of "memcg" not "curr". Because checking
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001149 * use_hierarchy of "curr" here make this function true if hierarchy is
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001150 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1151 * hierarchy(even if use_hierarchy is disabled in "memcg").
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001152 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001153 ret = mem_cgroup_same_or_subtree(memcg, curr);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001154 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001155 return ret;
1156}
1157
Johannes Weiner9b272972011-11-02 13:38:23 -07001158int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001159{
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001160 unsigned long inactive_ratio;
Johannes Weiner9b272972011-11-02 13:38:23 -07001161 int nid = zone_to_nid(zone);
1162 int zid = zone_idx(zone);
1163 unsigned long inactive;
1164 unsigned long active;
1165 unsigned long gb;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001166
Johannes Weiner9b272972011-11-02 13:38:23 -07001167 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1168 BIT(LRU_INACTIVE_ANON));
1169 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1170 BIT(LRU_ACTIVE_ANON));
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001171
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001172 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1173 if (gb)
1174 inactive_ratio = int_sqrt(10 * gb);
1175 else
1176 inactive_ratio = 1;
1177
Johannes Weiner9b272972011-11-02 13:38:23 -07001178 return inactive * inactive_ratio < active;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001179}
1180
Johannes Weiner9b272972011-11-02 13:38:23 -07001181int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone)
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001182{
1183 unsigned long active;
1184 unsigned long inactive;
Johannes Weiner9b272972011-11-02 13:38:23 -07001185 int zid = zone_idx(zone);
1186 int nid = zone_to_nid(zone);
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001187
Johannes Weiner9b272972011-11-02 13:38:23 -07001188 inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1189 BIT(LRU_INACTIVE_FILE));
1190 active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid,
1191 BIT(LRU_ACTIVE_FILE));
Rik van Riel56e49d22009-06-16 15:32:28 -07001192
1193 return (active > inactive);
1194}
1195
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001196struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1197 struct zone *zone)
1198{
KOSAKI Motohiro13d7e3a2010-08-10 18:03:06 -07001199 int nid = zone_to_nid(zone);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001200 int zid = zone_idx(zone);
1201 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1202
1203 return &mz->reclaim_stat;
1204}
1205
1206struct zone_reclaim_stat *
1207mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1208{
1209 struct page_cgroup *pc;
1210 struct mem_cgroup_per_zone *mz;
1211
1212 if (mem_cgroup_disabled())
1213 return NULL;
1214
1215 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -08001216 if (!PageCgroupUsed(pc))
1217 return NULL;
Johannes Weiner713735b2011-01-20 14:44:31 -08001218 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1219 smp_rmb();
Johannes Weiner97a6c372011-03-23 16:42:27 -07001220 mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001221 return &mz->reclaim_stat;
1222}
1223
Balbir Singh6d61ef42009-01-07 18:08:06 -08001224#define mem_cgroup_from_res_counter(counter, member) \
1225 container_of(counter, struct mem_cgroup, member)
1226
Johannes Weiner19942822011-02-01 15:52:43 -08001227/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001228 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1229 * @mem: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001230 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001231 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001232 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001233 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001234static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001235{
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001236 unsigned long long margin;
1237
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001238 margin = res_counter_margin(&memcg->res);
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001239 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001240 margin = min(margin, res_counter_margin(&memcg->memsw));
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001241 return margin >> PAGE_SHIFT;
Johannes Weiner19942822011-02-01 15:52:43 -08001242}
1243
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07001244int mem_cgroup_swappiness(struct mem_cgroup *memcg)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001245{
1246 struct cgroup *cgrp = memcg->css.cgroup;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001247
1248 /* root ? */
1249 if (cgrp->parent == NULL)
1250 return vm_swappiness;
1251
Johannes Weinerbf1ff262011-03-23 16:42:32 -07001252 return memcg->swappiness;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001253}
1254
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001255static void mem_cgroup_start_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001256{
1257 int cpu;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001258
1259 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001260 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001261 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001262 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1263 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1264 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001265 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001266
1267 synchronize_rcu();
1268}
1269
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001270static void mem_cgroup_end_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001271{
1272 int cpu;
1273
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001274 if (!memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001275 return;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001276 get_online_cpus();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001277 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001278 for_each_online_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001279 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1280 memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1281 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001282 put_online_cpus();
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001283}
1284/*
1285 * 2 routines for checking "mem" is under move_account() or not.
1286 *
1287 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1288 * for avoiding race in accounting. If true,
1289 * pc->mem_cgroup may be overwritten.
1290 *
1291 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1292 * under hierarchy of moving cgroups. This is for
1293 * waiting at hith-memory prressure caused by "move".
1294 */
1295
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001296static bool mem_cgroup_stealed(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001297{
1298 VM_BUG_ON(!rcu_read_lock_held());
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001299 return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001300}
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001301
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001302static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001303{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001304 struct mem_cgroup *from;
1305 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001306 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001307 /*
1308 * Unlike task_move routines, we access mc.to, mc.from not under
1309 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1310 */
1311 spin_lock(&mc.lock);
1312 from = mc.from;
1313 to = mc.to;
1314 if (!from)
1315 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001316
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001317 ret = mem_cgroup_same_or_subtree(memcg, from)
1318 || mem_cgroup_same_or_subtree(memcg, to);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001319unlock:
1320 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001321 return ret;
1322}
1323
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001324static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001325{
1326 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001327 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001328 DEFINE_WAIT(wait);
1329 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1330 /* moving charge context might have finished. */
1331 if (mc.moving_task)
1332 schedule();
1333 finish_wait(&mc.waitq, &wait);
1334 return true;
1335 }
1336 }
1337 return false;
1338}
1339
Balbir Singhe2224322009-04-02 16:57:39 -07001340/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001341 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001342 * @memcg: The memory cgroup that went over limit
1343 * @p: Task that is going to be killed
1344 *
1345 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1346 * enabled
1347 */
1348void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1349{
1350 struct cgroup *task_cgrp;
1351 struct cgroup *mem_cgrp;
1352 /*
1353 * Need a buffer in BSS, can't rely on allocations. The code relies
1354 * on the assumption that OOM is serialized for memory controller.
1355 * If this assumption is broken, revisit this code.
1356 */
1357 static char memcg_name[PATH_MAX];
1358 int ret;
1359
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001360 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001361 return;
1362
1363
1364 rcu_read_lock();
1365
1366 mem_cgrp = memcg->css.cgroup;
1367 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1368
1369 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1370 if (ret < 0) {
1371 /*
1372 * Unfortunately, we are unable to convert to a useful name
1373 * But we'll still print out the usage information
1374 */
1375 rcu_read_unlock();
1376 goto done;
1377 }
1378 rcu_read_unlock();
1379
1380 printk(KERN_INFO "Task in %s killed", memcg_name);
1381
1382 rcu_read_lock();
1383 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1384 if (ret < 0) {
1385 rcu_read_unlock();
1386 goto done;
1387 }
1388 rcu_read_unlock();
1389
1390 /*
1391 * Continues from above, so we don't need an KERN_ level
1392 */
1393 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1394done:
1395
1396 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1397 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1398 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1399 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1400 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1401 "failcnt %llu\n",
1402 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1403 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1404 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1405}
1406
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001407/*
1408 * This function returns the number of memcg under hierarchy tree. Returns
1409 * 1(self count) if no children.
1410 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001411static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001412{
1413 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001414 struct mem_cgroup *iter;
1415
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001416 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001417 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001418 return num;
1419}
1420
Balbir Singh6d61ef42009-01-07 18:08:06 -08001421/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001422 * Return the memory (and swap, if configured) limit for a memcg.
1423 */
1424u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1425{
1426 u64 limit;
1427 u64 memsw;
1428
Johannes Weinerf3e8eb72011-01-13 15:47:39 -08001429 limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1430 limit += total_swap_pages << PAGE_SHIFT;
1431
David Rientjesa63d83f2010-08-09 17:19:46 -07001432 memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1433 /*
1434 * If memsw is finite and limits the amount of swap space available
1435 * to this memcg, return that limit.
1436 */
1437 return min(limit, memsw);
1438}
1439
Johannes Weiner56600482012-01-12 17:17:59 -08001440static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1441 gfp_t gfp_mask,
1442 unsigned long flags)
1443{
1444 unsigned long total = 0;
1445 bool noswap = false;
1446 int loop;
1447
1448 if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1449 noswap = true;
1450 if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1451 noswap = true;
1452
1453 for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1454 if (loop)
1455 drain_all_stock_async(memcg);
1456 total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1457 /*
1458 * Allow limit shrinkers, which are triggered directly
1459 * by userspace, to catch signals and stop reclaim
1460 * after minimal progress, regardless of the margin.
1461 */
1462 if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1463 break;
1464 if (mem_cgroup_margin(memcg))
1465 break;
1466 /*
1467 * If nothing was reclaimed after two attempts, there
1468 * may be no reclaimable pages in this hierarchy.
1469 */
1470 if (loop && !total)
1471 break;
1472 }
1473 return total;
1474}
1475
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001476/**
1477 * test_mem_cgroup_node_reclaimable
1478 * @mem: the target memcg
1479 * @nid: the node ID to be checked.
1480 * @noswap : specify true here if the user wants flle only information.
1481 *
1482 * This function returns whether the specified memcg contains any
1483 * reclaimable pages on a node. Returns true if there are any reclaimable
1484 * pages in the node.
1485 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001486static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001487 int nid, bool noswap)
1488{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001489 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001490 return true;
1491 if (noswap || !total_swap_pages)
1492 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001493 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001494 return true;
1495 return false;
1496
1497}
Ying Han889976d2011-05-26 16:25:33 -07001498#if MAX_NUMNODES > 1
1499
1500/*
1501 * Always updating the nodemask is not very good - even if we have an empty
1502 * list or the wrong list here, we can start from some node and traverse all
1503 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1504 *
1505 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001506static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001507{
1508 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001509 /*
1510 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1511 * pagein/pageout changes since the last update.
1512 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001513 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001514 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001515 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001516 return;
1517
Ying Han889976d2011-05-26 16:25:33 -07001518 /* make a nodemask where this memcg uses memory from */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001519 memcg->scan_nodes = node_states[N_HIGH_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001520
1521 for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1522
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001523 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1524 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001525 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001526
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001527 atomic_set(&memcg->numainfo_events, 0);
1528 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001529}
1530
1531/*
1532 * Selecting a node where we start reclaim from. Because what we need is just
1533 * reducing usage counter, start from anywhere is O,K. Considering
1534 * memory reclaim from current node, there are pros. and cons.
1535 *
1536 * Freeing memory from current node means freeing memory from a node which
1537 * we'll use or we've used. So, it may make LRU bad. And if several threads
1538 * hit limits, it will see a contention on a node. But freeing from remote
1539 * node means more costs for memory reclaim because of memory latency.
1540 *
1541 * Now, we use round-robin. Better algorithm is welcomed.
1542 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001543int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001544{
1545 int node;
1546
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001547 mem_cgroup_may_update_nodemask(memcg);
1548 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001549
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001550 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001551 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001552 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001553 /*
1554 * We call this when we hit limit, not when pages are added to LRU.
1555 * No LRU may hold pages because all pages are UNEVICTABLE or
1556 * memcg is too small and all pages are not on LRU. In that case,
1557 * we use curret node.
1558 */
1559 if (unlikely(node == MAX_NUMNODES))
1560 node = numa_node_id();
1561
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001562 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001563 return node;
1564}
1565
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001566/*
1567 * Check all nodes whether it contains reclaimable pages or not.
1568 * For quick scan, we make use of scan_nodes. This will allow us to skip
1569 * unused nodes. But scan_nodes is lazily updated and may not cotain
1570 * enough new information. We need to do double check.
1571 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001572bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001573{
1574 int nid;
1575
1576 /*
1577 * quick check...making use of scan_node.
1578 * We can skip unused nodes.
1579 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001580 if (!nodes_empty(memcg->scan_nodes)) {
1581 for (nid = first_node(memcg->scan_nodes);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001582 nid < MAX_NUMNODES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583 nid = next_node(nid, memcg->scan_nodes)) {
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001585 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001586 return true;
1587 }
1588 }
1589 /*
1590 * Check rest of nodes.
1591 */
1592 for_each_node_state(nid, N_HIGH_MEMORY) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001593 if (node_isset(nid, memcg->scan_nodes))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001594 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001595 if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001596 return true;
1597 }
1598 return false;
1599}
1600
Ying Han889976d2011-05-26 16:25:33 -07001601#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001602int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001603{
1604 return 0;
1605}
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001606
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001607bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001608{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001609 return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001610}
Ying Han889976d2011-05-26 16:25:33 -07001611#endif
1612
Johannes Weiner56600482012-01-12 17:17:59 -08001613static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1614 struct zone *zone,
1615 gfp_t gfp_mask,
1616 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001617{
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001618 struct mem_cgroup *victim = NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001619 int total = 0;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001620 int loop = 0;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001621 unsigned long excess;
Johannes Weiner185efc02011-09-14 16:21:58 -07001622 unsigned long nr_scanned;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001623 struct mem_cgroup_reclaim_cookie reclaim = {
1624 .zone = zone,
1625 .priority = 0,
1626 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001627
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001628 excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001629
Balbir Singh4e416952009-09-23 15:56:39 -07001630 while (1) {
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001631 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001632 if (!victim) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001633 loop++;
Balbir Singh4e416952009-09-23 15:56:39 -07001634 if (loop >= 2) {
1635 /*
1636 * If we have not been able to reclaim
1637 * anything, it might because there are
1638 * no reclaimable pages under this hierarchy
1639 */
Johannes Weiner56600482012-01-12 17:17:59 -08001640 if (!total)
Balbir Singh4e416952009-09-23 15:56:39 -07001641 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001642 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001643 * We want to do more targeted reclaim.
Balbir Singh4e416952009-09-23 15:56:39 -07001644 * excess >> 2 is not to excessive so as to
1645 * reclaim too much, nor too less that we keep
1646 * coming back to reclaim from this cgroup
1647 */
1648 if (total >= (excess >> 2) ||
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001649 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
Balbir Singh4e416952009-09-23 15:56:39 -07001650 break;
Balbir Singh4e416952009-09-23 15:56:39 -07001651 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001652 continue;
Balbir Singh4e416952009-09-23 15:56:39 -07001653 }
Johannes Weiner56600482012-01-12 17:17:59 -08001654 if (!mem_cgroup_reclaimable(victim, false))
Balbir Singh6d61ef42009-01-07 18:08:06 -08001655 continue;
Johannes Weiner56600482012-01-12 17:17:59 -08001656 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1657 zone, &nr_scanned);
1658 *total_scanned += nr_scanned;
1659 if (!res_counter_soft_limit_excess(&root_memcg->res))
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001660 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001661 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001662 mem_cgroup_iter_break(root_memcg, victim);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001663 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001664}
1665
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001666/*
1667 * Check OOM-Killer is already running under our hierarchy.
1668 * If someone is running, return false.
Michal Hocko1af8efe2011-07-26 16:08:24 -07001669 * Has to be called with memcg_oom_lock
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001670 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001671static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001672{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001673 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001674
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001675 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001676 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001677 /*
1678 * this subtree of our hierarchy is already locked
1679 * so we cannot give a lock.
1680 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001681 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001682 mem_cgroup_iter_break(memcg, iter);
1683 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001684 } else
1685 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001686 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001687
Michal Hocko79dfdac2011-07-26 16:08:23 -07001688 if (!failed)
Johannes Weiner23751be2011-08-25 15:59:16 -07001689 return true;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001690
1691 /*
1692 * OK, we failed to lock the whole subtree so we have to clean up
1693 * what we set up to the failing subtree
1694 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001695 for_each_mem_cgroup_tree(iter, memcg) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001696 if (iter == failed) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001697 mem_cgroup_iter_break(memcg, iter);
1698 break;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001699 }
1700 iter->oom_lock = false;
1701 }
Johannes Weiner23751be2011-08-25 15:59:16 -07001702 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001703}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001704
Michal Hocko79dfdac2011-07-26 16:08:23 -07001705/*
Michal Hocko1af8efe2011-07-26 16:08:24 -07001706 * Has to be called with memcg_oom_lock
Michal Hocko79dfdac2011-07-26 16:08:23 -07001707 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001708static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001709{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001710 struct mem_cgroup *iter;
1711
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001712 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001713 iter->oom_lock = false;
1714 return 0;
1715}
1716
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001717static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001718{
1719 struct mem_cgroup *iter;
1720
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001721 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001722 atomic_inc(&iter->under_oom);
1723}
1724
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001725static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001726{
1727 struct mem_cgroup *iter;
1728
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001729 /*
1730 * When a new child is created while the hierarchy is under oom,
1731 * mem_cgroup_oom_lock() may not be called. We have to use
1732 * atomic_add_unless() here.
1733 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001734 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001735 atomic_add_unless(&iter->under_oom, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001736}
1737
Michal Hocko1af8efe2011-07-26 16:08:24 -07001738static DEFINE_SPINLOCK(memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001739static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1740
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001741struct oom_wait_info {
1742 struct mem_cgroup *mem;
1743 wait_queue_t wait;
1744};
1745
1746static int memcg_oom_wake_function(wait_queue_t *wait,
1747 unsigned mode, int sync, void *arg)
1748{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001749 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg,
1750 *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001751 struct oom_wait_info *oom_wait_info;
1752
1753 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001754 oom_wait_memcg = oom_wait_info->mem;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001755
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001756 /*
1757 * Both of oom_wait_info->mem and wake_mem are stable under us.
1758 * Then we can use css_is_ancestor without taking care of RCU.
1759 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001760 if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1761 && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001762 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001763 return autoremove_wake_function(wait, mode, sync, arg);
1764}
1765
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001766static void memcg_wakeup_oom(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001767{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001768 /* for filtering, pass "memcg" as argument. */
1769 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001770}
1771
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001772static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001773{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001774 if (memcg && atomic_read(&memcg->under_oom))
1775 memcg_wakeup_oom(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001776}
1777
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001778/*
1779 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1780 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001781bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001782{
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001783 struct oom_wait_info owait;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001784 bool locked, need_to_kill;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001785
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001786 owait.mem = memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001787 owait.wait.flags = 0;
1788 owait.wait.func = memcg_oom_wake_function;
1789 owait.wait.private = current;
1790 INIT_LIST_HEAD(&owait.wait.task_list);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001791 need_to_kill = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001792 mem_cgroup_mark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001793
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001794 /* At first, try to OOM lock hierarchy under memcg.*/
Michal Hocko1af8efe2011-07-26 16:08:24 -07001795 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001796 locked = mem_cgroup_oom_lock(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001797 /*
1798 * Even if signal_pending(), we can't quit charge() loop without
1799 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1800 * under OOM is always welcomed, use TASK_KILLABLE here.
1801 */
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001802 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001803 if (!locked || memcg->oom_kill_disable)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001804 need_to_kill = false;
1805 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001806 mem_cgroup_oom_notify(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001807 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001808
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001809 if (need_to_kill) {
1810 finish_wait(&memcg_oom_waitq, &owait.wait);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001811 mem_cgroup_out_of_memory(memcg, mask);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001812 } else {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001813 schedule();
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001814 finish_wait(&memcg_oom_waitq, &owait.wait);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001815 }
Michal Hocko1af8efe2011-07-26 16:08:24 -07001816 spin_lock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001817 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001818 mem_cgroup_oom_unlock(memcg);
1819 memcg_wakeup_oom(memcg);
Michal Hocko1af8efe2011-07-26 16:08:24 -07001820 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001821
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001822 mem_cgroup_unmark_under_oom(memcg);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001823
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001824 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1825 return false;
1826 /* Give chance to dying process */
KAMEZAWA Hiroyuki715a5ee2011-11-02 13:38:18 -07001827 schedule_timeout_uninterruptible(1);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001828 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001829}
1830
Balbir Singhd69b0422009-06-17 16:26:34 -07001831/*
1832 * Currently used to update mapped file statistics, but the routine can be
1833 * generalized to update other statistics as well.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001834 *
1835 * Notes: Race condition
1836 *
1837 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1838 * it tends to be costly. But considering some conditions, we doesn't need
1839 * to do so _always_.
1840 *
1841 * Considering "charge", lock_page_cgroup() is not required because all
1842 * file-stat operations happen after a page is attached to radix-tree. There
1843 * are no race with "charge".
1844 *
1845 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1846 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1847 * if there are race with "uncharge". Statistics itself is properly handled
1848 * by flags.
1849 *
1850 * Considering "move", this is an only case we see a race. To make the race
1851 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1852 * possibility of race condition. If there is, we take a lock.
Balbir Singhd69b0422009-06-17 16:26:34 -07001853 */
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001854
Greg Thelen2a7106f2011-01-13 15:47:37 -08001855void mem_cgroup_update_page_stat(struct page *page,
1856 enum mem_cgroup_page_stat_item idx, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001857{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001858 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001859 struct page_cgroup *pc = lookup_page_cgroup(page);
1860 bool need_unlock = false;
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001861 unsigned long uninitialized_var(flags);
Balbir Singhd69b0422009-06-17 16:26:34 -07001862
Johannes Weinercfa44942012-01-12 17:18:38 -08001863 if (mem_cgroup_disabled())
Balbir Singhd69b0422009-06-17 16:26:34 -07001864 return;
1865
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001866 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001867 memcg = pc->mem_cgroup;
1868 if (unlikely(!memcg || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001869 goto out;
1870 /* pc->mem_cgroup is unstable ? */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001871 if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) {
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001872 /* take a lock against to access pc->mem_cgroup */
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001873 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001874 need_unlock = true;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001875 memcg = pc->mem_cgroup;
1876 if (!memcg || !PageCgroupUsed(pc))
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001877 goto out;
1878 }
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001879
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001880 switch (idx) {
Greg Thelen2a7106f2011-01-13 15:47:37 -08001881 case MEMCG_NR_FILE_MAPPED:
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001882 if (val > 0)
1883 SetPageCgroupFileMapped(pc);
1884 else if (!page_mapped(page))
KAMEZAWA Hiroyuki0c270f82010-10-27 15:33:39 -07001885 ClearPageCgroupFileMapped(pc);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001886 idx = MEM_CGROUP_STAT_FILE_MAPPED;
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001887 break;
1888 default:
1889 BUG();
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07001890 }
Balbir Singhd69b0422009-06-17 16:26:34 -07001891
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 this_cpu_add(memcg->stat->count[idx], val);
Greg Thelen2a7106f2011-01-13 15:47:37 -08001893
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001894out:
1895 if (unlikely(need_unlock))
KAMEZAWA Hiroyukidbd4ea72011-01-13 15:47:38 -08001896 move_unlock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001897 rcu_read_unlock();
1898 return;
Balbir Singhd69b0422009-06-17 16:26:34 -07001899}
Greg Thelen2a7106f2011-01-13 15:47:37 -08001900EXPORT_SYMBOL(mem_cgroup_update_page_stat);
KAMEZAWA Hiroyuki26174ef2010-10-27 15:33:43 -07001901
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001902/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001903 * size of first charge trial. "32" comes from vmscan.c's magic value.
1904 * TODO: maybe necessary to use big numbers in big irons.
1905 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001906#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001907struct memcg_stock_pcp {
1908 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001909 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001910 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001911 unsigned long flags;
1912#define FLUSHING_CACHED_CHARGE (0)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001913};
1914static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001915static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001916
1917/*
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001918 * Try to consume stocked charge on this cpu. If success, one page is consumed
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001919 * from local stock and true is returned. If the stock is 0 or charges from a
1920 * cgroup which is not current target, returns false. This stock will be
1921 * refilled.
1922 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001923static bool consume_stock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001924{
1925 struct memcg_stock_pcp *stock;
1926 bool ret = true;
1927
1928 stock = &get_cpu_var(memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001929 if (memcg == stock->cached && stock->nr_pages)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001930 stock->nr_pages--;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001931 else /* need to call res_counter_charge */
1932 ret = false;
1933 put_cpu_var(memcg_stock);
1934 return ret;
1935}
1936
1937/*
1938 * Returns stocks cached in percpu to res_counter and reset cached information.
1939 */
1940static void drain_stock(struct memcg_stock_pcp *stock)
1941{
1942 struct mem_cgroup *old = stock->cached;
1943
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001944 if (stock->nr_pages) {
1945 unsigned long bytes = stock->nr_pages * PAGE_SIZE;
1946
1947 res_counter_uncharge(&old->res, bytes);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001948 if (do_swap_account)
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001949 res_counter_uncharge(&old->memsw, bytes);
1950 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001951 }
1952 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001953}
1954
1955/*
1956 * This must be called under preempt disabled or must be called by
1957 * a thread which is pinned to local cpu.
1958 */
1959static void drain_local_stock(struct work_struct *dummy)
1960{
1961 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1962 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001963 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001964}
1965
1966/*
1967 * Cache charges(val) which is from res_counter, to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001968 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001969 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001970static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001971{
1972 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1973
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001974 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001975 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001976 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001977 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001978 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001979 put_cpu_var(memcg_stock);
1980}
1981
1982/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001983 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Michal Hockod38144b2011-07-26 16:08:28 -07001984 * of the hierarchy under it. sync flag says whether we should block
1985 * until the work is done.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001986 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001987static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001988{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001989 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001990
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001991 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001992 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001993 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001994 for_each_online_cpu(cpu) {
1995 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001996 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001997
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001998 memcg = stock->cached;
1999 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002000 continue;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002001 if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07002002 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07002003 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2004 if (cpu == curcpu)
2005 drain_local_stock(&stock->work);
2006 else
2007 schedule_work_on(cpu, &stock->work);
2008 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002009 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002010 put_cpu();
Michal Hockod38144b2011-07-26 16:08:28 -07002011
2012 if (!sync)
2013 goto out;
2014
2015 for_each_online_cpu(cpu) {
2016 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002017 if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
Michal Hockod38144b2011-07-26 16:08:28 -07002018 flush_work(&stock->work);
2019 }
2020out:
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002021 put_online_cpus();
Michal Hockod38144b2011-07-26 16:08:28 -07002022}
2023
2024/*
2025 * Tries to drain stocked charges in other cpus. This function is asynchronous
2026 * and just put a work per cpu for draining localy on each cpu. Caller can
2027 * expects some charges will be back to res_counter later but cannot wait for
2028 * it.
2029 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002030static void drain_all_stock_async(struct mem_cgroup *root_memcg)
Michal Hockod38144b2011-07-26 16:08:28 -07002031{
Michal Hocko9f50fad2011-08-09 11:56:26 +02002032 /*
2033 * If someone calls draining, avoid adding more kworker runs.
2034 */
2035 if (!mutex_trylock(&percpu_charge_mutex))
2036 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002037 drain_all_stock(root_memcg, false);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002038 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002039}
2040
2041/* This is a synchronous drain interface. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002042static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002043{
2044 /* called when force_empty is called */
Michal Hocko9f50fad2011-08-09 11:56:26 +02002045 mutex_lock(&percpu_charge_mutex);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002046 drain_all_stock(root_memcg, true);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002047 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002048}
2049
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002050/*
2051 * This function drains percpu counter value from DEAD cpu and
2052 * move it to local cpu. Note that this function can be preempted.
2053 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002054static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002055{
2056 int i;
2057
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002058 spin_lock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002059 for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002060 long x = per_cpu(memcg->stat->count[i], cpu);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002061
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002062 per_cpu(memcg->stat->count[i], cpu) = 0;
2063 memcg->nocpu_base.count[i] += x;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002064 }
Johannes Weinere9f89742011-03-23 16:42:37 -07002065 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002066 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -07002067
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002068 per_cpu(memcg->stat->events[i], cpu) = 0;
2069 memcg->nocpu_base.events[i] += x;
Johannes Weinere9f89742011-03-23 16:42:37 -07002070 }
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002071 /* need to clear ON_MOVE value, works as a kind of lock. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002072 per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
2073 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002074}
2075
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002076static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002077{
2078 int idx = MEM_CGROUP_ON_MOVE;
2079
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002080 spin_lock(&memcg->pcp_counter_lock);
2081 per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx];
2082 spin_unlock(&memcg->pcp_counter_lock);
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002083}
2084
2085static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002086 unsigned long action,
2087 void *hcpu)
2088{
2089 int cpu = (unsigned long)hcpu;
2090 struct memcg_stock_pcp *stock;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002091 struct mem_cgroup *iter;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002092
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002093 if ((action == CPU_ONLINE)) {
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002094 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07002095 synchronize_mem_cgroup_on_move(iter, cpu);
2096 return NOTIFY_OK;
2097 }
2098
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002099 if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002100 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002101
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08002102 for_each_mem_cgroup(iter)
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07002103 mem_cgroup_drain_pcp_counter(iter, cpu);
2104
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002105 stock = &per_cpu(memcg_stock, cpu);
2106 drain_stock(stock);
2107 return NOTIFY_OK;
2108}
2109
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002110
2111/* See __mem_cgroup_try_charge() for details */
2112enum {
2113 CHARGE_OK, /* success */
2114 CHARGE_RETRY, /* need to retry but retry is not bad */
2115 CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2116 CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2117 CHARGE_OOM_DIE, /* the current is killed because of OOM */
2118};
2119
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002120static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002121 unsigned int nr_pages, bool oom_check)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002122{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002123 unsigned long csize = nr_pages * PAGE_SIZE;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002124 struct mem_cgroup *mem_over_limit;
2125 struct res_counter *fail_res;
2126 unsigned long flags = 0;
2127 int ret;
2128
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002129 ret = res_counter_charge(&memcg->res, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002130
2131 if (likely(!ret)) {
2132 if (!do_swap_account)
2133 return CHARGE_OK;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002134 ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002135 if (likely(!ret))
2136 return CHARGE_OK;
2137
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002138 res_counter_uncharge(&memcg->res, csize);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002139 mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2140 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2141 } else
2142 mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
Johannes Weiner9221edb2011-02-01 15:52:42 -08002143 /*
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002144 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2145 * of regular pages (CHARGE_BATCH), or a single regular page (1).
Johannes Weiner9221edb2011-02-01 15:52:42 -08002146 *
2147 * Never reclaim on behalf of optional batching, retry with a
2148 * single page instead.
2149 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002150 if (nr_pages == CHARGE_BATCH)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002151 return CHARGE_RETRY;
2152
2153 if (!(gfp_mask & __GFP_WAIT))
2154 return CHARGE_WOULDBLOCK;
2155
Johannes Weiner56600482012-01-12 17:17:59 -08002156 ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002157 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner19942822011-02-01 15:52:43 -08002158 return CHARGE_RETRY;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002159 /*
Johannes Weiner19942822011-02-01 15:52:43 -08002160 * Even though the limit is exceeded at this point, reclaim
2161 * may have been able to free some pages. Retry the charge
2162 * before killing the task.
2163 *
2164 * Only for regular pages, though: huge pages are rather
2165 * unlikely to succeed so close to the limit, and we fall back
2166 * to regular pages anyway in case of failure.
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002167 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002168 if (nr_pages == 1 && ret)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002169 return CHARGE_RETRY;
2170
2171 /*
2172 * At task move, charge accounts can be doubly counted. So, it's
2173 * better to wait until the end of task_move if something is going on.
2174 */
2175 if (mem_cgroup_wait_acct_move(mem_over_limit))
2176 return CHARGE_RETRY;
2177
2178 /* If we don't need to call oom-killer at el, return immediately */
2179 if (!oom_check)
2180 return CHARGE_NOMEM;
2181 /* check OOM */
2182 if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
2183 return CHARGE_OOM_DIE;
2184
2185 return CHARGE_RETRY;
2186}
2187
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002188/*
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002189 * __mem_cgroup_try_charge() does
2190 * 1. detect memcg to be charged against from passed *mm and *ptr,
2191 * 2. update res_counter
2192 * 3. call memory reclaim if necessary.
2193 *
2194 * In some special case, if the task is fatal, fatal_signal_pending() or
2195 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2196 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2197 * as possible without any hazards. 2: all pages should have a valid
2198 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2199 * pointer, that is treated as a charge to root_mem_cgroup.
2200 *
2201 * So __mem_cgroup_try_charge() will return
2202 * 0 ... on success, filling *ptr with a valid memcg pointer.
2203 * -ENOMEM ... charge failure because of resource limits.
2204 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2205 *
2206 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2207 * the oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002208 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002209static int __mem_cgroup_try_charge(struct mm_struct *mm,
Andrea Arcangeliec168512011-01-13 15:46:56 -08002210 gfp_t gfp_mask,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002211 unsigned int nr_pages,
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002212 struct mem_cgroup **ptr,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002213 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002214{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002215 unsigned int batch = max(CHARGE_BATCH, nr_pages);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002216 int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002217 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002218 int ret;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002219
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002220 /*
2221 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2222 * in system level. So, allow to go ahead dying process in addition to
2223 * MEMDIE process.
2224 */
2225 if (unlikely(test_thread_flag(TIF_MEMDIE)
2226 || fatal_signal_pending(current)))
2227 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002228
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002229 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08002230 * We always charge the cgroup the mm_struct belongs to.
2231 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002232 * thread group leader migrates. It's possible that mm is not
2233 * set, if so charge the init_mm (happens for pagecache usage).
2234 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002235 if (!*ptr && !mm)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002236 *ptr = root_mem_cgroup;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002237again:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002238 if (*ptr) { /* css should be a valid one */
2239 memcg = *ptr;
2240 VM_BUG_ON(css_is_removed(&memcg->css));
2241 if (mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002242 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002243 if (nr_pages == 1 && consume_stock(memcg))
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002244 goto done;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002245 css_get(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002246 } else {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002247 struct task_struct *p;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002248
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002249 rcu_read_lock();
2250 p = rcu_dereference(mm->owner);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002251 /*
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002252 * Because we don't have task_lock(), "p" can exit.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002253 * In that case, "memcg" can point to root or p can be NULL with
KAMEZAWA Hiroyukiebb76ce2010-12-29 14:07:11 -08002254 * race with swapoff. Then, we have small risk of mis-accouning.
2255 * But such kind of mis-account by race always happens because
2256 * we don't have cgroup_mutex(). It's overkill and we allo that
2257 * small race, here.
2258 * (*) swapoff at el will charge against mm-struct not against
2259 * task-struct. So, mm->owner can be NULL.
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002260 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002261 memcg = mem_cgroup_from_task(p);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002262 if (!memcg)
2263 memcg = root_mem_cgroup;
2264 if (mem_cgroup_is_root(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002265 rcu_read_unlock();
2266 goto done;
2267 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002268 if (nr_pages == 1 && consume_stock(memcg)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002269 /*
2270 * It seems dagerous to access memcg without css_get().
2271 * But considering how consume_stok works, it's not
2272 * necessary. If consume_stock success, some charges
2273 * from this memcg are cached on this cpu. So, we
2274 * don't need to call css_get()/css_tryget() before
2275 * calling consume_stock().
2276 */
2277 rcu_read_unlock();
2278 goto done;
2279 }
2280 /* after here, we may be blocked. we need to get refcnt */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002281 if (!css_tryget(&memcg->css)) {
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002282 rcu_read_unlock();
2283 goto again;
2284 }
2285 rcu_read_unlock();
2286 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002287
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002288 do {
2289 bool oom_check;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002290
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002291 /* If killed, bypass charge */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002292 if (fatal_signal_pending(current)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002293 css_put(&memcg->css);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002294 goto bypass;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002295 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002296
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002297 oom_check = false;
2298 if (oom && !nr_oom_retries) {
2299 oom_check = true;
2300 nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2301 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002302
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002303 ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002304 switch (ret) {
2305 case CHARGE_OK:
2306 break;
2307 case CHARGE_RETRY: /* not in OOM situation but retry */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002308 batch = nr_pages;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002309 css_put(&memcg->css);
2310 memcg = NULL;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002311 goto again;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002312 case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002313 css_put(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002314 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002315 case CHARGE_NOMEM: /* OOM routine works */
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002316 if (!oom) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002317 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002318 goto nomem;
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002319 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002320 /* If oom, we never return -ENOMEM */
2321 nr_oom_retries--;
2322 break;
2323 case CHARGE_OOM_DIE: /* Killed by OOM Killer */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002324 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002325 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08002326 }
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002327 } while (ret != CHARGE_OK);
2328
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002329 if (batch > nr_pages)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002330 refill_stock(memcg, batch - nr_pages);
2331 css_put(&memcg->css);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002332done:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002333 *ptr = memcg;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002334 return 0;
2335nomem:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002336 *ptr = NULL;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002337 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002338bypass:
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002339 *ptr = root_mem_cgroup;
2340 return -EINTR;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002341}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002342
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002343/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002344 * Somemtimes we have to undo a charge we got by try_charge().
2345 * This function is for that and do uncharge, put css's refcnt.
2346 * gotten by try_charge().
2347 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002348static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
Johannes Weinere7018b82011-03-23 16:42:33 -07002349 unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002350{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002351 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinere7018b82011-03-23 16:42:33 -07002352 unsigned long bytes = nr_pages * PAGE_SIZE;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002353
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002354 res_counter_uncharge(&memcg->res, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002355 if (do_swap_account)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002356 res_counter_uncharge(&memcg->memsw, bytes);
Johannes Weinere7018b82011-03-23 16:42:33 -07002357 }
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002358}
2359
2360/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002361 * A helper function to get mem_cgroup from ID. must be called under
2362 * rcu_read_lock(). The caller must check css_is_removed() or some if
2363 * it's concern. (dropping refcnt from swap can be called against removed
2364 * memcg.)
2365 */
2366static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2367{
2368 struct cgroup_subsys_state *css;
2369
2370 /* ID 0 is unused ID */
2371 if (!id)
2372 return NULL;
2373 css = css_lookup(&mem_cgroup_subsys, id);
2374 if (!css)
2375 return NULL;
2376 return container_of(css, struct mem_cgroup, css);
2377}
2378
Wu Fengguange42d9d52009-12-16 12:19:59 +01002379struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002380{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002381 struct mem_cgroup *memcg = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002382 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002383 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002384 swp_entry_t ent;
2385
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002386 VM_BUG_ON(!PageLocked(page));
2387
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002388 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002389 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002390 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002391 memcg = pc->mem_cgroup;
2392 if (memcg && !css_tryget(&memcg->css))
2393 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002394 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002395 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002396 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002397 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002398 memcg = mem_cgroup_lookup(id);
2399 if (memcg && !css_tryget(&memcg->css))
2400 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002401 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002402 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07002403 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002404 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002405}
2406
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002407static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
Johannes Weiner5564e882011-03-23 16:42:29 -07002408 struct page *page,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002409 unsigned int nr_pages,
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002410 struct page_cgroup *pc,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002411 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002412{
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002413 lock_page_cgroup(pc);
2414 if (unlikely(PageCgroupUsed(pc))) {
2415 unlock_page_cgroup(pc);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002416 __mem_cgroup_cancel_charge(memcg, nr_pages);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002417 return;
2418 }
2419 /*
2420 * we don't need page_cgroup_lock about tail pages, becase they are not
2421 * accessed by any other context at this point.
2422 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002423 pc->mem_cgroup = memcg;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07002424 /*
2425 * We access a page_cgroup asynchronously without lock_page_cgroup().
2426 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2427 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2428 * before USED bit, we need memory barrier here.
2429 * See mem_cgroup_add_lru_list(), etc.
2430 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002431 smp_wmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -07002432 switch (ctype) {
2433 case MEM_CGROUP_CHARGE_TYPE_CACHE:
2434 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2435 SetPageCgroupCache(pc);
2436 SetPageCgroupUsed(pc);
2437 break;
2438 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2439 ClearPageCgroupCache(pc);
2440 SetPageCgroupUsed(pc);
2441 break;
2442 default:
2443 break;
2444 }
Hugh Dickins3be91272008-02-07 00:14:19 -08002445
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002446 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002447 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002448 WARN_ON_ONCE(PageLRU(page));
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002449 /*
2450 * "charge_statistics" updated event counter. Then, check it.
2451 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2452 * if they exceeds softlimit.
2453 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002454 memcg_check_events(memcg, page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002455}
2456
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002457#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2458
2459#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002460 (1 << PCG_MIGRATION))
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002461/*
2462 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002463 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2464 * charge/uncharge will be never happen and move_account() is done under
2465 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002466 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002467void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002468{
2469 struct page_cgroup *head_pc = lookup_page_cgroup(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002470 struct page_cgroup *pc;
2471 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002472
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002473 if (mem_cgroup_disabled())
2474 return;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002475 for (i = 1; i < HPAGE_PMD_NR; i++) {
2476 pc = head_pc + i;
2477 pc->mem_cgroup = head_pc->mem_cgroup;
2478 smp_wmb();/* see __commit_charge() */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002479 pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2480 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002481}
Hugh Dickins12d27102012-01-12 17:19:52 -08002482#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002483
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002484/**
Johannes Weinerde3638d2011-03-23 16:42:28 -07002485 * mem_cgroup_move_account - move account of the page
Johannes Weiner5564e882011-03-23 16:42:29 -07002486 * @page: the page
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002487 * @nr_pages: number of regular pages (>1 for huge pages)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002488 * @pc: page_cgroup of the page.
2489 * @from: mem_cgroup which the page is moved from.
2490 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002491 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002492 *
2493 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002494 * - page is not on LRU (isolate_page() is useful.)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002495 * - compound_lock is held when nr_pages > 1
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002496 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002497 * This function doesn't do "charge" nor css_get to new cgroup. It should be
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002498 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002499 * true, this function does "uncharge" from old cgroup, but it doesn't if
2500 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002501 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002502static int mem_cgroup_move_account(struct page *page,
2503 unsigned int nr_pages,
2504 struct page_cgroup *pc,
2505 struct mem_cgroup *from,
2506 struct mem_cgroup *to,
2507 bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002508{
Johannes Weinerde3638d2011-03-23 16:42:28 -07002509 unsigned long flags;
2510 int ret;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002511
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002512 VM_BUG_ON(from == to);
Johannes Weiner5564e882011-03-23 16:42:29 -07002513 VM_BUG_ON(PageLRU(page));
Johannes Weinerde3638d2011-03-23 16:42:28 -07002514 /*
2515 * The page is isolated from LRU. So, collapse function
2516 * will not handle this page. But page splitting can happen.
2517 * Do this check under compound_page_lock(). The caller should
2518 * hold it.
2519 */
2520 ret = -EBUSY;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002521 if (nr_pages > 1 && !PageTransHuge(page))
Johannes Weinerde3638d2011-03-23 16:42:28 -07002522 goto out;
2523
2524 lock_page_cgroup(pc);
2525
2526 ret = -EINVAL;
2527 if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2528 goto unlock;
2529
2530 move_lock_page_cgroup(pc, &flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002531
KAMEZAWA Hiroyuki8725d542010-04-06 14:35:05 -07002532 if (PageCgroupFileMapped(pc)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002533 /* Update mapped_file data for mem_cgroup */
2534 preempt_disable();
2535 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2536 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2537 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07002538 }
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002539 mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002540 if (uncharge)
2541 /* This is not "cancel", but cancel_charge does all we need. */
Johannes Weinere7018b82011-03-23 16:42:33 -07002542 __mem_cgroup_cancel_charge(from, nr_pages);
Balbir Singhd69b0422009-06-17 16:26:34 -07002543
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002544 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002545 pc->mem_cgroup = to;
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002546 mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002547 /*
2548 * We charges against "to" which may not have any tasks. Then, "to"
2549 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002550 * this function is just force_empty() and move charge, so it's
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002551 * guaranteed that "to" is never removed. So, we don't check rmdir
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08002552 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002553 */
Johannes Weinerde3638d2011-03-23 16:42:28 -07002554 move_unlock_page_cgroup(pc, &flags);
2555 ret = 0;
2556unlock:
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002557 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002558 /*
2559 * check events
2560 */
Johannes Weiner5564e882011-03-23 16:42:29 -07002561 memcg_check_events(to, page);
2562 memcg_check_events(from, page);
Johannes Weinerde3638d2011-03-23 16:42:28 -07002563out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002564 return ret;
2565}
2566
2567/*
2568 * move charges to its parent.
2569 */
2570
Johannes Weiner5564e882011-03-23 16:42:29 -07002571static int mem_cgroup_move_parent(struct page *page,
2572 struct page_cgroup *pc,
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002573 struct mem_cgroup *child,
2574 gfp_t gfp_mask)
2575{
2576 struct cgroup *cg = child->css.cgroup;
2577 struct cgroup *pcg = cg->parent;
2578 struct mem_cgroup *parent;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002579 unsigned int nr_pages;
Andrew Morton4be44892011-03-23 16:42:39 -07002580 unsigned long uninitialized_var(flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002581 int ret;
2582
2583 /* Is ROOT ? */
2584 if (!pcg)
2585 return -EINVAL;
2586
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002587 ret = -EBUSY;
2588 if (!get_page_unless_zero(page))
2589 goto out;
2590 if (isolate_lru_page(page))
2591 goto put;
KAMEZAWA Hiroyuki52dbb902011-01-25 15:07:29 -08002592
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002593 nr_pages = hpage_nr_pages(page);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002594
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002595 parent = mem_cgroup_from_cont(pcg);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002596 ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002597 if (ret)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002598 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002599
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002600 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002601 flags = compound_lock_irqsave(page);
2602
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002603 ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08002604 if (ret)
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002605 __mem_cgroup_cancel_charge(parent, nr_pages);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002606
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002607 if (nr_pages > 1)
KAMEZAWA Hiroyuki987eba62011-01-20 14:44:25 -08002608 compound_unlock_irqrestore(page, flags);
Jesper Juhl8dba4742011-01-25 15:07:24 -08002609put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002610 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002611put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08002612 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08002613out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002614 return ret;
2615}
2616
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002617/*
2618 * Charge the memory controller for page usage.
2619 * Return
2620 * 0 if the charge was successful
2621 * < 0 if the cgroup is over its limit
2622 */
2623static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002624 gfp_t gfp_mask, enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002625{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002626 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002627 unsigned int nr_pages = 1;
Johannes Weiner8493ae42011-02-01 15:52:44 -08002628 struct page_cgroup *pc;
2629 bool oom = true;
2630 int ret;
Andrea Arcangeliec168512011-01-13 15:46:56 -08002631
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002632 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002633 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002634 VM_BUG_ON(!PageTransHuge(page));
Johannes Weiner8493ae42011-02-01 15:52:44 -08002635 /*
2636 * Never OOM-kill a process for a huge page. The
2637 * fault handler will fall back to regular pages.
2638 */
2639 oom = false;
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002640 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002641
2642 pc = lookup_page_cgroup(page);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002643 ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002644 if (ret == -ENOMEM)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002645 return ret;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002646 __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002647 return 0;
2648}
2649
2650int mem_cgroup_newpage_charge(struct page *page,
2651 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002652{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002653 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002654 return 0;
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002655 VM_BUG_ON(page_mapped(page));
2656 VM_BUG_ON(page->mapping && !PageAnon(page));
2657 VM_BUG_ON(!mm);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002658 return mem_cgroup_charge_common(page, mm, gfp_mask,
Johannes Weiner7a0524c2012-01-12 17:18:43 -08002659 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08002660}
2661
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002662static void
2663__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2664 enum charge_type ctype);
2665
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002666static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002667__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002668 enum charge_type ctype)
2669{
2670 struct page_cgroup *pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002671 struct zone *zone = page_zone(page);
2672 unsigned long flags;
2673 bool removed = false;
2674
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002675 /*
2676 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2677 * is already on LRU. It means the page may on some other page_cgroup's
2678 * LRU. Take care of it.
2679 */
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002680 spin_lock_irqsave(&zone->lru_lock, flags);
2681 if (PageLRU(page)) {
2682 del_page_from_lru_list(zone, page, page_lru(page));
2683 ClearPageLRU(page);
2684 removed = true;
2685 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002686 __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype);
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08002687 if (removed) {
2688 add_page_to_lru_list(zone, page, page_lru(page));
2689 SetPageLRU(page);
2690 }
2691 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002692 return;
2693}
2694
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002695int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2696 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08002697{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002698 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002699 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002700 int ret;
2701
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002702 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002703 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002704 if (PageCompound(page))
2705 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002706
Daisuke Nishimura73045c42010-08-10 18:02:59 -07002707 if (unlikely(!mm))
Balbir Singh8697d332008-02-07 00:13:59 -08002708 mm = &init_mm;
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002709 if (!page_is_file_cache(page))
2710 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07002711
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002712 if (!PageSwapCache(page))
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002713 ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002714 else { /* page is swapcache/shmem */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002715 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002716 if (!ret)
KAMEZAWA Hiroyukidc67d502012-01-12 17:18:55 -08002717 __mem_cgroup_commit_charge_swapin(page, memcg, type);
2718 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002719 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002720}
2721
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002722/*
2723 * While swap-in, try_charge -> commit or cancel, the page is locked.
2724 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002725 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002726 * "commit()" or removed by "cancel()"
2727 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002728int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2729 struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08002730 gfp_t mask, struct mem_cgroup **memcgp)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002731{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002732 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002733 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002734
Johannes Weiner72835c82012-01-12 17:18:32 -08002735 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002736
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002737 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002738 return 0;
2739
2740 if (!do_swap_account)
2741 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002742 /*
2743 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002744 * the pte, and even removed page from swap cache: in those cases
2745 * do_swap_page()'s pte_same() test will fail; but there's also a
2746 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002747 */
2748 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002749 goto charge_cur_mm;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002750 memcg = try_get_mem_cgroup_from_page(page);
2751 if (!memcg)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002752 goto charge_cur_mm;
Johannes Weiner72835c82012-01-12 17:18:32 -08002753 *memcgp = memcg;
2754 ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002755 css_put(&memcg->css);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002756 if (ret == -EINTR)
2757 ret = 0;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002758 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002759charge_cur_mm:
2760 if (unlikely(!mm))
2761 mm = &init_mm;
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08002762 ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2763 if (ret == -EINTR)
2764 ret = 0;
2765 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002766}
2767
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002768static void
Johannes Weiner72835c82012-01-12 17:18:32 -08002769__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002770 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002771{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002772 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002773 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002774 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002775 return;
Johannes Weiner72835c82012-01-12 17:18:32 -08002776 cgroup_exclude_rmdir(&memcg->css);
KAMEZAWA Hiroyuki5a6475a2011-03-23 16:42:42 -07002777
Johannes Weiner72835c82012-01-12 17:18:32 -08002778 __mem_cgroup_commit_charge_lrucare(page, memcg, ctype);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002779 /*
2780 * Now swap is on-memory. This means this page may be
2781 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002782 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2783 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2784 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002785 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002786 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002787 swp_entry_t ent = {.val = page_private(page)};
Johannes Weiner72835c82012-01-12 17:18:32 -08002788 struct mem_cgroup *swap_memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002789 unsigned short id;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002790
2791 id = swap_cgroup_record(ent, 0);
2792 rcu_read_lock();
Johannes Weiner72835c82012-01-12 17:18:32 -08002793 swap_memcg = mem_cgroup_lookup(id);
2794 if (swap_memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002795 /*
2796 * This recorded memcg can be obsolete one. So, avoid
2797 * calling css_tryget
2798 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002799 if (!mem_cgroup_is_root(swap_memcg))
2800 res_counter_uncharge(&swap_memcg->memsw,
2801 PAGE_SIZE);
2802 mem_cgroup_swap_statistics(swap_memcg, false);
2803 mem_cgroup_put(swap_memcg);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002804 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002805 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002806 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002807 /*
2808 * At swapin, we may charge account against cgroup which has no tasks.
2809 * So, rmdir()->pre_destroy() can be called while we do this charge.
2810 * In that case, we need to call pre_destroy() again. check it here.
2811 */
Johannes Weiner72835c82012-01-12 17:18:32 -08002812 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002813}
2814
Johannes Weiner72835c82012-01-12 17:18:32 -08002815void mem_cgroup_commit_charge_swapin(struct page *page,
2816 struct mem_cgroup *memcg)
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002817{
Johannes Weiner72835c82012-01-12 17:18:32 -08002818 __mem_cgroup_commit_charge_swapin(page, memcg,
2819 MEM_CGROUP_CHARGE_TYPE_MAPPED);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002820}
2821
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002822void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002823{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002824 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002825 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002826 if (!memcg)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002827 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002828 __mem_cgroup_cancel_charge(memcg, 1);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002829}
2830
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002831static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002832 unsigned int nr_pages,
2833 const enum charge_type ctype)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002834{
2835 struct memcg_batch_info *batch = NULL;
2836 bool uncharge_memsw = true;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002837
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002838 /* If swapout, usage of swap doesn't decrease */
2839 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2840 uncharge_memsw = false;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002841
2842 batch = &current->memcg_batch;
2843 /*
2844 * In usual, we do css_get() when we remember memcg pointer.
2845 * But in this case, we keep res->usage until end of a series of
2846 * uncharges. Then, it's ok to ignore memcg's refcnt.
2847 */
2848 if (!batch->memcg)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002849 batch->memcg = memcg;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002850 /*
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002851 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002852 * In those cases, all pages freed continuously can be expected to be in
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002853 * the same cgroup and we have chance to coalesce uncharges.
2854 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2855 * because we want to do uncharge as soon as possible.
2856 */
2857
2858 if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2859 goto direct_uncharge;
2860
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002861 if (nr_pages > 1)
Andrea Arcangeliec168512011-01-13 15:46:56 -08002862 goto direct_uncharge;
2863
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002864 /*
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002865 * In typical case, batch->memcg == mem. This means we can
2866 * merge a series of uncharges to an uncharge of res_counter.
2867 * If not, we uncharge res_counter ony by one.
2868 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002869 if (batch->memcg != memcg)
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002870 goto direct_uncharge;
2871 /* remember freed charge and uncharge it later */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002872 batch->nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002873 if (uncharge_memsw)
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002874 batch->memsw_nr_pages++;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002875 return;
2876direct_uncharge:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002877 res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002878 if (uncharge_memsw)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002879 res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2880 if (unlikely(batch->memcg != memcg))
2881 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002882 return;
2883}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002884
Balbir Singh8697d332008-02-07 00:13:59 -08002885/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002886 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002887 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002888static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002889__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002890{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002891 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002892 unsigned int nr_pages = 1;
2893 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002894
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002895 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002896 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002897
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002898 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002899 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002900
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002901 if (PageTransHuge(page)) {
Johannes Weiner7ec99d62011-03-23 16:42:36 -07002902 nr_pages <<= compound_order(page);
Andrea Arcangeli37c2ac72011-01-13 15:47:16 -08002903 VM_BUG_ON(!PageTransHuge(page));
2904 }
Balbir Singh8697d332008-02-07 00:13:59 -08002905 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002906 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002907 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002908 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08002909 if (unlikely(!PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002910 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002911
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002912 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002913
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002914 memcg = pc->mem_cgroup;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002915
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002916 if (!PageCgroupUsed(pc))
2917 goto unlock_out;
2918
2919 switch (ctype) {
2920 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002921 case MEM_CGROUP_CHARGE_TYPE_DROP:
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07002922 /* See mem_cgroup_prepare_migration() */
2923 if (page_mapped(page) || PageCgroupMigration(pc))
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002924 goto unlock_out;
2925 break;
2926 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2927 if (!PageAnon(page)) { /* Shared memory */
2928 if (page->mapping && !page_is_file_cache(page))
2929 goto unlock_out;
2930 } else if (page_mapped(page)) /* Anon */
2931 goto unlock_out;
2932 break;
2933 default:
2934 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002935 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002936
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002937 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002938
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002939 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002940 /*
2941 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2942 * freed from LRU. This is safe because uncharged page is expected not
2943 * to be reused (freed soon). Exception is SwapCache, it's handled by
2944 * special functions.
2945 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08002946
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002947 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002948 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002949 * even after unlock, we have memcg->res.usage here and this memcg
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002950 * will never be freed.
2951 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002952 memcg_check_events(memcg, page);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002953 if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002954 mem_cgroup_swap_statistics(memcg, true);
2955 mem_cgroup_get(memcg);
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07002956 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002957 if (!mem_cgroup_is_root(memcg))
2958 mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002959
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002960 return memcg;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002961
2962unlock_out:
2963 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002964 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08002965}
2966
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002967void mem_cgroup_uncharge_page(struct page *page)
2968{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002969 /* early check. */
2970 if (page_mapped(page))
2971 return;
Johannes Weiner40f23a22012-01-12 17:18:45 -08002972 VM_BUG_ON(page->mapping && !PageAnon(page));
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002973 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2974}
2975
2976void mem_cgroup_uncharge_cache_page(struct page *page)
2977{
2978 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07002979 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002980 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2981}
2982
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002983/*
2984 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2985 * In that cases, pages are freed continuously and we can expect pages
2986 * are in the same memcg. All these calls itself limits the number of
2987 * pages freed at once, then uncharge_start/end() is called properly.
2988 * This may be called prural(2) times in a context,
2989 */
2990
2991void mem_cgroup_uncharge_start(void)
2992{
2993 current->memcg_batch.do_batch++;
2994 /* We can do nest. */
2995 if (current->memcg_batch.do_batch == 1) {
2996 current->memcg_batch.memcg = NULL;
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07002997 current->memcg_batch.nr_pages = 0;
2998 current->memcg_batch.memsw_nr_pages = 0;
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002999 }
3000}
3001
3002void mem_cgroup_uncharge_end(void)
3003{
3004 struct memcg_batch_info *batch = &current->memcg_batch;
3005
3006 if (!batch->do_batch)
3007 return;
3008
3009 batch->do_batch--;
3010 if (batch->do_batch) /* If stacked, do nothing. */
3011 return;
3012
3013 if (!batch->memcg)
3014 return;
3015 /*
3016 * This "batch->memcg" is valid without any css_get/put etc...
3017 * bacause we hide charges behind us.
3018 */
Johannes Weiner7ffd4ca2011-03-23 16:42:35 -07003019 if (batch->nr_pages)
3020 res_counter_uncharge(&batch->memcg->res,
3021 batch->nr_pages * PAGE_SIZE);
3022 if (batch->memsw_nr_pages)
3023 res_counter_uncharge(&batch->memcg->memsw,
3024 batch->memsw_nr_pages * PAGE_SIZE);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003025 memcg_oom_recover(batch->memcg);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08003026 /* forget this pointer (for sanity check) */
3027 batch->memcg = NULL;
3028}
3029
KAMEZAWA Hiroyuki4e5f01c2012-01-12 17:18:58 -08003030/*
3031 * A function for resetting pc->mem_cgroup for newly allocated pages.
3032 * This function should be called if the newpage will be added to LRU
3033 * before start accounting.
3034 */
3035void mem_cgroup_reset_owner(struct page *newpage)
3036{
3037 struct page_cgroup *pc;
3038
3039 if (mem_cgroup_disabled())
3040 return;
3041
3042 pc = lookup_page_cgroup(newpage);
3043 VM_BUG_ON(PageCgroupUsed(pc));
3044 pc->mem_cgroup = root_mem_cgroup;
3045}
3046
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003047#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003048/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003049 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003050 * memcg information is recorded to swap_cgroup of "ent"
3051 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003052void
3053mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003054{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003055 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003056 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003057
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07003058 if (!swapout) /* this was a swap cache but the swap is unused ! */
3059 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3060
3061 memcg = __mem_cgroup_uncharge_common(page, ctype);
3062
KAMEZAWA Hiroyukif75ca962010-08-10 18:03:02 -07003063 /*
3064 * record memcg information, if swapout && memcg != NULL,
3065 * mem_cgroup_get() was called in uncharge().
3066 */
3067 if (do_swap_account && swapout && memcg)
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003068 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003069}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07003070#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08003071
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003072#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3073/*
3074 * called from swap_entry_free(). remove record in swap_cgroup and
3075 * uncharge "memsw" account.
3076 */
3077void mem_cgroup_uncharge_swap(swp_entry_t ent)
3078{
3079 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003080 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003081
3082 if (!do_swap_account)
3083 return;
3084
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003085 id = swap_cgroup_record(ent, 0);
3086 rcu_read_lock();
3087 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003088 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003089 /*
3090 * We uncharge this because swap is freed.
3091 * This memcg can be obsolete one. We avoid calling css_tryget
3092 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003093 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07003094 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003095 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003096 mem_cgroup_put(memcg);
3097 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07003098 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003099}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003100
3101/**
3102 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3103 * @entry: swap entry to be moved
3104 * @from: mem_cgroup which the entry is moved from
3105 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003106 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08003107 *
3108 * It succeeds only when the swap_cgroup's record for this entry is the same
3109 * as the mem_cgroup's id of @from.
3110 *
3111 * Returns 0 on success, -EINVAL on failure.
3112 *
3113 * The caller must have charged to @to, IOW, called res_counter_charge() about
3114 * both res and memsw, and called css_get().
3115 */
3116static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003117 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003118{
3119 unsigned short old_id, new_id;
3120
3121 old_id = css_id(&from->css);
3122 new_id = css_id(&to->css);
3123
3124 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08003125 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003126 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003127 /*
3128 * This function is only called from task migration context now.
3129 * It postpones res_counter and refcount handling till the end
3130 * of task migration(mem_cgroup_clear_mc()) for performance
3131 * improvement. But we cannot postpone mem_cgroup_get(to)
3132 * because if the process that has been moved to @to does
3133 * swap-in, the refcount of @to might be decreased to 0.
3134 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08003135 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003136 if (need_fixup) {
3137 if (!mem_cgroup_is_root(from))
3138 res_counter_uncharge(&from->memsw, PAGE_SIZE);
3139 mem_cgroup_put(from);
3140 /*
3141 * we charged both to->res and to->memsw, so we should
3142 * uncharge to->res.
3143 */
3144 if (!mem_cgroup_is_root(to))
3145 res_counter_uncharge(&to->res, PAGE_SIZE);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003146 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08003147 return 0;
3148 }
3149 return -EINVAL;
3150}
3151#else
3152static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003153 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003154{
3155 return -EINVAL;
3156}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003157#endif
3158
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003159/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003160 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3161 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003162 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003163int mem_cgroup_prepare_migration(struct page *page,
Johannes Weiner72835c82012-01-12 17:18:32 -08003164 struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003165{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003166 struct mem_cgroup *memcg = NULL;
Johannes Weiner7ec99d62011-03-23 16:42:36 -07003167 struct page_cgroup *pc;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003168 enum charge_type ctype;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003169 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003170
Johannes Weiner72835c82012-01-12 17:18:32 -08003171 *memcgp = NULL;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07003172
Andrea Arcangeliec168512011-01-13 15:46:56 -08003173 VM_BUG_ON(PageTransHuge(page));
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003174 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07003175 return 0;
3176
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003177 pc = lookup_page_cgroup(page);
3178 lock_page_cgroup(pc);
3179 if (PageCgroupUsed(pc)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003180 memcg = pc->mem_cgroup;
3181 css_get(&memcg->css);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003182 /*
3183 * At migrating an anonymous page, its mapcount goes down
3184 * to 0 and uncharge() will be called. But, even if it's fully
3185 * unmapped, migration may fail and this page has to be
3186 * charged again. We set MIGRATION flag here and delay uncharge
3187 * until end_migration() is called
3188 *
3189 * Corner Case Thinking
3190 * A)
3191 * When the old page was mapped as Anon and it's unmap-and-freed
3192 * while migration was ongoing.
3193 * If unmap finds the old page, uncharge() of it will be delayed
3194 * until end_migration(). If unmap finds a new page, it's
3195 * uncharged when it make mapcount to be 1->0. If unmap code
3196 * finds swap_migration_entry, the new page will not be mapped
3197 * and end_migration() will find it(mapcount==0).
3198 *
3199 * B)
3200 * When the old page was mapped but migraion fails, the kernel
3201 * remaps it. A charge for it is kept by MIGRATION flag even
3202 * if mapcount goes down to 0. We can do remap successfully
3203 * without charging it again.
3204 *
3205 * C)
3206 * The "old" page is under lock_page() until the end of
3207 * migration, so, the old page itself will not be swapped-out.
3208 * If the new page is swapped out before end_migraton, our
3209 * hook to usual swap-out path will catch the event.
3210 */
3211 if (PageAnon(page))
3212 SetPageCgroupMigration(pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08003213 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003214 unlock_page_cgroup(pc);
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003215 /*
3216 * If the page is not charged at this point,
3217 * we return here.
3218 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003219 if (!memcg)
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003220 return 0;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003221
Johannes Weiner72835c82012-01-12 17:18:32 -08003222 *memcgp = memcg;
3223 ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003224 css_put(&memcg->css);/* drop extra refcnt */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003225 if (ret) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003226 if (PageAnon(page)) {
3227 lock_page_cgroup(pc);
3228 ClearPageCgroupMigration(pc);
3229 unlock_page_cgroup(pc);
3230 /*
3231 * The old page may be fully unmapped while we kept it.
3232 */
3233 mem_cgroup_uncharge_page(page);
3234 }
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003235 /* we'll need to revisit this error code (we have -EINTR) */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003236 return -ENOMEM;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003237 }
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003238 /*
3239 * We charge new page before it's used/mapped. So, even if unlock_page()
3240 * is called before end_migration, we can catch all events on this new
3241 * page. In the case new page is migrated but not remapped, new page's
3242 * mapcount will be finally 0 and we call uncharge in end_migration().
3243 */
3244 pc = lookup_page_cgroup(newpage);
3245 if (PageAnon(page))
3246 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3247 else if (page_is_file_cache(page))
3248 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3249 else
3250 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
Johannes Weiner6568d4a2012-01-20 14:34:12 -08003251 __mem_cgroup_commit_charge(memcg, newpage, 1, pc, ctype);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003252 return ret;
3253}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08003254
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003255/* remove redundant charge if migration failed*/
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003256void mem_cgroup_end_migration(struct mem_cgroup *memcg,
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003257 struct page *oldpage, struct page *newpage, bool migration_ok)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07003258{
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003259 struct page *used, *unused;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003260 struct page_cgroup *pc;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003261
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003262 if (!memcg)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003263 return;
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003264 /* blocks rmdir() */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003265 cgroup_exclude_rmdir(&memcg->css);
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -08003266 if (!migration_ok) {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003267 used = oldpage;
3268 unused = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003269 } else {
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003270 used = newpage;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003271 unused = oldpage;
3272 }
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003273 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003274 * We disallowed uncharge of pages under migration because mapcount
3275 * of the page goes down to zero, temporarly.
3276 * Clear the flag and check the page should be charged.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07003277 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003278 pc = lookup_page_cgroup(oldpage);
3279 lock_page_cgroup(pc);
3280 ClearPageCgroupMigration(pc);
3281 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003282
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003283 __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3284
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003285 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003286 * If a page is a file cache, radix-tree replacement is very atomic
3287 * and we can skip this check. When it was an Anon page, its mapcount
3288 * goes down to 0. But because we added MIGRATION flage, it's not
3289 * uncharged yet. There are several case but page->mapcount check
3290 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3291 * check. (see prepare_charge() also)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08003292 */
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003293 if (PageAnon(used))
3294 mem_cgroup_uncharge_page(used);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003295 /*
akpm@linux-foundation.orgac39cf82010-05-26 14:42:46 -07003296 * At migration, we may charge account against cgroup which has no
3297 * tasks.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003298 * So, rmdir()->pre_destroy() can be called while we do this charge.
3299 * In that case, we need to call pre_destroy() again. check it here.
3300 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003301 cgroup_release_and_wakeup_rmdir(&memcg->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08003302}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003303
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003304/*
3305 * At replace page cache, newpage is not under any memcg but it's on
3306 * LRU. So, this function doesn't touch res_counter but handles LRU
3307 * in correct way. Both pages are locked so we cannot race with uncharge.
3308 */
3309void mem_cgroup_replace_page_cache(struct page *oldpage,
3310 struct page *newpage)
3311{
3312 struct mem_cgroup *memcg;
3313 struct page_cgroup *pc;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003314 enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003315
3316 if (mem_cgroup_disabled())
3317 return;
3318
3319 pc = lookup_page_cgroup(oldpage);
3320 /* fix accounting on old pages */
3321 lock_page_cgroup(pc);
3322 memcg = pc->mem_cgroup;
3323 mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -1);
3324 ClearPageCgroupUsed(pc);
3325 unlock_page_cgroup(pc);
3326
3327 if (PageSwapBacked(oldpage))
3328 type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3329
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003330 /*
3331 * Even if newpage->mapping was NULL before starting replacement,
3332 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3333 * LRU while we overwrite pc->mem_cgroup.
3334 */
KAMEZAWA Hiroyuki36b62ad2012-01-12 17:18:57 -08003335 __mem_cgroup_commit_charge_lrucare(newpage, memcg, type);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -08003336}
3337
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003338#ifdef CONFIG_DEBUG_VM
3339static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3340{
3341 struct page_cgroup *pc;
3342
3343 pc = lookup_page_cgroup(page);
Johannes Weinercfa44942012-01-12 17:18:38 -08003344 /*
3345 * Can be NULL while feeding pages into the page allocator for
3346 * the first time, i.e. during boot or memory hotplug;
3347 * or when mem_cgroup_disabled().
3348 */
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003349 if (likely(pc) && PageCgroupUsed(pc))
3350 return pc;
3351 return NULL;
3352}
3353
3354bool mem_cgroup_bad_page_check(struct page *page)
3355{
3356 if (mem_cgroup_disabled())
3357 return false;
3358
3359 return lookup_page_cgroup_used(page) != NULL;
3360}
3361
3362void mem_cgroup_print_bad_page(struct page *page)
3363{
3364 struct page_cgroup *pc;
3365
3366 pc = lookup_page_cgroup_used(page);
3367 if (pc) {
Hugh Dickins90b3fea2012-01-12 17:19:54 -08003368 printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003369 pc, pc->flags, pc->mem_cgroup);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003370 }
3371}
3372#endif
3373
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003374static DEFINE_MUTEX(set_limit_mutex);
3375
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08003376static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003377 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003378{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003379 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003380 u64 memswlimit, memlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003381 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003382 int children = mem_cgroup_count_children(memcg);
3383 u64 curusage, oldusage;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003384 int enlarge;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003385
3386 /*
3387 * For keeping hierarchical_reclaim simple, how long we should retry
3388 * is depends on callers. We set our retry-count to be function
3389 * of # of children which we should visit in this loop.
3390 */
3391 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3392
3393 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003394
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003395 enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003396 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003397 if (signal_pending(current)) {
3398 ret = -EINTR;
3399 break;
3400 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003401 /*
3402 * Rather than hide all in some function, I do this in
3403 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003404 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003405 */
3406 mutex_lock(&set_limit_mutex);
3407 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3408 if (memswlimit < val) {
3409 ret = -EINVAL;
3410 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003411 break;
3412 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003413
3414 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3415 if (memlimit < val)
3416 enlarge = 1;
3417
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003418 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003419 if (!ret) {
3420 if (memswlimit == val)
3421 memcg->memsw_is_minimum = true;
3422 else
3423 memcg->memsw_is_minimum = false;
3424 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003425 mutex_unlock(&set_limit_mutex);
3426
3427 if (!ret)
3428 break;
3429
Johannes Weiner56600482012-01-12 17:17:59 -08003430 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3431 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003432 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3433 /* Usage is reduced ? */
3434 if (curusage >= oldusage)
3435 retry_count--;
3436 else
3437 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003438 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003439 if (!ret && enlarge)
3440 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08003441
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003442 return ret;
3443}
3444
Li Zefan338c8432009-06-17 16:27:15 -07003445static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3446 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003447{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003448 int retry_count;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003449 u64 memlimit, memswlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003450 int children = mem_cgroup_count_children(memcg);
3451 int ret = -EBUSY;
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003452 int enlarge = 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003453
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003454 /* see mem_cgroup_resize_res_limit */
3455 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3456 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003457 while (retry_count) {
3458 if (signal_pending(current)) {
3459 ret = -EINTR;
3460 break;
3461 }
3462 /*
3463 * Rather than hide all in some function, I do this in
3464 * open coded manner. You see what this really does.
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003465 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003466 */
3467 mutex_lock(&set_limit_mutex);
3468 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3469 if (memlimit > val) {
3470 ret = -EINVAL;
3471 mutex_unlock(&set_limit_mutex);
3472 break;
3473 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003474 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3475 if (memswlimit < val)
3476 enlarge = 1;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003477 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07003478 if (!ret) {
3479 if (memlimit == val)
3480 memcg->memsw_is_minimum = true;
3481 else
3482 memcg->memsw_is_minimum = false;
3483 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003484 mutex_unlock(&set_limit_mutex);
3485
3486 if (!ret)
3487 break;
3488
Johannes Weiner56600482012-01-12 17:17:59 -08003489 mem_cgroup_reclaim(memcg, GFP_KERNEL,
3490 MEM_CGROUP_RECLAIM_NOSWAP |
3491 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003492 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003493 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003494 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003495 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003496 else
3497 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003498 }
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003499 if (!ret && enlarge)
3500 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003501 return ret;
3502}
3503
Balbir Singh4e416952009-09-23 15:56:39 -07003504unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
Ying Han0ae5e892011-05-26 16:25:25 -07003505 gfp_t gfp_mask,
3506 unsigned long *total_scanned)
Balbir Singh4e416952009-09-23 15:56:39 -07003507{
3508 unsigned long nr_reclaimed = 0;
3509 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3510 unsigned long reclaimed;
3511 int loop = 0;
3512 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003513 unsigned long long excess;
Ying Han0ae5e892011-05-26 16:25:25 -07003514 unsigned long nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003515
3516 if (order > 0)
3517 return 0;
3518
KOSAKI Motohiro00918b62010-08-10 18:03:05 -07003519 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
Balbir Singh4e416952009-09-23 15:56:39 -07003520 /*
3521 * This loop can run a while, specially if mem_cgroup's continuously
3522 * keep exceeding their soft limit and putting the system under
3523 * pressure
3524 */
3525 do {
3526 if (next_mz)
3527 mz = next_mz;
3528 else
3529 mz = mem_cgroup_largest_soft_limit_node(mctz);
3530 if (!mz)
3531 break;
3532
Ying Han0ae5e892011-05-26 16:25:25 -07003533 nr_scanned = 0;
Johannes Weiner56600482012-01-12 17:17:59 -08003534 reclaimed = mem_cgroup_soft_reclaim(mz->mem, zone,
3535 gfp_mask, &nr_scanned);
Balbir Singh4e416952009-09-23 15:56:39 -07003536 nr_reclaimed += reclaimed;
Ying Han0ae5e892011-05-26 16:25:25 -07003537 *total_scanned += nr_scanned;
Balbir Singh4e416952009-09-23 15:56:39 -07003538 spin_lock(&mctz->lock);
3539
3540 /*
3541 * If we failed to reclaim anything from this memory cgroup
3542 * it is time to move on to the next cgroup
3543 */
3544 next_mz = NULL;
3545 if (!reclaimed) {
3546 do {
3547 /*
3548 * Loop until we find yet another one.
3549 *
3550 * By the time we get the soft_limit lock
3551 * again, someone might have aded the
3552 * group back on the RB tree. Iterate to
3553 * make sure we get a different mem.
3554 * mem_cgroup_largest_soft_limit_node returns
3555 * NULL if no other cgroup is present on
3556 * the tree
3557 */
3558 next_mz =
3559 __mem_cgroup_largest_soft_limit_node(mctz);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003560 if (next_mz == mz)
Balbir Singh4e416952009-09-23 15:56:39 -07003561 css_put(&next_mz->mem->css);
Michal Hocko39cc98f2011-05-26 16:25:28 -07003562 else /* next_mz == NULL or other memcg */
Balbir Singh4e416952009-09-23 15:56:39 -07003563 break;
3564 } while (1);
3565 }
Balbir Singh4e416952009-09-23 15:56:39 -07003566 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003567 excess = res_counter_soft_limit_excess(&mz->mem->res);
Balbir Singh4e416952009-09-23 15:56:39 -07003568 /*
3569 * One school of thought says that we should not add
3570 * back the node to the tree if reclaim returns 0.
3571 * But our reclaim could return 0, simply because due
3572 * to priority we are exposing a smaller subset of
3573 * memory to reclaim from. Consider this as a longer
3574 * term TODO.
3575 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07003576 /* If excess == 0, no tree ops */
3577 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07003578 spin_unlock(&mctz->lock);
3579 css_put(&mz->mem->css);
3580 loop++;
3581 /*
3582 * Could not reclaim anything and there are no more
3583 * mem cgroups to try or we seem to be looping without
3584 * reclaiming anything.
3585 */
3586 if (!nr_reclaimed &&
3587 (next_mz == NULL ||
3588 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3589 break;
3590 } while (!nr_reclaimed);
3591 if (next_mz)
3592 css_put(&next_mz->mem->css);
3593 return nr_reclaimed;
3594}
3595
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07003596/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003597 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003598 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3599 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003600static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003601 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003602{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003603 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003604 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003605 struct list_head *list;
Johannes Weiner925b7672012-01-12 17:18:15 -08003606 struct page *busy;
3607 struct zone *zone;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003608 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08003609
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003610 zone = &NODE_DATA(node)->node_zones[zid];
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003611 mz = mem_cgroup_zoneinfo(memcg, node, zid);
Johannes Weiner6290df52012-01-12 17:18:10 -08003612 list = &mz->lruvec.lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003613
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003614 loop = MEM_CGROUP_ZSTAT(mz, lru);
3615 /* give some margin against EBUSY etc...*/
3616 loop += 256;
3617 busy = NULL;
3618 while (loop--) {
Johannes Weiner925b7672012-01-12 17:18:15 -08003619 struct page_cgroup *pc;
Johannes Weiner5564e882011-03-23 16:42:29 -07003620 struct page *page;
3621
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003622 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003623 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003624 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003625 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003626 break;
3627 }
Johannes Weiner925b7672012-01-12 17:18:15 -08003628 page = list_entry(list->prev, struct page, lru);
3629 if (busy == page) {
3630 list_move(&page->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08003631 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003632 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003633 continue;
3634 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003635 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003636
Johannes Weiner925b7672012-01-12 17:18:15 -08003637 pc = lookup_page_cgroup(page);
Johannes Weiner5564e882011-03-23 16:42:29 -07003638
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003639 ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08003640 if (ret == -ENOMEM || ret == -EINTR)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003641 break;
3642
3643 if (ret == -EBUSY || ret == -EINVAL) {
3644 /* found lock contention or "pc" is obsolete. */
Johannes Weiner925b7672012-01-12 17:18:15 -08003645 busy = page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003646 cond_resched();
3647 } else
3648 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003649 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003650
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003651 if (!ret && !list_empty(list))
3652 return -EBUSY;
3653 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003654}
3655
3656/*
3657 * make mem_cgroup's charge to be 0 if there is no task.
3658 * This enables deleting this mem_cgroup.
3659 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003660static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003661{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003662 int ret;
3663 int node, zid, shrink;
3664 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003665 struct cgroup *cgrp = memcg->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08003666
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003667 css_get(&memcg->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003668
3669 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003670 /* should free all ? */
3671 if (free_all)
3672 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003673move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003674 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003675 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003676 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003677 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003678 ret = -EINTR;
3679 if (signal_pending(current))
3680 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003681 /* This is for making all *used* pages to be on LRU. */
3682 lru_add_drain_all();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003683 drain_all_stock_sync(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003684 ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003685 mem_cgroup_start_move(memcg);
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08003686 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003687 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07003688 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003689 for_each_lru(l) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003690 ret = mem_cgroup_force_empty_list(memcg,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003691 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003692 if (ret)
3693 break;
3694 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003695 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003696 if (ret)
3697 break;
3698 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003699 mem_cgroup_end_move(memcg);
3700 memcg_oom_recover(memcg);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003701 /* it seems parent cgroup doesn't have enough mem */
3702 if (ret == -ENOMEM)
3703 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07003704 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003705 /* "ret" should also be checked to ensure all lists are empty. */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003706 } while (memcg->res.usage > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003707out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003708 css_put(&memcg->css);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003709 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003710
3711try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003712 /* returns EBUSY if there is a task or if we come here twice. */
3713 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003714 ret = -EBUSY;
3715 goto out;
3716 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003717 /* we call try-to-free pages for make this cgroup empty */
3718 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003719 /* try to free all pages in this cgroup */
3720 shrink = 1;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003721 while (nr_retries && memcg->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003722 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003723
3724 if (signal_pending(current)) {
3725 ret = -EINTR;
3726 goto out;
3727 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003728 progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
Johannes Weiner185efc02011-09-14 16:21:58 -07003729 false);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003730 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003731 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003732 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003733 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003734 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003735
3736 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003737 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003738 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08003739 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003740}
3741
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003742int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3743{
3744 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3745}
3746
3747
Balbir Singh18f59ea2009-01-07 18:08:07 -08003748static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3749{
3750 return mem_cgroup_from_cont(cont)->use_hierarchy;
3751}
3752
3753static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3754 u64 val)
3755{
3756 int retval = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003757 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003758 struct cgroup *parent = cont->parent;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003759 struct mem_cgroup *parent_memcg = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003760
3761 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003762 parent_memcg = mem_cgroup_from_cont(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003763
3764 cgroup_lock();
3765 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003766 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003767 * in the child subtrees. If it is unset, then the change can
3768 * occur, provided the current cgroup has no children.
3769 *
3770 * For the root cgroup, parent_mem is NULL, we allow value to be
3771 * set if there are no children.
3772 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003773 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003774 (val == 1 || val == 0)) {
3775 if (list_empty(&cont->children))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003776 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003777 else
3778 retval = -EBUSY;
3779 } else
3780 retval = -EINVAL;
3781 cgroup_unlock();
3782
3783 return retval;
3784}
3785
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003786
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003787static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003788 enum mem_cgroup_stat_index idx)
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003789{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003790 struct mem_cgroup *iter;
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003791 long val = 0;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003792
Johannes Weiner7a159cc2011-03-23 16:42:38 -07003793 /* Per-cpu values can be negative, use a signed accumulator */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003794 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003795 val += mem_cgroup_read_stat(iter, idx);
3796
3797 if (val < 0) /* race ? */
3798 val = 0;
3799 return val;
Balbir Singh0c3e73e2009-09-23 15:56:42 -07003800}
3801
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003802static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003803{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003804 u64 val;
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003805
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003806 if (!mem_cgroup_is_root(memcg)) {
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003807 if (!swap)
Glauber Costa65c64ce2011-12-22 01:02:27 +00003808 return res_counter_read_u64(&memcg->res, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003809 else
Glauber Costa65c64ce2011-12-22 01:02:27 +00003810 return res_counter_read_u64(&memcg->memsw, RES_USAGE);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003811 }
3812
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003813 val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3814 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003815
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003816 if (swap)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003817 val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003818
3819 return val << PAGE_SHIFT;
3820}
3821
Paul Menage2c3daa72008-04-29 00:59:58 -07003822static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003823{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003824 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003825 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003826 int type, name;
3827
3828 type = MEMFILE_TYPE(cft->private);
3829 name = MEMFILE_ATTR(cft->private);
3830 switch (type) {
3831 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003832 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003833 val = mem_cgroup_usage(memcg, false);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003834 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003835 val = res_counter_read_u64(&memcg->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003836 break;
3837 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003838 if (name == RES_USAGE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003839 val = mem_cgroup_usage(memcg, true);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08003840 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003841 val = res_counter_read_u64(&memcg->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003842 break;
3843 default:
3844 BUG();
3845 break;
3846 }
3847 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003848}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003849/*
3850 * The user of this function is...
3851 * RES_LIMIT.
3852 */
Paul Menage856c13a2008-07-25 01:47:04 -07003853static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3854 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003855{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003856 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003857 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003858 unsigned long long val;
3859 int ret;
3860
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003861 type = MEMFILE_TYPE(cft->private);
3862 name = MEMFILE_ATTR(cft->private);
3863 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003864 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003865 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3866 ret = -EINVAL;
3867 break;
3868 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003869 /* This function does all necessary parse...reuse it */
3870 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003871 if (ret)
3872 break;
3873 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003874 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003875 else
3876 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003877 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003878 case RES_SOFT_LIMIT:
3879 ret = res_counter_memparse_write_strategy(buffer, &val);
3880 if (ret)
3881 break;
3882 /*
3883 * For memsw, soft limits are hard to implement in terms
3884 * of semantics, for now, we support soft limits for
3885 * control without swap
3886 */
3887 if (type == _MEM)
3888 ret = res_counter_set_soft_limit(&memcg->res, val);
3889 else
3890 ret = -EINVAL;
3891 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003892 default:
3893 ret = -EINVAL; /* should be BUG() ? */
3894 break;
3895 }
3896 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003897}
3898
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003899static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3900 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3901{
3902 struct cgroup *cgroup;
3903 unsigned long long min_limit, min_memsw_limit, tmp;
3904
3905 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3906 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3907 cgroup = memcg->css.cgroup;
3908 if (!memcg->use_hierarchy)
3909 goto out;
3910
3911 while (cgroup->parent) {
3912 cgroup = cgroup->parent;
3913 memcg = mem_cgroup_from_cont(cgroup);
3914 if (!memcg->use_hierarchy)
3915 break;
3916 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3917 min_limit = min(min_limit, tmp);
3918 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3919 min_memsw_limit = min(min_memsw_limit, tmp);
3920 }
3921out:
3922 *mem_limit = min_limit;
3923 *memsw_limit = min_memsw_limit;
3924 return;
3925}
3926
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003927static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003928{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003929 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003930 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003931
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003932 memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003933 type = MEMFILE_TYPE(event);
3934 name = MEMFILE_ATTR(event);
3935 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003936 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003937 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003938 res_counter_reset_max(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003939 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003940 res_counter_reset_max(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003941 break;
3942 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003943 if (type == _MEM)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003944 res_counter_reset_failcnt(&memcg->res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003945 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003946 res_counter_reset_failcnt(&memcg->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003947 break;
3948 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003949
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07003950 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003951}
3952
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003953static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3954 struct cftype *cft)
3955{
3956 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3957}
3958
Daisuke Nishimura02491442010-03-10 15:22:17 -08003959#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003960static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3961 struct cftype *cft, u64 val)
3962{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003963 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003964
3965 if (val >= (1 << NR_MOVE_TYPE))
3966 return -EINVAL;
3967 /*
3968 * We check this value several times in both in can_attach() and
3969 * attach(), so we need cgroup lock to prevent this value from being
3970 * inconsistent.
3971 */
3972 cgroup_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003973 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003974 cgroup_unlock();
3975
3976 return 0;
3977}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003978#else
3979static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3980 struct cftype *cft, u64 val)
3981{
3982 return -ENOSYS;
3983}
3984#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003985
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003986
3987/* For read statistics */
3988enum {
3989 MCS_CACHE,
3990 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08003991 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003992 MCS_PGPGIN,
3993 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003994 MCS_SWAP,
Ying Han456f9982011-05-26 16:25:38 -07003995 MCS_PGFAULT,
3996 MCS_PGMAJFAULT,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003997 MCS_INACTIVE_ANON,
3998 MCS_ACTIVE_ANON,
3999 MCS_INACTIVE_FILE,
4000 MCS_ACTIVE_FILE,
4001 MCS_UNEVICTABLE,
4002 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004003};
4004
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004005struct mcs_total_stat {
4006 s64 stat[NR_MCS_STAT];
4007};
4008
4009struct {
4010 char *local_name;
4011 char *total_name;
4012} memcg_stat_strings[NR_MCS_STAT] = {
4013 {"cache", "total_cache"},
4014 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07004015 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004016 {"pgpgin", "total_pgpgin"},
4017 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004018 {"swap", "total_swap"},
Ying Han456f9982011-05-26 16:25:38 -07004019 {"pgfault", "total_pgfault"},
4020 {"pgmajfault", "total_pgmajfault"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004021 {"inactive_anon", "total_inactive_anon"},
4022 {"active_anon", "total_active_anon"},
4023 {"inactive_file", "total_inactive_file"},
4024 {"active_file", "total_active_file"},
4025 {"unevictable", "total_unevictable"}
4026};
4027
4028
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004029static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004030mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004031{
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004032 s64 val;
4033
4034 /* per cpu stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004035 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004036 s->stat[MCS_CACHE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004037 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004038 s->stat[MCS_RSS] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004039 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08004040 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004041 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004042 s->stat[MCS_PGPGIN] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004043 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004044 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004045 if (do_swap_account) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004046 val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004047 s->stat[MCS_SWAP] += val * PAGE_SIZE;
4048 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004049 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004050 s->stat[MCS_PGFAULT] += val;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004051 val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07004052 s->stat[MCS_PGMAJFAULT] += val;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004053
4054 /* per zone stat */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004055 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004056 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004057 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004058 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004059 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004060 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004061 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004062 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004063 val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004064 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004065}
4066
4067static void
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004068mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004069{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004070 struct mem_cgroup *iter;
4071
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004072 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004073 mem_cgroup_get_local_stat(iter, s);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004074}
4075
Ying Han406eb0c2011-05-26 16:25:37 -07004076#ifdef CONFIG_NUMA
4077static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4078{
4079 int nid;
4080 unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4081 unsigned long node_nr;
4082 struct cgroup *cont = m->private;
4083 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4084
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004085 total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004086 seq_printf(m, "total=%lu", total_nr);
4087 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004088 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL);
Ying Han406eb0c2011-05-26 16:25:37 -07004089 seq_printf(m, " N%d=%lu", nid, node_nr);
4090 }
4091 seq_putc(m, '\n');
4092
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004093 file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004094 seq_printf(m, "file=%lu", file_nr);
4095 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004096 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4097 LRU_ALL_FILE);
Ying Han406eb0c2011-05-26 16:25:37 -07004098 seq_printf(m, " N%d=%lu", nid, node_nr);
4099 }
4100 seq_putc(m, '\n');
4101
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004102 anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004103 seq_printf(m, "anon=%lu", anon_nr);
4104 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004105 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4106 LRU_ALL_ANON);
Ying Han406eb0c2011-05-26 16:25:37 -07004107 seq_printf(m, " N%d=%lu", nid, node_nr);
4108 }
4109 seq_putc(m, '\n');
4110
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004111 unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004112 seq_printf(m, "unevictable=%lu", unevictable_nr);
4113 for_each_node_state(nid, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -07004114 node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid,
4115 BIT(LRU_UNEVICTABLE));
Ying Han406eb0c2011-05-26 16:25:37 -07004116 seq_printf(m, " N%d=%lu", nid, node_nr);
4117 }
4118 seq_putc(m, '\n');
4119 return 0;
4120}
4121#endif /* CONFIG_NUMA */
4122
Paul Menagec64745c2008-04-29 01:00:02 -07004123static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4124 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004125{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004126 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004127 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004128 int i;
4129
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004130 memset(&mystat, 0, sizeof(mystat));
4131 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004132
Ying Han406eb0c2011-05-26 16:25:37 -07004133
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004134 for (i = 0; i < NR_MCS_STAT; i++) {
4135 if (i == MCS_SWAP && !do_swap_account)
4136 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004137 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004138 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004139
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004140 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004141 {
4142 unsigned long long limit, memsw_limit;
4143 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
4144 cb->fill(cb, "hierarchical_memory_limit", limit);
4145 if (do_swap_account)
4146 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4147 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004148
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004149 memset(&mystat, 0, sizeof(mystat));
4150 mem_cgroup_get_total_stat(mem_cont, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004151 for (i = 0; i < NR_MCS_STAT; i++) {
4152 if (i == MCS_SWAP && !do_swap_account)
4153 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004154 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004155 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004156
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004157#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004158 {
4159 int nid, zid;
4160 struct mem_cgroup_per_zone *mz;
4161 unsigned long recent_rotated[2] = {0, 0};
4162 unsigned long recent_scanned[2] = {0, 0};
4163
4164 for_each_online_node(nid)
4165 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4166 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
4167
4168 recent_rotated[0] +=
4169 mz->reclaim_stat.recent_rotated[0];
4170 recent_rotated[1] +=
4171 mz->reclaim_stat.recent_rotated[1];
4172 recent_scanned[0] +=
4173 mz->reclaim_stat.recent_scanned[0];
4174 recent_scanned[1] +=
4175 mz->reclaim_stat.recent_scanned[1];
4176 }
4177 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4178 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4179 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4180 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4181 }
4182#endif
4183
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004184 return 0;
4185}
4186
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004187static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4188{
4189 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4190
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004191 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004192}
4193
4194static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4195 u64 val)
4196{
4197 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4198 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08004199
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004200 if (val > 100)
4201 return -EINVAL;
4202
4203 if (cgrp->parent == NULL)
4204 return -EINVAL;
4205
4206 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08004207
4208 cgroup_lock();
4209
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004210 /* If under hierarchy, only empty-root can set this value */
4211 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08004212 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4213 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004214 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08004215 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004216
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004217 memcg->swappiness = val;
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004218
Li Zefan068b38c2009-01-15 13:51:26 -08004219 cgroup_unlock();
4220
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004221 return 0;
4222}
4223
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004224static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4225{
4226 struct mem_cgroup_threshold_ary *t;
4227 u64 usage;
4228 int i;
4229
4230 rcu_read_lock();
4231 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004232 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004233 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004234 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004235
4236 if (!t)
4237 goto unlock;
4238
4239 usage = mem_cgroup_usage(memcg, swap);
4240
4241 /*
4242 * current_threshold points to threshold just below usage.
4243 * If it's not true, a threshold was crossed after last
4244 * call of __mem_cgroup_threshold().
4245 */
Phil Carmody5407a562010-05-26 14:42:42 -07004246 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004247
4248 /*
4249 * Iterate backward over array of thresholds starting from
4250 * current_threshold and check if a threshold is crossed.
4251 * If none of thresholds below usage is crossed, we read
4252 * only one element of the array here.
4253 */
4254 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4255 eventfd_signal(t->entries[i].eventfd, 1);
4256
4257 /* i = current_threshold + 1 */
4258 i++;
4259
4260 /*
4261 * Iterate forward over array of thresholds starting from
4262 * current_threshold+1 and check if a threshold is crossed.
4263 * If none of thresholds above usage is crossed, we read
4264 * only one element of the array here.
4265 */
4266 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4267 eventfd_signal(t->entries[i].eventfd, 1);
4268
4269 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004270 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004271unlock:
4272 rcu_read_unlock();
4273}
4274
4275static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4276{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004277 while (memcg) {
4278 __mem_cgroup_threshold(memcg, false);
4279 if (do_swap_account)
4280 __mem_cgroup_threshold(memcg, true);
4281
4282 memcg = parent_mem_cgroup(memcg);
4283 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004284}
4285
4286static int compare_thresholds(const void *a, const void *b)
4287{
4288 const struct mem_cgroup_threshold *_a = a;
4289 const struct mem_cgroup_threshold *_b = b;
4290
4291 return _a->threshold - _b->threshold;
4292}
4293
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004294static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004295{
4296 struct mem_cgroup_eventfd_list *ev;
4297
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004298 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004299 eventfd_signal(ev->eventfd, 1);
4300 return 0;
4301}
4302
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004303static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004304{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004305 struct mem_cgroup *iter;
4306
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004307 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004308 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004309}
4310
4311static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4312 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004313{
4314 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004315 struct mem_cgroup_thresholds *thresholds;
4316 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004317 int type = MEMFILE_TYPE(cft->private);
4318 u64 threshold, usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004319 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004320
4321 ret = res_counter_memparse_write_strategy(args, &threshold);
4322 if (ret)
4323 return ret;
4324
4325 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004326
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004327 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004328 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004329 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004330 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004331 else
4332 BUG();
4333
4334 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4335
4336 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004337 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004338 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4339
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004340 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004341
4342 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004343 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004344 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004345 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004346 ret = -ENOMEM;
4347 goto unlock;
4348 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004349 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004350
4351 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004352 if (thresholds->primary) {
4353 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004354 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004355 }
4356
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004357 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004358 new->entries[size - 1].eventfd = eventfd;
4359 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004360
4361 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004362 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004363 compare_thresholds, NULL);
4364
4365 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004366 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004367 for (i = 0; i < size; i++) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004368 if (new->entries[i].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004369 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004370 * new->current_threshold will not be used until
4371 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004372 * it here.
4373 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004374 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004375 }
4376 }
4377
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004378 /* Free old spare buffer and save old primary buffer as spare */
4379 kfree(thresholds->spare);
4380 thresholds->spare = thresholds->primary;
4381
4382 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004383
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004384 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004385 synchronize_rcu();
4386
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004387unlock:
4388 mutex_unlock(&memcg->thresholds_lock);
4389
4390 return ret;
4391}
4392
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004393static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004394 struct cftype *cft, struct eventfd_ctx *eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004395{
4396 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 struct mem_cgroup_thresholds *thresholds;
4398 struct mem_cgroup_threshold_ary *new;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004399 int type = MEMFILE_TYPE(cft->private);
4400 u64 usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004401 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004402
4403 mutex_lock(&memcg->thresholds_lock);
4404 if (type == _MEM)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004405 thresholds = &memcg->thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004406 else if (type == _MEMSWAP)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004407 thresholds = &memcg->memsw_thresholds;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004408 else
4409 BUG();
4410
4411 /*
4412 * Something went wrong if we trying to unregister a threshold
4413 * if we don't have thresholds
4414 */
4415 BUG_ON(!thresholds);
4416
Anton Vorontsov371528c2012-02-24 05:14:46 +04004417 if (!thresholds->primary)
4418 goto unlock;
4419
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004420 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4421
4422 /* Check if a threshold crossed before removing */
4423 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4424
4425 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004426 size = 0;
4427 for (i = 0; i < thresholds->primary->size; i++) {
4428 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004429 size++;
4430 }
4431
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004432 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004433
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004434 /* Set thresholds array to NULL if we don't have thresholds */
4435 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004436 kfree(new);
4437 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004438 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004439 }
4440
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004441 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004442
4443 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004444 new->current_threshold = -1;
4445 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4446 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004447 continue;
4448
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004449 new->entries[j] = thresholds->primary->entries[i];
4450 if (new->entries[j].threshold < usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004451 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004452 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004453 * until rcu_assign_pointer(), so it's safe to increment
4454 * it here.
4455 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004456 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004457 }
4458 j++;
4459 }
4460
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004461swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004462 /* Swap primary and spare array */
4463 thresholds->spare = thresholds->primary;
4464 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004465
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004466 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004467 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04004468unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004469 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004470}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004471
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004472static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4473 struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4474{
4475 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4476 struct mem_cgroup_eventfd_list *event;
4477 int type = MEMFILE_TYPE(cft->private);
4478
4479 BUG_ON(type != _OOM_TYPE);
4480 event = kmalloc(sizeof(*event), GFP_KERNEL);
4481 if (!event)
4482 return -ENOMEM;
4483
Michal Hocko1af8efe2011-07-26 16:08:24 -07004484 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004485
4486 event->eventfd = eventfd;
4487 list_add(&event->list, &memcg->oom_notify);
4488
4489 /* already in OOM ? */
Michal Hocko79dfdac2011-07-26 16:08:23 -07004490 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004491 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004492 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004493
4494 return 0;
4495}
4496
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004497static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004498 struct cftype *cft, struct eventfd_ctx *eventfd)
4499{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004500 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004501 struct mem_cgroup_eventfd_list *ev, *tmp;
4502 int type = MEMFILE_TYPE(cft->private);
4503
4504 BUG_ON(type != _OOM_TYPE);
4505
Michal Hocko1af8efe2011-07-26 16:08:24 -07004506 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004507
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004508 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004509 if (ev->eventfd == eventfd) {
4510 list_del(&ev->list);
4511 kfree(ev);
4512 }
4513 }
4514
Michal Hocko1af8efe2011-07-26 16:08:24 -07004515 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004516}
4517
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004518static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4519 struct cftype *cft, struct cgroup_map_cb *cb)
4520{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004521 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004522
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004523 cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004524
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004525 if (atomic_read(&memcg->under_oom))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004526 cb->fill(cb, "under_oom", 1);
4527 else
4528 cb->fill(cb, "under_oom", 0);
4529 return 0;
4530}
4531
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004532static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4533 struct cftype *cft, u64 val)
4534{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004535 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004536 struct mem_cgroup *parent;
4537
4538 /* cannot set to root cgroup and only 0 and 1 are allowed */
4539 if (!cgrp->parent || !((val == 0) || (val == 1)))
4540 return -EINVAL;
4541
4542 parent = mem_cgroup_from_cont(cgrp->parent);
4543
4544 cgroup_lock();
4545 /* oom-kill-disable is a flag for subhierarchy. */
4546 if ((parent->use_hierarchy) ||
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004547 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004548 cgroup_unlock();
4549 return -EINVAL;
4550 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004551 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004552 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004553 memcg_oom_recover(memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004554 cgroup_unlock();
4555 return 0;
4556}
4557
Ying Han406eb0c2011-05-26 16:25:37 -07004558#ifdef CONFIG_NUMA
4559static const struct file_operations mem_control_numa_stat_file_operations = {
4560 .read = seq_read,
4561 .llseek = seq_lseek,
4562 .release = single_release,
4563};
4564
4565static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4566{
4567 struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4568
4569 file->f_op = &mem_control_numa_stat_file_operations;
4570 return single_open(file, mem_control_numa_stat_show, cont);
4571}
4572#endif /* CONFIG_NUMA */
4573
Glauber Costae5671df2011-12-11 21:47:01 +00004574#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
Glauber Costae5671df2011-12-11 21:47:01 +00004575static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4576{
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004577 /*
4578 * Part of this would be better living in a separate allocation
4579 * function, leaving us with just the cgroup tree population work.
4580 * We, however, depend on state such as network's proto_list that
4581 * is only initialized after cgroup creation. I found the less
4582 * cumbersome way to deal with it to defer it all to populate time
4583 */
Glauber Costa65c64ce2011-12-22 01:02:27 +00004584 return mem_cgroup_sockets_init(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004585};
4586
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004587static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4588 struct cgroup *cont)
4589{
4590 mem_cgroup_sockets_destroy(cont, ss);
4591}
Glauber Costae5671df2011-12-11 21:47:01 +00004592#else
4593static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss)
4594{
4595 return 0;
4596}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004597
4598static void kmem_cgroup_destroy(struct cgroup_subsys *ss,
4599 struct cgroup *cont)
4600{
4601}
Glauber Costae5671df2011-12-11 21:47:01 +00004602#endif
4603
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004604static struct cftype mem_cgroup_files[] = {
4605 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004606 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004607 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07004608 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004609 .register_event = mem_cgroup_usage_register_event,
4610 .unregister_event = mem_cgroup_usage_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004611 },
4612 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004613 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004614 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004615 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004616 .read_u64 = mem_cgroup_read,
4617 },
4618 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004619 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004620 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07004621 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07004622 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004623 },
4624 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004625 .name = "soft_limit_in_bytes",
4626 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4627 .write_string = mem_cgroup_write,
4628 .read_u64 = mem_cgroup_read,
4629 },
4630 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004631 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004632 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004633 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07004634 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004635 },
Balbir Singh8697d332008-02-07 00:13:59 -08004636 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004637 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07004638 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004639 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004640 {
4641 .name = "force_empty",
4642 .trigger = mem_cgroup_force_empty_write,
4643 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004644 {
4645 .name = "use_hierarchy",
4646 .write_u64 = mem_cgroup_hierarchy_write,
4647 .read_u64 = mem_cgroup_hierarchy_read,
4648 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004649 {
4650 .name = "swappiness",
4651 .read_u64 = mem_cgroup_swappiness_read,
4652 .write_u64 = mem_cgroup_swappiness_write,
4653 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004654 {
4655 .name = "move_charge_at_immigrate",
4656 .read_u64 = mem_cgroup_move_charge_read,
4657 .write_u64 = mem_cgroup_move_charge_write,
4658 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004659 {
4660 .name = "oom_control",
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004661 .read_map = mem_cgroup_oom_control_read,
4662 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004663 .register_event = mem_cgroup_oom_register_event,
4664 .unregister_event = mem_cgroup_oom_unregister_event,
4665 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4666 },
Ying Han406eb0c2011-05-26 16:25:37 -07004667#ifdef CONFIG_NUMA
4668 {
4669 .name = "numa_stat",
4670 .open = mem_control_numa_stat_open,
KAMEZAWA Hiroyuki89577122011-06-15 15:08:41 -07004671 .mode = S_IRUGO,
Ying Han406eb0c2011-05-26 16:25:37 -07004672 },
4673#endif
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004674};
4675
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004676#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4677static struct cftype memsw_cgroup_files[] = {
4678 {
4679 .name = "memsw.usage_in_bytes",
4680 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4681 .read_u64 = mem_cgroup_read,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004682 .register_event = mem_cgroup_usage_register_event,
4683 .unregister_event = mem_cgroup_usage_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004684 },
4685 {
4686 .name = "memsw.max_usage_in_bytes",
4687 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4688 .trigger = mem_cgroup_reset,
4689 .read_u64 = mem_cgroup_read,
4690 },
4691 {
4692 .name = "memsw.limit_in_bytes",
4693 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4694 .write_string = mem_cgroup_write,
4695 .read_u64 = mem_cgroup_read,
4696 },
4697 {
4698 .name = "memsw.failcnt",
4699 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4700 .trigger = mem_cgroup_reset,
4701 .read_u64 = mem_cgroup_read,
4702 },
4703};
4704
4705static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4706{
4707 if (!do_swap_account)
4708 return 0;
4709 return cgroup_add_files(cont, ss, memsw_cgroup_files,
4710 ARRAY_SIZE(memsw_cgroup_files));
4711};
4712#else
4713static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4714{
4715 return 0;
4716}
4717#endif
4718
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004719static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004720{
4721 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004722 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07004723 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004724 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004725 /*
4726 * This routine is called against possible nodes.
4727 * But it's BUG to call kmalloc() against offline node.
4728 *
4729 * TODO: this routine can waste much memory for nodes which will
4730 * never be onlined. It's better to use memory hotplug callback
4731 * function.
4732 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004733 if (!node_state(node, N_NORMAL_MEMORY))
4734 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004735 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004736 if (!pn)
4737 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004738
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004739 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4740 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07004741 for_each_lru(l)
Johannes Weiner6290df52012-01-12 17:18:10 -08004742 INIT_LIST_HEAD(&mz->lruvec.lists[l]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07004743 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07004744 mz->on_tree = false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004745 mz->mem = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004746 }
Igor Mammedov0a619e52011-11-02 13:38:21 -07004747 memcg->info.nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004748 return 0;
4749}
4750
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004751static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004752{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004753 kfree(memcg->info.nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004754}
4755
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004756static struct mem_cgroup *mem_cgroup_alloc(void)
4757{
4758 struct mem_cgroup *mem;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004759 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004760
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004761 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08004762 if (size < PAGE_SIZE)
Jesper Juhl17295c82011-01-13 15:47:42 -08004763 mem = kzalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004764 else
Jesper Juhl17295c82011-01-13 15:47:42 -08004765 mem = vzalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004766
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004767 if (!mem)
4768 return NULL;
4769
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004770 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004771 if (!mem->stat)
4772 goto out_free;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004773 spin_lock_init(&mem->pcp_counter_lock);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004774 return mem;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004775
4776out_free:
4777 if (size < PAGE_SIZE)
4778 kfree(mem);
4779 else
4780 vfree(mem);
4781 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004782}
4783
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004784/*
4785 * At destroying mem_cgroup, references from swap_cgroup can remain.
4786 * (scanning all at force_empty is too costly...)
4787 *
4788 * Instead of clearing all references at force_empty, we remember
4789 * the number of reference from swap_cgroup and free mem_cgroup when
4790 * it goes down to 0.
4791 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004792 * Removal of cgroup itself succeeds regardless of refs from swap.
4793 */
4794
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004795static void __mem_cgroup_free(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004796{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004797 int node;
4798
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004799 mem_cgroup_remove_from_trees(memcg);
4800 free_css_id(&mem_cgroup_subsys, &memcg->css);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004801
Bob Liu3ed28fa2012-01-12 17:19:04 -08004802 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004803 free_mem_cgroup_per_zone_info(memcg, node);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08004804
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004805 free_percpu(memcg->stat);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08004806 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004807 kfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004808 else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004809 vfree(memcg);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004810}
4811
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004812static void mem_cgroup_get(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004813{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004814 atomic_inc(&memcg->refcnt);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004815}
4816
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004817static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004818{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004819 if (atomic_sub_and_test(count, &memcg->refcnt)) {
4820 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4821 __mem_cgroup_free(memcg);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004822 if (parent)
4823 mem_cgroup_put(parent);
4824 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004825}
4826
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004827static void mem_cgroup_put(struct mem_cgroup *memcg)
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004828{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004829 __mem_cgroup_put(memcg, 1);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004830}
4831
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004832/*
4833 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4834 */
Glauber Costae1aab162011-12-11 21:47:03 +00004835struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004836{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004837 if (!memcg->res.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004838 return NULL;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004839 return mem_cgroup_from_res_counter(memcg->res.parent, res);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004840}
Glauber Costae1aab162011-12-11 21:47:03 +00004841EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004842
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004843#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4844static void __init enable_swap_cgroup(void)
4845{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08004846 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004847 do_swap_account = 1;
4848}
4849#else
4850static void __init enable_swap_cgroup(void)
4851{
4852}
4853#endif
4854
Balbir Singhf64c3f52009-09-23 15:56:37 -07004855static int mem_cgroup_soft_limit_tree_init(void)
4856{
4857 struct mem_cgroup_tree_per_node *rtpn;
4858 struct mem_cgroup_tree_per_zone *rtpz;
4859 int tmp, node, zone;
4860
Bob Liu3ed28fa2012-01-12 17:19:04 -08004861 for_each_node(node) {
Balbir Singhf64c3f52009-09-23 15:56:37 -07004862 tmp = node;
4863 if (!node_state(node, N_NORMAL_MEMORY))
4864 tmp = -1;
4865 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4866 if (!rtpn)
Michal Hockoc3cecc62012-01-12 17:18:50 -08004867 goto err_cleanup;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004868
4869 soft_limit_tree.rb_tree_per_node[node] = rtpn;
4870
4871 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4872 rtpz = &rtpn->rb_tree_per_zone[zone];
4873 rtpz->rb_root = RB_ROOT;
4874 spin_lock_init(&rtpz->lock);
4875 }
4876 }
4877 return 0;
Michal Hockoc3cecc62012-01-12 17:18:50 -08004878
4879err_cleanup:
Bob Liu3ed28fa2012-01-12 17:19:04 -08004880 for_each_node(node) {
Michal Hockoc3cecc62012-01-12 17:18:50 -08004881 if (!soft_limit_tree.rb_tree_per_node[node])
4882 break;
4883 kfree(soft_limit_tree.rb_tree_per_node[node]);
4884 soft_limit_tree.rb_tree_per_node[node] = NULL;
4885 }
4886 return 1;
4887
Balbir Singhf64c3f52009-09-23 15:56:37 -07004888}
4889
Li Zefan0eb253e2009-01-15 13:51:25 -08004890static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004891mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4892{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004893 struct mem_cgroup *memcg, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004894 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004895 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004896
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004897 memcg = mem_cgroup_alloc();
4898 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004899 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004900
Bob Liu3ed28fa2012-01-12 17:19:04 -08004901 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004902 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004903 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004904
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004905 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004906 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004907 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004908 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004909 parent = NULL;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004910 if (mem_cgroup_soft_limit_tree_init())
4911 goto free_out;
Hillf Dantona41c58a2011-12-19 17:11:57 -08004912 root_mem_cgroup = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08004913 for_each_possible_cpu(cpu) {
4914 struct memcg_stock_pcp *stock =
4915 &per_cpu(memcg_stock, cpu);
4916 INIT_WORK(&stock->work, drain_local_stock);
4917 }
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07004918 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004919 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004920 parent = mem_cgroup_from_cont(cont->parent);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004921 memcg->use_hierarchy = parent->use_hierarchy;
4922 memcg->oom_kill_disable = parent->oom_kill_disable;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004923 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004924
Balbir Singh18f59ea2009-01-07 18:08:07 -08004925 if (parent && parent->use_hierarchy) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004926 res_counter_init(&memcg->res, &parent->res);
4927 res_counter_init(&memcg->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004928 /*
4929 * We increment refcnt of the parent to ensure that we can
4930 * safely access it on res_counter_charge/uncharge.
4931 * This refcnt will be decremented when freeing this
4932 * mem_cgroup(see mem_cgroup_put).
4933 */
4934 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004935 } else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004936 res_counter_init(&memcg->res, NULL);
4937 res_counter_init(&memcg->memsw, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004938 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004939 memcg->last_scanned_node = MAX_NUMNODES;
4940 INIT_LIST_HEAD(&memcg->oom_notify);
Balbir Singh6d61ef42009-01-07 18:08:06 -08004941
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004942 if (parent)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004943 memcg->swappiness = mem_cgroup_swappiness(parent);
4944 atomic_set(&memcg->refcnt, 1);
4945 memcg->move_charge_at_immigrate = 0;
4946 mutex_init(&memcg->thresholds_lock);
4947 return &memcg->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004948free_out:
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004949 __mem_cgroup_free(memcg);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004950 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004951}
4952
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004953static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004954 struct cgroup *cont)
4955{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004956 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004957
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004958 return mem_cgroup_force_empty(memcg, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004959}
4960
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004961static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4962 struct cgroup *cont)
4963{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004964 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004965
Glauber Costad1a4c0b2011-12-11 21:47:04 +00004966 kmem_cgroup_destroy(ss, cont);
4967
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004968 mem_cgroup_put(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004969}
4970
4971static int mem_cgroup_populate(struct cgroup_subsys *ss,
4972 struct cgroup *cont)
4973{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004974 int ret;
4975
4976 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4977 ARRAY_SIZE(mem_cgroup_files));
4978
4979 if (!ret)
4980 ret = register_memsw_files(cont, ss);
Glauber Costae5671df2011-12-11 21:47:01 +00004981
4982 if (!ret)
4983 ret = register_kmem_files(cont, ss);
4984
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08004985 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004986}
4987
Daisuke Nishimura02491442010-03-10 15:22:17 -08004988#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004989/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004990#define PRECHARGE_COUNT_AT_ONCE 256
4991static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004992{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004993 int ret = 0;
4994 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004995 struct mem_cgroup *memcg = mc.to;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004996
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004997 if (mem_cgroup_is_root(memcg)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004998 mc.precharge += count;
4999 /* we don't need css_get for root */
5000 return ret;
5001 }
5002 /* try to charge at once */
5003 if (count > 1) {
5004 struct res_counter *dummy;
5005 /*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005006 * "memcg" cannot be under rmdir() because we've already checked
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005007 * by cgroup_lock_live_cgroup() that it is not removed and we
5008 * are still under the same cgroup_mutex. So we can postpone
5009 * css_get().
5010 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005011 if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005012 goto one_by_one;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005013 if (do_swap_account && res_counter_charge(&memcg->memsw,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005014 PAGE_SIZE * count, &dummy)) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005015 res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005016 goto one_by_one;
5017 }
5018 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005019 return ret;
5020 }
5021one_by_one:
5022 /* fall back to one by one charge */
5023 while (count--) {
5024 if (signal_pending(current)) {
5025 ret = -EINTR;
5026 break;
5027 }
5028 if (!batch_count--) {
5029 batch_count = PRECHARGE_COUNT_AT_ONCE;
5030 cond_resched();
5031 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005032 ret = __mem_cgroup_try_charge(NULL,
5033 GFP_KERNEL, 1, &memcg, false);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005034 if (ret)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005035 /* mem_cgroup_clear_mc() will do uncharge later */
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005036 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005037 mc.precharge++;
5038 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005039 return ret;
5040}
5041
5042/**
5043 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5044 * @vma: the vma the pte to be checked belongs
5045 * @addr: the address corresponding to the pte to be checked
5046 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08005047 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005048 *
5049 * Returns
5050 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5051 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5052 * move charge. if @target is not NULL, the page is stored in target->page
5053 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08005054 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5055 * target for charge migration. if @target is not NULL, the entry is stored
5056 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005057 *
5058 * Called with pte lock held.
5059 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005060union mc_target {
5061 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005062 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005063};
5064
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005065enum mc_target_type {
5066 MC_TARGET_NONE, /* not used */
5067 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005068 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005069};
5070
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005071static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5072 unsigned long addr, pte_t ptent)
5073{
5074 struct page *page = vm_normal_page(vma, addr, ptent);
5075
5076 if (!page || !page_mapped(page))
5077 return NULL;
5078 if (PageAnon(page)) {
5079 /* we don't move shared anon */
5080 if (!move_anon() || page_mapcount(page) > 2)
5081 return NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005082 } else if (!move_file())
5083 /* we ignore mapcount for file pages */
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005084 return NULL;
5085 if (!get_page_unless_zero(page))
5086 return NULL;
5087
5088 return page;
5089}
5090
5091static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5092 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5093{
5094 int usage_count;
5095 struct page *page = NULL;
5096 swp_entry_t ent = pte_to_swp_entry(ptent);
5097
5098 if (!move_anon() || non_swap_entry(ent))
5099 return NULL;
5100 usage_count = mem_cgroup_count_swap_user(ent, &page);
5101 if (usage_count > 1) { /* we don't move shared anon */
5102 if (page)
5103 put_page(page);
5104 return NULL;
5105 }
5106 if (do_swap_account)
5107 entry->val = ent.val;
5108
5109 return page;
5110}
5111
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005112static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5113 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5114{
5115 struct page *page = NULL;
5116 struct inode *inode;
5117 struct address_space *mapping;
5118 pgoff_t pgoff;
5119
5120 if (!vma->vm_file) /* anonymous vma */
5121 return NULL;
5122 if (!move_file())
5123 return NULL;
5124
5125 inode = vma->vm_file->f_path.dentry->d_inode;
5126 mapping = vma->vm_file->f_mapping;
5127 if (pte_none(ptent))
5128 pgoff = linear_page_index(vma, addr);
5129 else /* pte_file(ptent) is true */
5130 pgoff = pte_to_pgoff(ptent);
5131
5132 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005133 page = find_get_page(mapping, pgoff);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005134
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005135#ifdef CONFIG_SWAP
5136 /* shmem/tmpfs may report page out on swap: account for that too. */
5137 if (radix_tree_exceptional_entry(page)) {
5138 swp_entry_t swap = radix_to_swp_entry(page);
5139 if (do_swap_account)
5140 *entry = swap;
5141 page = find_get_page(&swapper_space, swap.val);
5142 }
5143#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005144 return page;
5145}
5146
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005147static int is_target_pte_for_mc(struct vm_area_struct *vma,
5148 unsigned long addr, pte_t ptent, union mc_target *target)
5149{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005150 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005151 struct page_cgroup *pc;
5152 int ret = 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005153 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005154
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005155 if (pte_present(ptent))
5156 page = mc_handle_present_pte(vma, addr, ptent);
5157 else if (is_swap_pte(ptent))
5158 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005159 else if (pte_none(ptent) || pte_file(ptent))
5160 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005161
5162 if (!page && !ent.val)
Daisuke Nishimura02491442010-03-10 15:22:17 -08005163 return 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005164 if (page) {
5165 pc = lookup_page_cgroup(page);
5166 /*
5167 * Do only loose check w/o page_cgroup lock.
5168 * mem_cgroup_move_account() checks the pc is valid or not under
5169 * the lock.
5170 */
5171 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5172 ret = MC_TARGET_PAGE;
5173 if (target)
5174 target->page = page;
5175 }
5176 if (!ret || !target)
5177 put_page(page);
5178 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005179 /* There is a swap entry and a page doesn't exist or isn't charged */
5180 if (ent.val && !ret &&
Bob Liu9fb4b7c2012-01-12 17:18:48 -08005181 css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005182 ret = MC_TARGET_SWAP;
5183 if (target)
5184 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005185 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005186 return ret;
5187}
5188
5189static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5190 unsigned long addr, unsigned long end,
5191 struct mm_walk *walk)
5192{
5193 struct vm_area_struct *vma = walk->private;
5194 pte_t *pte;
5195 spinlock_t *ptl;
5196
Dave Hansen03319322011-03-22 16:32:56 -07005197 split_huge_page_pmd(walk->mm, pmd);
5198
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005199 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5200 for (; addr != end; pte++, addr += PAGE_SIZE)
5201 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
5202 mc.precharge++; /* increment precharge temporarily */
5203 pte_unmap_unlock(pte - 1, ptl);
5204 cond_resched();
5205
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005206 return 0;
5207}
5208
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005209static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5210{
5211 unsigned long precharge;
5212 struct vm_area_struct *vma;
5213
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005214 down_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005215 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5216 struct mm_walk mem_cgroup_count_precharge_walk = {
5217 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5218 .mm = mm,
5219 .private = vma,
5220 };
5221 if (is_vm_hugetlb_page(vma))
5222 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005223 walk_page_range(vma->vm_start, vma->vm_end,
5224 &mem_cgroup_count_precharge_walk);
5225 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005226 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005227
5228 precharge = mc.precharge;
5229 mc.precharge = 0;
5230
5231 return precharge;
5232}
5233
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005234static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5235{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005236 unsigned long precharge = mem_cgroup_count_precharge(mm);
5237
5238 VM_BUG_ON(mc.moving_task);
5239 mc.moving_task = current;
5240 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005241}
5242
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005243/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5244static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005245{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005246 struct mem_cgroup *from = mc.from;
5247 struct mem_cgroup *to = mc.to;
5248
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005249 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005250 if (mc.precharge) {
5251 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5252 mc.precharge = 0;
5253 }
5254 /*
5255 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5256 * we must uncharge here.
5257 */
5258 if (mc.moved_charge) {
5259 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5260 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005261 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005262 /* we must fixup refcnts and charges */
5263 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005264 /* uncharge swap account from the old cgroup */
5265 if (!mem_cgroup_is_root(mc.from))
5266 res_counter_uncharge(&mc.from->memsw,
5267 PAGE_SIZE * mc.moved_swap);
5268 __mem_cgroup_put(mc.from, mc.moved_swap);
5269
5270 if (!mem_cgroup_is_root(mc.to)) {
5271 /*
5272 * we charged both to->res and to->memsw, so we should
5273 * uncharge to->res.
5274 */
5275 res_counter_uncharge(&mc.to->res,
5276 PAGE_SIZE * mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005277 }
5278 /* we've already done mem_cgroup_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005279 mc.moved_swap = 0;
5280 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005281 memcg_oom_recover(from);
5282 memcg_oom_recover(to);
5283 wake_up_all(&mc.waitq);
5284}
5285
5286static void mem_cgroup_clear_mc(void)
5287{
5288 struct mem_cgroup *from = mc.from;
5289
5290 /*
5291 * we must clear moving_task before waking up waiters at the end of
5292 * task migration.
5293 */
5294 mc.moving_task = NULL;
5295 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005296 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005297 mc.from = NULL;
5298 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005299 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005300 mem_cgroup_end_move(from);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005301}
5302
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005303static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5304 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005305 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005306{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005307 struct task_struct *p = cgroup_taskset_first(tset);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005308 int ret = 0;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005309 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005310
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005311 if (memcg->move_charge_at_immigrate) {
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005312 struct mm_struct *mm;
5313 struct mem_cgroup *from = mem_cgroup_from_task(p);
5314
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005315 VM_BUG_ON(from == memcg);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005316
5317 mm = get_task_mm(p);
5318 if (!mm)
5319 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005320 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005321 if (mm->owner == p) {
5322 VM_BUG_ON(mc.from);
5323 VM_BUG_ON(mc.to);
5324 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005325 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005326 VM_BUG_ON(mc.moved_swap);
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07005327 mem_cgroup_start_move(from);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005328 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005329 mc.from = from;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07005330 mc.to = memcg;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005331 spin_unlock(&mc.lock);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005332 /* We set mc.moving_task later */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005333
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005334 ret = mem_cgroup_precharge_mc(mm);
5335 if (ret)
5336 mem_cgroup_clear_mc();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005337 }
5338 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005339 }
5340 return ret;
5341}
5342
5343static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5344 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005345 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005346{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005347 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005348}
5349
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005350static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5351 unsigned long addr, unsigned long end,
5352 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005353{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005354 int ret = 0;
5355 struct vm_area_struct *vma = walk->private;
5356 pte_t *pte;
5357 spinlock_t *ptl;
5358
Dave Hansen03319322011-03-22 16:32:56 -07005359 split_huge_page_pmd(walk->mm, pmd);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005360retry:
5361 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5362 for (; addr != end; addr += PAGE_SIZE) {
5363 pte_t ptent = *(pte++);
5364 union mc_target target;
5365 int type;
5366 struct page *page;
5367 struct page_cgroup *pc;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005368 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005369
5370 if (!mc.precharge)
5371 break;
5372
5373 type = is_target_pte_for_mc(vma, addr, ptent, &target);
5374 switch (type) {
5375 case MC_TARGET_PAGE:
5376 page = target.page;
5377 if (isolate_lru_page(page))
5378 goto put;
5379 pc = lookup_page_cgroup(page);
Johannes Weiner7ec99d62011-03-23 16:42:36 -07005380 if (!mem_cgroup_move_account(page, 1, pc,
5381 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005382 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005383 /* we uncharge from mc.from later. */
5384 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005385 }
5386 putback_lru_page(page);
5387put: /* is_target_pte_for_mc() gets the page */
5388 put_page(page);
5389 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005390 case MC_TARGET_SWAP:
5391 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005392 if (!mem_cgroup_move_swap_account(ent,
5393 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005394 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005395 /* we fixup refcnts and charges later. */
5396 mc.moved_swap++;
5397 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005398 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005399 default:
5400 break;
5401 }
5402 }
5403 pte_unmap_unlock(pte - 1, ptl);
5404 cond_resched();
5405
5406 if (addr != end) {
5407 /*
5408 * We have consumed all precharges we got in can_attach().
5409 * We try charge one by one, but don't do any additional
5410 * charges to mc.to if we have failed in charge once in attach()
5411 * phase.
5412 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005413 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005414 if (!ret)
5415 goto retry;
5416 }
5417
5418 return ret;
5419}
5420
5421static void mem_cgroup_move_charge(struct mm_struct *mm)
5422{
5423 struct vm_area_struct *vma;
5424
5425 lru_add_drain_all();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005426retry:
5427 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5428 /*
5429 * Someone who are holding the mmap_sem might be waiting in
5430 * waitq. So we cancel all extra charges, wake up all waiters,
5431 * and retry. Because we cancel precharges, we might not be able
5432 * to move enough charges, but moving charge is a best-effort
5433 * feature anyway, so it wouldn't be a big problem.
5434 */
5435 __mem_cgroup_clear_mc();
5436 cond_resched();
5437 goto retry;
5438 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005439 for (vma = mm->mmap; vma; vma = vma->vm_next) {
5440 int ret;
5441 struct mm_walk mem_cgroup_move_charge_walk = {
5442 .pmd_entry = mem_cgroup_move_charge_pte_range,
5443 .mm = mm,
5444 .private = vma,
5445 };
5446 if (is_vm_hugetlb_page(vma))
5447 continue;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005448 ret = walk_page_range(vma->vm_start, vma->vm_end,
5449 &mem_cgroup_move_charge_walk);
5450 if (ret)
5451 /*
5452 * means we have consumed all precharges and failed in
5453 * doing additional charge. Just abandon here.
5454 */
5455 break;
5456 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005457 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005458}
5459
Balbir Singh67e465a2008-02-07 00:13:54 -08005460static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5461 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005462 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005463{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005464 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005465 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005466
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005467 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005468 if (mc.to)
5469 mem_cgroup_move_charge(mm);
5470 put_swap_token(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005471 mmput(mm);
5472 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005473 if (mc.to)
5474 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005475}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005476#else /* !CONFIG_MMU */
5477static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5478 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005479 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005480{
5481 return 0;
5482}
5483static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5484 struct cgroup *cgroup,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005485 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005486{
5487}
5488static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5489 struct cgroup *cont,
Tejun Heo2f7ee562011-12-12 18:12:21 -08005490 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005491{
5492}
5493#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005494
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005495struct cgroup_subsys mem_cgroup_subsys = {
5496 .name = "memory",
5497 .subsys_id = mem_cgroup_subsys_id,
5498 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005499 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005500 .destroy = mem_cgroup_destroy,
5501 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005502 .can_attach = mem_cgroup_can_attach,
5503 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005504 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005505 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07005506 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005507};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005508
5509#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Michal Hockoa42c3902010-11-24 12:57:08 -08005510static int __init enable_swap_account(char *s)
5511{
5512 /* consider enabled if no parameter or 1 is given */
Michal Hockoa2c89902011-05-24 17:12:50 -07005513 if (!strcmp(s, "1"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005514 really_do_swap_account = 1;
Michal Hockoa2c89902011-05-24 17:12:50 -07005515 else if (!strcmp(s, "0"))
Michal Hockoa42c3902010-11-24 12:57:08 -08005516 really_do_swap_account = 0;
5517 return 1;
5518}
Michal Hockoa2c89902011-05-24 17:12:50 -07005519__setup("swapaccount=", enable_swap_account);
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005520
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005521#endif