blob: 7973b5221fb8e86b816c8e9667aff5f6c5fa7e21 [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>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080036#include <linux/mutex.h>
Balbir Singhf64c3f52009-09-23 15:56:37 -070037#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070038#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080039#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080040#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080041#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080042#include <linux/eventfd.h>
43#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080045#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070046#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070047#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070048#include <linux/page_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080049#include <linux/cpu.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080050#include "internal.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080051
Balbir Singh8697d332008-02-07 00:13:59 -080052#include <asm/uaccess.h>
53
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070054struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070055#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh4b3bde42009-09-23 15:56:32 -070056struct mem_cgroup *root_mem_cgroup __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080057
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080058#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070059/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080060int do_swap_account __read_mostly;
61static int really_do_swap_account __initdata = 1; /* for remember boot option*/
62#else
63#define do_swap_account (0)
64#endif
65
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -080066/*
67 * Per memcg event counter is incremented at every pagein/pageout. This counter
68 * is used for trigger some periodic events. This is straightforward and better
69 * than using jiffies etc. to handle periodic memcg event.
70 *
71 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
72 */
73#define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
74#define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080075
Balbir Singh8cdea7c2008-02-07 00:13:50 -080076/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080077 * Statistics for memory cgroup.
78 */
79enum mem_cgroup_stat_index {
80 /*
81 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
82 */
83 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070084 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -080085 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
Balaji Rao55e462b2008-05-01 04:35:12 -070086 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
87 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
Balbir Singh0c3e73e2009-09-23 15:56:42 -070088 MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -080089 MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080090
91 MEM_CGROUP_STAT_NSTATS,
92};
93
94struct mem_cgroup_stat_cpu {
95 s64 count[MEM_CGROUP_STAT_NSTATS];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080096};
97
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080098/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -080099 * per-zone information in memory controller.
100 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800101struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800102 /*
103 * spin_lock to protect the per cgroup LRU
104 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700105 struct list_head lists[NR_LRU_LISTS];
106 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800107
108 struct zone_reclaim_stat reclaim_stat;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700109 struct rb_node tree_node; /* RB tree node */
110 unsigned long long usage_in_excess;/* Set to the value by which */
111 /* the soft limit is exceeded*/
112 bool on_tree;
Balbir Singh4e416952009-09-23 15:56:39 -0700113 struct mem_cgroup *mem; /* Back pointer, we cannot */
114 /* use container_of */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800115};
116/* Macro for accessing counter */
117#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
118
119struct mem_cgroup_per_node {
120 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
121};
122
123struct mem_cgroup_lru_info {
124 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
125};
126
127/*
Balbir Singhf64c3f52009-09-23 15:56:37 -0700128 * Cgroups above their limits are maintained in a RB-Tree, independent of
129 * their hierarchy representation
130 */
131
132struct mem_cgroup_tree_per_zone {
133 struct rb_root rb_root;
134 spinlock_t lock;
135};
136
137struct mem_cgroup_tree_per_node {
138 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
139};
140
141struct mem_cgroup_tree {
142 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
143};
144
145static struct mem_cgroup_tree soft_limit_tree __read_mostly;
146
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800147struct mem_cgroup_threshold {
148 struct eventfd_ctx *eventfd;
149 u64 threshold;
150};
151
152struct mem_cgroup_threshold_ary {
153 /* An array index points to threshold just below usage. */
154 atomic_t current_threshold;
155 /* Size of entries[] */
156 unsigned int size;
157 /* Array of thresholds */
158 struct mem_cgroup_threshold entries[0];
159};
160
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800161static void mem_cgroup_threshold(struct mem_cgroup *mem);
162
Balbir Singhf64c3f52009-09-23 15:56:37 -0700163/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800164 * The memory controller data structure. The memory controller controls both
165 * page cache and RSS per cgroup. We would eventually like to provide
166 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
167 * to help the administrator determine what knobs to tune.
168 *
169 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800170 * we hit the water mark. May be even add a low water mark, such that
171 * no reclaim occurs from a cgroup at it's low water mark, this is
172 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800173 */
174struct mem_cgroup {
175 struct cgroup_subsys_state css;
176 /*
177 * the counter to account for memory usage
178 */
179 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800180 /*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800181 * the counter to account for mem+swap usage.
182 */
183 struct res_counter memsw;
184 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800185 * Per cgroup active and inactive list, similar to the
186 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800187 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800188 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800189
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800190 /*
191 protect against reclaim related member.
192 */
193 spinlock_t reclaim_param_lock;
194
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800195 int prev_priority; /* for recording reclaim priority */
Balbir Singh6d61ef42009-01-07 18:08:06 -0800196
197 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200198 * While reclaiming in a hierarchy, we cache the last child we
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700199 * reclaimed from.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800200 */
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700201 int last_scanned_child;
Balbir Singh18f59ea2009-01-07 18:08:07 -0800202 /*
203 * Should the accounting and control be hierarchical, per subtree?
204 */
205 bool use_hierarchy;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -0800206 atomic_t oom_lock;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800207 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800208
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800209 unsigned int swappiness;
210
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700211 /* set when res.limit == memsw.limit */
212 bool memsw_is_minimum;
213
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800214 /* protect arrays of thresholds */
215 struct mutex thresholds_lock;
216
217 /* thresholds for memory usage. RCU-protected */
218 struct mem_cgroup_threshold_ary *thresholds;
219
220 /* thresholds for mem+swap usage. RCU-protected */
221 struct mem_cgroup_threshold_ary *memsw_thresholds;
222
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800223 /*
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800224 * Should we move charges of a task when a task is moved into this
225 * mem_cgroup ? And what type of charges should we move ?
226 */
227 unsigned long move_charge_at_immigrate;
228
229 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800230 * percpu counter.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800231 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800232 struct mem_cgroup_stat_cpu *stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800233};
234
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800235/* Stuffs for move charges at task migration. */
236/*
237 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
238 * left-shifted bitmap of these types.
239 */
240enum move_type {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800241 MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800242 NR_MOVE_TYPE,
243};
244
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800245/* "mc" and its members are protected by cgroup_mutex */
246static struct move_charge_struct {
247 struct mem_cgroup *from;
248 struct mem_cgroup *to;
249 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800250 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800251 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800252 struct task_struct *moving_task; /* a task moving charges */
253 wait_queue_head_t waitq; /* a waitq for other context */
254} mc = {
255 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
256};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800257
Balbir Singh4e416952009-09-23 15:56:39 -0700258/*
259 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
260 * limit reclaim to prevent infinite loops, if they ever occur.
261 */
262#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
263#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
264
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800265enum charge_type {
266 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
267 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700268 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700269 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800270 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700271 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700272 NR_CHARGE_TYPE,
273};
274
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700275/* only for here (for easy reading.) */
276#define PCGF_CACHE (1UL << PCG_CACHE)
277#define PCGF_USED (1UL << PCG_USED)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700278#define PCGF_LOCK (1UL << PCG_LOCK)
Balbir Singh4b3bde42009-09-23 15:56:32 -0700279/* Not used, but added here for completeness */
280#define PCGF_ACCT (1UL << PCG_ACCT)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800281
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800282/* for encoding cft->private value on file */
283#define _MEM (0)
284#define _MEMSWAP (1)
285#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
286#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
287#define MEMFILE_ATTR(val) ((val) & 0xffff)
288
Balbir Singh75822b42009-09-23 15:56:38 -0700289/*
290 * Reclaim flags for mem_cgroup_hierarchical_reclaim
291 */
292#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
293#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
294#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
295#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
Balbir Singh4e416952009-09-23 15:56:39 -0700296#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
297#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
Balbir Singh75822b42009-09-23 15:56:38 -0700298
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800299static void mem_cgroup_get(struct mem_cgroup *mem);
300static void mem_cgroup_put(struct mem_cgroup *mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -0800301static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -0800302static void drain_all_stock_async(void);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800303
Balbir Singhf64c3f52009-09-23 15:56:37 -0700304static struct mem_cgroup_per_zone *
305mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
306{
307 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
308}
309
Wu Fengguangd3242362009-12-16 12:19:59 +0100310struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
311{
312 return &mem->css;
313}
314
Balbir Singhf64c3f52009-09-23 15:56:37 -0700315static struct mem_cgroup_per_zone *
316page_cgroup_zoneinfo(struct page_cgroup *pc)
317{
318 struct mem_cgroup *mem = pc->mem_cgroup;
319 int nid = page_cgroup_nid(pc);
320 int zid = page_cgroup_zid(pc);
321
322 if (!mem)
323 return NULL;
324
325 return mem_cgroup_zoneinfo(mem, nid, zid);
326}
327
328static struct mem_cgroup_tree_per_zone *
329soft_limit_tree_node_zone(int nid, int zid)
330{
331 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
332}
333
334static struct mem_cgroup_tree_per_zone *
335soft_limit_tree_from_page(struct page *page)
336{
337 int nid = page_to_nid(page);
338 int zid = page_zonenum(page);
339
340 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
341}
342
343static void
Balbir Singh4e416952009-09-23 15:56:39 -0700344__mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
Balbir Singhf64c3f52009-09-23 15:56:37 -0700345 struct mem_cgroup_per_zone *mz,
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700346 struct mem_cgroup_tree_per_zone *mctz,
347 unsigned long long new_usage_in_excess)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700348{
349 struct rb_node **p = &mctz->rb_root.rb_node;
350 struct rb_node *parent = NULL;
351 struct mem_cgroup_per_zone *mz_node;
352
353 if (mz->on_tree)
354 return;
355
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700356 mz->usage_in_excess = new_usage_in_excess;
357 if (!mz->usage_in_excess)
358 return;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700359 while (*p) {
360 parent = *p;
361 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
362 tree_node);
363 if (mz->usage_in_excess < mz_node->usage_in_excess)
364 p = &(*p)->rb_left;
365 /*
366 * We can't avoid mem cgroups that are over their soft
367 * limit by the same amount
368 */
369 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
370 p = &(*p)->rb_right;
371 }
372 rb_link_node(&mz->tree_node, parent, p);
373 rb_insert_color(&mz->tree_node, &mctz->rb_root);
374 mz->on_tree = true;
Balbir Singh4e416952009-09-23 15:56:39 -0700375}
376
377static void
378__mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
379 struct mem_cgroup_per_zone *mz,
380 struct mem_cgroup_tree_per_zone *mctz)
381{
382 if (!mz->on_tree)
383 return;
384 rb_erase(&mz->tree_node, &mctz->rb_root);
385 mz->on_tree = false;
386}
387
388static void
Balbir Singhf64c3f52009-09-23 15:56:37 -0700389mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
390 struct mem_cgroup_per_zone *mz,
391 struct mem_cgroup_tree_per_zone *mctz)
392{
393 spin_lock(&mctz->lock);
Balbir Singh4e416952009-09-23 15:56:39 -0700394 __mem_cgroup_remove_exceeded(mem, mz, mctz);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700395 spin_unlock(&mctz->lock);
396}
397
Balbir Singhf64c3f52009-09-23 15:56:37 -0700398
399static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
400{
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700401 unsigned long long excess;
Balbir Singhf64c3f52009-09-23 15:56:37 -0700402 struct mem_cgroup_per_zone *mz;
403 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700404 int nid = page_to_nid(page);
405 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700406 mctz = soft_limit_tree_from_page(page);
407
408 /*
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700409 * Necessary to update all ancestors when hierarchy is used.
410 * because their event counter is not touched.
Balbir Singhf64c3f52009-09-23 15:56:37 -0700411 */
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700412 for (; mem; mem = parent_mem_cgroup(mem)) {
413 mz = mem_cgroup_zoneinfo(mem, nid, zid);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700414 excess = res_counter_soft_limit_excess(&mem->res);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700415 /*
416 * We have to update the tree if mz is on RB-tree or
417 * mem is over its softlimit.
418 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700419 if (excess || mz->on_tree) {
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700420 spin_lock(&mctz->lock);
421 /* if on-tree, remove it */
422 if (mz->on_tree)
423 __mem_cgroup_remove_exceeded(mem, mz, mctz);
424 /*
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700425 * Insert again. mz->usage_in_excess will be updated.
426 * If excess is 0, no tree ops.
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700427 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -0700428 __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -0700429 spin_unlock(&mctz->lock);
430 }
Balbir Singhf64c3f52009-09-23 15:56:37 -0700431 }
432}
433
434static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
435{
436 int node, zone;
437 struct mem_cgroup_per_zone *mz;
438 struct mem_cgroup_tree_per_zone *mctz;
439
440 for_each_node_state(node, N_POSSIBLE) {
441 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
442 mz = mem_cgroup_zoneinfo(mem, node, zone);
443 mctz = soft_limit_tree_node_zone(node, zone);
444 mem_cgroup_remove_exceeded(mem, mz, mctz);
445 }
446 }
447}
448
Balbir Singh4e416952009-09-23 15:56:39 -0700449static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
450{
451 return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
452}
453
454static struct mem_cgroup_per_zone *
455__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
456{
457 struct rb_node *rightmost = NULL;
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700458 struct mem_cgroup_per_zone *mz;
Balbir Singh4e416952009-09-23 15:56:39 -0700459
460retry:
KAMEZAWA Hiroyuki26251ea2009-10-01 15:44:08 -0700461 mz = NULL;
Balbir Singh4e416952009-09-23 15:56:39 -0700462 rightmost = rb_last(&mctz->rb_root);
463 if (!rightmost)
464 goto done; /* Nothing to reclaim from */
465
466 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
467 /*
468 * Remove the node now but someone else can add it back,
469 * we will to add it back at the end of reclaim to its correct
470 * position in the tree.
471 */
472 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
473 if (!res_counter_soft_limit_excess(&mz->mem->res) ||
474 !css_tryget(&mz->mem->css))
475 goto retry;
476done:
477 return mz;
478}
479
480static struct mem_cgroup_per_zone *
481mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
482{
483 struct mem_cgroup_per_zone *mz;
484
485 spin_lock(&mctz->lock);
486 mz = __mem_cgroup_largest_soft_limit_node(mctz);
487 spin_unlock(&mctz->lock);
488 return mz;
489}
490
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800491static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
492 enum mem_cgroup_stat_index idx)
493{
494 int cpu;
495 s64 val = 0;
496
497 for_each_possible_cpu(cpu)
498 val += per_cpu(mem->stat->count[idx], cpu);
499 return val;
500}
501
502static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
503{
504 s64 ret;
505
506 ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
507 ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
508 return ret;
509}
510
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700511static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
512 bool charge)
513{
514 int val = (charge) ? 1 : -1;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800515 this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700516}
517
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700518static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
519 struct page_cgroup *pc,
520 bool charge)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800521{
Balbir Singh0c3e73e2009-09-23 15:56:42 -0700522 int val = (charge) ? 1 : -1;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800523
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800524 preempt_disable();
525
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700526 if (PageCgroupCache(pc))
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800527 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800528 else
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800529 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
Balaji Rao55e462b2008-05-01 04:35:12 -0700530
531 if (charge)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800532 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
Balaji Rao55e462b2008-05-01 04:35:12 -0700533 else
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800534 __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800535 __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800536
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800537 preempt_enable();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800538}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800539
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700540static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
Christoph Lameterb69408e2008-10-18 20:26:14 -0700541 enum lru_list idx)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800542{
543 int nid, zid;
544 struct mem_cgroup_per_zone *mz;
545 u64 total = 0;
546
547 for_each_online_node(nid)
548 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
549 mz = mem_cgroup_zoneinfo(mem, nid, zid);
550 total += MEM_CGROUP_ZSTAT(mz, idx);
551 }
552 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800553}
554
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800555static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
556{
557 s64 val;
558
559 val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
560
561 return !(val & ((1 << event_mask_shift) - 1));
562}
563
564/*
565 * Check events in order.
566 *
567 */
568static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
569{
570 /* threshold event is triggered in finer grain than soft limit */
571 if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
572 mem_cgroup_threshold(mem);
573 if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
574 mem_cgroup_update_tree(mem, page);
575 }
576}
577
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800578static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800579{
580 return container_of(cgroup_subsys_state(cont,
581 mem_cgroup_subsys_id), struct mem_cgroup,
582 css);
583}
584
Balbir Singhcf475ad2008-04-29 01:00:16 -0700585struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800586{
Balbir Singh31a78f22008-09-28 23:09:31 +0100587 /*
588 * mm_update_next_owner() may clear mm->owner to NULL
589 * if it races with swapoff, page migration, etc.
590 * So this can be called with p == NULL.
591 */
592 if (unlikely(!p))
593 return NULL;
594
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800595 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
596 struct mem_cgroup, css);
597}
598
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800599static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
600{
601 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700602
603 if (!mm)
604 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800605 /*
606 * Because we have no locks, mm->owner's may be being moved to other
607 * cgroup. We use css_tryget() here even if this looks
608 * pessimistic (rather than adding locks here).
609 */
610 rcu_read_lock();
611 do {
612 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
613 if (unlikely(!mem))
614 break;
615 } while (!css_tryget(&mem->css));
616 rcu_read_unlock();
617 return mem;
618}
619
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700620/*
621 * Call callback function against all cgroup under hierarchy tree.
622 */
623static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
624 int (*func)(struct mem_cgroup *, void *))
625{
626 int found, ret, nextid;
627 struct cgroup_subsys_state *css;
628 struct mem_cgroup *mem;
629
630 if (!root->use_hierarchy)
631 return (*func)(root, data);
632
633 nextid = 1;
634 do {
635 ret = 0;
636 mem = NULL;
637
638 rcu_read_lock();
639 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
640 &found);
641 if (css && css_tryget(css))
642 mem = container_of(css, struct mem_cgroup, css);
643 rcu_read_unlock();
644
645 if (mem) {
646 ret = (*func)(mem, data);
647 css_put(&mem->css);
648 }
649 nextid = found + 1;
650 } while (!ret && css);
651
652 return ret;
653}
654
Balbir Singh4b3bde42009-09-23 15:56:32 -0700655static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
656{
657 return (mem == root_mem_cgroup);
658}
659
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800660/*
661 * Following LRU functions are allowed to be used without PCG_LOCK.
662 * Operations are called by routine of global LRU independently from memcg.
663 * What we have to take care of here is validness of pc->mem_cgroup.
664 *
665 * Changes to pc->mem_cgroup happens when
666 * 1. charge
667 * 2. moving account
668 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
669 * It is added to LRU before charge.
670 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
671 * When moving account, the page is not on LRU. It's isolated.
672 */
673
674void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800675{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800676 struct page_cgroup *pc;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800677 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700678
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800679 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800680 return;
681 pc = lookup_page_cgroup(page);
682 /* can happen while we handle swapcache. */
Balbir Singh4b3bde42009-09-23 15:56:32 -0700683 if (!TestClearPageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800684 return;
Balbir Singh4b3bde42009-09-23 15:56:32 -0700685 VM_BUG_ON(!pc->mem_cgroup);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800686 /*
687 * We don't check PCG_USED bit. It's cleared when the "page" is finally
688 * removed from global LRU.
689 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800690 mz = page_cgroup_zoneinfo(pc);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700691 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
Balbir Singh4b3bde42009-09-23 15:56:32 -0700692 if (mem_cgroup_is_root(pc->mem_cgroup))
693 return;
694 VM_BUG_ON(list_empty(&pc->lru));
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800695 list_del_init(&pc->lru);
696 return;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800697}
698
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800699void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800700{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800701 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800702}
703
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800704void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -0800705{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800706 struct mem_cgroup_per_zone *mz;
707 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800708
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800709 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700710 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700711
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800712 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800713 /*
714 * Used bit is set without atomic ops but after smp_wmb().
715 * For making pc->mem_cgroup visible, insert smp_rmb() here.
716 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800717 smp_rmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -0700718 /* unused or root page is not rotated. */
719 if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800720 return;
721 mz = page_cgroup_zoneinfo(pc);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700722 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -0800723}
724
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800725void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
726{
727 struct page_cgroup *pc;
728 struct mem_cgroup_per_zone *mz;
729
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800730 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800731 return;
732 pc = lookup_page_cgroup(page);
Balbir Singh4b3bde42009-09-23 15:56:32 -0700733 VM_BUG_ON(PageCgroupAcctLRU(pc));
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800734 /*
735 * Used bit is set without atomic ops but after smp_wmb().
736 * For making pc->mem_cgroup visible, insert smp_rmb() here.
737 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800738 smp_rmb();
739 if (!PageCgroupUsed(pc))
740 return;
741
742 mz = page_cgroup_zoneinfo(pc);
743 MEM_CGROUP_ZSTAT(mz, lru) += 1;
Balbir Singh4b3bde42009-09-23 15:56:32 -0700744 SetPageCgroupAcctLRU(pc);
745 if (mem_cgroup_is_root(pc->mem_cgroup))
746 return;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800747 list_add(&pc->lru, &mz->lists[lru]);
748}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800749
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800750/*
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800751 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
752 * lru because the page may.be reused after it's fully uncharged (because of
753 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
754 * it again. This function is only used to charge SwapCache. It's done under
755 * lock_page and expected that zone->lru_lock is never held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800756 */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800757static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800758{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800759 unsigned long flags;
760 struct zone *zone = page_zone(page);
761 struct page_cgroup *pc = lookup_page_cgroup(page);
762
763 spin_lock_irqsave(&zone->lru_lock, flags);
764 /*
765 * Forget old LRU when this page_cgroup is *not* used. This Used bit
766 * is guarded by lock_page() because the page is SwapCache.
767 */
768 if (!PageCgroupUsed(pc))
769 mem_cgroup_del_lru_list(page, page_lru(page));
770 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800771}
772
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800773static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
774{
775 unsigned long flags;
776 struct zone *zone = page_zone(page);
777 struct page_cgroup *pc = lookup_page_cgroup(page);
778
779 spin_lock_irqsave(&zone->lru_lock, flags);
780 /* link when the page is linked to LRU but page_cgroup isn't */
Balbir Singh4b3bde42009-09-23 15:56:32 -0700781 if (PageLRU(page) && !PageCgroupAcctLRU(pc))
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800782 mem_cgroup_add_lru_list(page, page_lru(page));
783 spin_unlock_irqrestore(&zone->lru_lock, flags);
784}
785
786
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800787void mem_cgroup_move_lists(struct page *page,
788 enum lru_list from, enum lru_list to)
789{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800790 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800791 return;
792 mem_cgroup_del_lru_list(page, from);
793 mem_cgroup_add_lru_list(page, to);
794}
795
David Rientjes4c4a2212008-02-07 00:14:06 -0800796int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
797{
798 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700799 struct mem_cgroup *curr = NULL;
David Rientjes4c4a2212008-02-07 00:14:06 -0800800
801 task_lock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700802 rcu_read_lock();
803 curr = try_get_mem_cgroup_from_mm(task->mm);
804 rcu_read_unlock();
David Rientjes4c4a2212008-02-07 00:14:06 -0800805 task_unlock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700806 if (!curr)
807 return 0;
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800808 /*
809 * We should check use_hierarchy of "mem" not "curr". Because checking
810 * use_hierarchy of "curr" here make this function true if hierarchy is
811 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
812 * hierarchy(even if use_hierarchy is disabled in "mem").
813 */
814 if (mem->use_hierarchy)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700815 ret = css_is_ancestor(&curr->css, &mem->css);
816 else
817 ret = (curr == mem);
818 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -0800819 return ret;
820}
821
Balbir Singh66e17072008-02-07 00:13:56 -0800822/*
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800823 * prev_priority control...this will be used in memory reclaim path.
824 */
825int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
826{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800827 int prev_priority;
828
829 spin_lock(&mem->reclaim_param_lock);
830 prev_priority = mem->prev_priority;
831 spin_unlock(&mem->reclaim_param_lock);
832
833 return prev_priority;
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800834}
835
836void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
837{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800838 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800839 if (priority < mem->prev_priority)
840 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800841 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800842}
843
844void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
845{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800846 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800847 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800848 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800849}
850
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800851static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800852{
853 unsigned long active;
854 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800855 unsigned long gb;
856 unsigned long inactive_ratio;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800857
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700858 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
859 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800860
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800861 gb = (inactive + active) >> (30 - PAGE_SHIFT);
862 if (gb)
863 inactive_ratio = int_sqrt(10 * gb);
864 else
865 inactive_ratio = 1;
866
867 if (present_pages) {
868 present_pages[0] = inactive;
869 present_pages[1] = active;
870 }
871
872 return inactive_ratio;
873}
874
875int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
876{
877 unsigned long active;
878 unsigned long inactive;
879 unsigned long present_pages[2];
880 unsigned long inactive_ratio;
881
882 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
883
884 inactive = present_pages[0];
885 active = present_pages[1];
886
887 if (inactive * inactive_ratio < active)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800888 return 1;
889
890 return 0;
891}
892
Rik van Riel56e49d22009-06-16 15:32:28 -0700893int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
894{
895 unsigned long active;
896 unsigned long inactive;
897
898 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
899 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
900
901 return (active > inactive);
902}
903
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800904unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
905 struct zone *zone,
906 enum lru_list lru)
907{
908 int nid = zone->zone_pgdat->node_id;
909 int zid = zone_idx(zone);
910 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
911
912 return MEM_CGROUP_ZSTAT(mz, lru);
913}
914
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800915struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
916 struct zone *zone)
917{
918 int nid = zone->zone_pgdat->node_id;
919 int zid = zone_idx(zone);
920 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
921
922 return &mz->reclaim_stat;
923}
924
925struct zone_reclaim_stat *
926mem_cgroup_get_reclaim_stat_from_page(struct page *page)
927{
928 struct page_cgroup *pc;
929 struct mem_cgroup_per_zone *mz;
930
931 if (mem_cgroup_disabled())
932 return NULL;
933
934 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800935 /*
936 * Used bit is set without atomic ops but after smp_wmb().
937 * For making pc->mem_cgroup visible, insert smp_rmb() here.
938 */
939 smp_rmb();
940 if (!PageCgroupUsed(pc))
941 return NULL;
942
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800943 mz = page_cgroup_zoneinfo(pc);
944 if (!mz)
945 return NULL;
946
947 return &mz->reclaim_stat;
948}
949
Balbir Singh66e17072008-02-07 00:13:56 -0800950unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
951 struct list_head *dst,
952 unsigned long *scanned, int order,
953 int mode, struct zone *z,
954 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700955 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800956{
957 unsigned long nr_taken = 0;
958 struct page *page;
959 unsigned long scan;
960 LIST_HEAD(pc_list);
961 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800962 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800963 int nid = z->zone_pgdat->node_id;
964 int zid = zone_idx(z);
965 struct mem_cgroup_per_zone *mz;
Johannes Weinerb7c46d12009-09-21 17:02:56 -0700966 int lru = LRU_FILE * file + active;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700967 int ret;
Balbir Singh66e17072008-02-07 00:13:56 -0800968
Balbir Singhcf475ad2008-04-29 01:00:16 -0700969 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800970 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700971 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -0800972
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800973 scan = 0;
974 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800975 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800976 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800977
978 page = pc->page;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700979 if (unlikely(!PageCgroupUsed(pc)))
980 continue;
Hugh Dickins436c65412008-02-07 00:14:12 -0800981 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800982 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800983
Hugh Dickins436c65412008-02-07 00:14:12 -0800984 scan++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700985 ret = __isolate_lru_page(page, mode, file);
986 switch (ret) {
987 case 0:
Balbir Singh66e17072008-02-07 00:13:56 -0800988 list_move(&page->lru, dst);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700989 mem_cgroup_del_lru(page);
Balbir Singh66e17072008-02-07 00:13:56 -0800990 nr_taken++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700991 break;
992 case -EBUSY:
993 /* we don't affect global LRU but rotate in our LRU */
994 mem_cgroup_rotate_lru_list(page, page_lru(page));
995 break;
996 default:
997 break;
Balbir Singh66e17072008-02-07 00:13:56 -0800998 }
999 }
1000
Balbir Singh66e17072008-02-07 00:13:56 -08001001 *scanned = scan;
1002 return nr_taken;
1003}
1004
Balbir Singh6d61ef42009-01-07 18:08:06 -08001005#define mem_cgroup_from_res_counter(counter, member) \
1006 container_of(counter, struct mem_cgroup, member)
1007
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -08001008static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
1009{
1010 if (do_swap_account) {
1011 if (res_counter_check_under_limit(&mem->res) &&
1012 res_counter_check_under_limit(&mem->memsw))
1013 return true;
1014 } else
1015 if (res_counter_check_under_limit(&mem->res))
1016 return true;
1017 return false;
1018}
1019
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001020static unsigned int get_swappiness(struct mem_cgroup *memcg)
1021{
1022 struct cgroup *cgrp = memcg->css.cgroup;
1023 unsigned int swappiness;
1024
1025 /* root ? */
1026 if (cgrp->parent == NULL)
1027 return vm_swappiness;
1028
1029 spin_lock(&memcg->reclaim_param_lock);
1030 swappiness = memcg->swappiness;
1031 spin_unlock(&memcg->reclaim_param_lock);
1032
1033 return swappiness;
1034}
1035
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001036static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
1037{
1038 int *val = data;
1039 (*val)++;
1040 return 0;
1041}
Balbir Singhe2224322009-04-02 16:57:39 -07001042
1043/**
Kirill A. Shutemov6a6135b2010-03-10 15:22:25 -08001044 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
Balbir Singhe2224322009-04-02 16:57:39 -07001045 * @memcg: The memory cgroup that went over limit
1046 * @p: Task that is going to be killed
1047 *
1048 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1049 * enabled
1050 */
1051void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1052{
1053 struct cgroup *task_cgrp;
1054 struct cgroup *mem_cgrp;
1055 /*
1056 * Need a buffer in BSS, can't rely on allocations. The code relies
1057 * on the assumption that OOM is serialized for memory controller.
1058 * If this assumption is broken, revisit this code.
1059 */
1060 static char memcg_name[PATH_MAX];
1061 int ret;
1062
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -08001063 if (!memcg || !p)
Balbir Singhe2224322009-04-02 16:57:39 -07001064 return;
1065
1066
1067 rcu_read_lock();
1068
1069 mem_cgrp = memcg->css.cgroup;
1070 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1071
1072 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1073 if (ret < 0) {
1074 /*
1075 * Unfortunately, we are unable to convert to a useful name
1076 * But we'll still print out the usage information
1077 */
1078 rcu_read_unlock();
1079 goto done;
1080 }
1081 rcu_read_unlock();
1082
1083 printk(KERN_INFO "Task in %s killed", memcg_name);
1084
1085 rcu_read_lock();
1086 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1087 if (ret < 0) {
1088 rcu_read_unlock();
1089 goto done;
1090 }
1091 rcu_read_unlock();
1092
1093 /*
1094 * Continues from above, so we don't need an KERN_ level
1095 */
1096 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1097done:
1098
1099 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1100 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1101 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1102 res_counter_read_u64(&memcg->res, RES_FAILCNT));
1103 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1104 "failcnt %llu\n",
1105 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1106 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1107 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1108}
1109
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001110/*
1111 * This function returns the number of memcg under hierarchy tree. Returns
1112 * 1(self count) if no children.
1113 */
1114static int mem_cgroup_count_children(struct mem_cgroup *mem)
1115{
1116 int num = 0;
1117 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1118 return num;
1119}
1120
Balbir Singh6d61ef42009-01-07 18:08:06 -08001121/*
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001122 * Visit the first child (need not be the first child as per the ordering
1123 * of the cgroup list, since we track last_scanned_child) of @mem and use
1124 * that to reclaim free pages from.
1125 */
1126static struct mem_cgroup *
1127mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1128{
1129 struct mem_cgroup *ret = NULL;
1130 struct cgroup_subsys_state *css;
1131 int nextid, found;
1132
1133 if (!root_mem->use_hierarchy) {
1134 css_get(&root_mem->css);
1135 ret = root_mem;
1136 }
1137
1138 while (!ret) {
1139 rcu_read_lock();
1140 nextid = root_mem->last_scanned_child + 1;
1141 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1142 &found);
1143 if (css && css_tryget(css))
1144 ret = container_of(css, struct mem_cgroup, css);
1145
1146 rcu_read_unlock();
1147 /* Updates scanning parameter */
1148 spin_lock(&root_mem->reclaim_param_lock);
1149 if (!css) {
1150 /* this means start scan from ID:1 */
1151 root_mem->last_scanned_child = 0;
1152 } else
1153 root_mem->last_scanned_child = found;
1154 spin_unlock(&root_mem->reclaim_param_lock);
1155 }
1156
1157 return ret;
1158}
1159
1160/*
1161 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1162 * we reclaimed from, so that we don't end up penalizing one child extensively
1163 * based on its position in the children list.
Balbir Singh6d61ef42009-01-07 18:08:06 -08001164 *
1165 * root_mem is the original ancestor that we've been reclaim from.
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001166 *
1167 * We give up and return to the caller when we visit root_mem twice.
1168 * (other groups can be removed while we're walking....)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001169 *
1170 * If shrink==true, for avoiding to free too much, this returns immedieately.
Balbir Singh6d61ef42009-01-07 18:08:06 -08001171 */
1172static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
Balbir Singh4e416952009-09-23 15:56:39 -07001173 struct zone *zone,
Balbir Singh75822b42009-09-23 15:56:38 -07001174 gfp_t gfp_mask,
1175 unsigned long reclaim_options)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001176{
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001177 struct mem_cgroup *victim;
1178 int ret, total = 0;
1179 int loop = 0;
Balbir Singh75822b42009-09-23 15:56:38 -07001180 bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1181 bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
Balbir Singh4e416952009-09-23 15:56:39 -07001182 bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1183 unsigned long excess = mem_cgroup_get_excess(root_mem);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001184
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001185 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1186 if (root_mem->memsw_is_minimum)
1187 noswap = true;
1188
Balbir Singh4e416952009-09-23 15:56:39 -07001189 while (1) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001190 victim = mem_cgroup_select_victim(root_mem);
Balbir Singh4e416952009-09-23 15:56:39 -07001191 if (victim == root_mem) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001192 loop++;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001193 if (loop >= 1)
1194 drain_all_stock_async();
Balbir Singh4e416952009-09-23 15:56:39 -07001195 if (loop >= 2) {
1196 /*
1197 * If we have not been able to reclaim
1198 * anything, it might because there are
1199 * no reclaimable pages under this hierarchy
1200 */
1201 if (!check_soft || !total) {
1202 css_put(&victim->css);
1203 break;
1204 }
1205 /*
1206 * We want to do more targetted reclaim.
1207 * excess >> 2 is not to excessive so as to
1208 * reclaim too much, nor too less that we keep
1209 * coming back to reclaim from this cgroup
1210 */
1211 if (total >= (excess >> 2) ||
1212 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1213 css_put(&victim->css);
1214 break;
1215 }
1216 }
1217 }
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08001218 if (!mem_cgroup_local_usage(victim)) {
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001219 /* this cgroup's local usage == 0 */
1220 css_put(&victim->css);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001221 continue;
1222 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001223 /* we use swappiness of local cgroup */
Balbir Singh4e416952009-09-23 15:56:39 -07001224 if (check_soft)
1225 ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1226 noswap, get_swappiness(victim), zone,
1227 zone->zone_pgdat->node_id);
1228 else
1229 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1230 noswap, get_swappiness(victim));
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001231 css_put(&victim->css);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001232 /*
1233 * At shrinking usage, we can't check we should stop here or
1234 * reclaim more. It's depends on callers. last_scanned_child
1235 * will work enough for keeping fairness under tree.
1236 */
1237 if (shrink)
1238 return ret;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001239 total += ret;
Balbir Singh4e416952009-09-23 15:56:39 -07001240 if (check_soft) {
1241 if (res_counter_check_under_soft_limit(&root_mem->res))
1242 return total;
1243 } else if (mem_cgroup_check_under_limit(root_mem))
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001244 return 1 + total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001245 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001246 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001247}
1248
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001249static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data)
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001250{
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001251 int *val = (int *)data;
1252 int x;
1253 /*
1254 * Logically, we can stop scanning immediately when we find
1255 * a memcg is already locked. But condidering unlock ops and
1256 * creation/removal of memcg, scan-all is simple operation.
1257 */
1258 x = atomic_inc_return(&mem->oom_lock);
1259 *val = max(x, *val);
1260 return 0;
1261}
1262/*
1263 * Check OOM-Killer is already running under our hierarchy.
1264 * If someone is running, return false.
1265 */
1266static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1267{
1268 int lock_count = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001269
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001270 mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb);
1271
1272 if (lock_count == 1)
1273 return true;
1274 return false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001275}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001276
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001277static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001278{
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001279 /*
1280 * When a new child is created while the hierarchy is under oom,
1281 * mem_cgroup_oom_lock() may not be called. We have to use
1282 * atomic_add_unless() here.
1283 */
1284 atomic_add_unless(&mem->oom_lock, -1, 0);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001285 return 0;
1286}
1287
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001288static void mem_cgroup_oom_unlock(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001289{
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001290 mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb);
1291}
1292
1293static DEFINE_MUTEX(memcg_oom_mutex);
1294static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1295
1296/*
1297 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1298 */
1299bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1300{
1301 DEFINE_WAIT(wait);
1302 bool locked;
1303
1304 /* At first, try to OOM lock hierarchy under mem.*/
1305 mutex_lock(&memcg_oom_mutex);
1306 locked = mem_cgroup_oom_lock(mem);
1307 /*
1308 * Even if signal_pending(), we can't quit charge() loop without
1309 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1310 * under OOM is always welcomed, use TASK_KILLABLE here.
1311 */
1312 if (!locked)
1313 prepare_to_wait(&memcg_oom_waitq, &wait, TASK_KILLABLE);
1314 mutex_unlock(&memcg_oom_mutex);
1315
1316 if (locked)
1317 mem_cgroup_out_of_memory(mem, mask);
1318 else {
1319 schedule();
1320 finish_wait(&memcg_oom_waitq, &wait);
1321 }
1322 mutex_lock(&memcg_oom_mutex);
1323 mem_cgroup_oom_unlock(mem);
1324 /*
1325 * Here, we use global waitq .....more fine grained waitq ?
1326 * Assume following hierarchy.
1327 * A/
1328 * 01
1329 * 02
1330 * assume OOM happens both in A and 01 at the same time. Tthey are
1331 * mutually exclusive by lock. (kill in 01 helps A.)
1332 * When we use per memcg waitq, we have to wake up waiters on A and 02
1333 * in addtion to waiters on 01. We use global waitq for avoiding mess.
1334 * It will not be a big problem.
1335 * (And a task may be moved to other groups while it's waiting for OOM.)
1336 */
1337 wake_up_all(&memcg_oom_waitq);
1338 mutex_unlock(&memcg_oom_mutex);
1339
1340 if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1341 return false;
1342 /* Give chance to dying process */
1343 schedule_timeout(1);
1344 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001345}
1346
Balbir Singhd69b0422009-06-17 16:26:34 -07001347/*
1348 * Currently used to update mapped file statistics, but the routine can be
1349 * generalized to update other statistics as well.
1350 */
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08001351void mem_cgroup_update_file_mapped(struct page *page, int val)
Balbir Singhd69b0422009-06-17 16:26:34 -07001352{
1353 struct mem_cgroup *mem;
Balbir Singhd69b0422009-06-17 16:26:34 -07001354 struct page_cgroup *pc;
1355
Balbir Singhd69b0422009-06-17 16:26:34 -07001356 pc = lookup_page_cgroup(page);
1357 if (unlikely(!pc))
1358 return;
1359
1360 lock_page_cgroup(pc);
1361 mem = pc->mem_cgroup;
1362 if (!mem)
1363 goto done;
1364
1365 if (!PageCgroupUsed(pc))
1366 goto done;
1367
1368 /*
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08001369 * Preemption is already disabled. We can use __this_cpu_xxx
Balbir Singhd69b0422009-06-17 16:26:34 -07001370 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08001371 __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], val);
Balbir Singhd69b0422009-06-17 16:26:34 -07001372
Balbir Singhd69b0422009-06-17 16:26:34 -07001373done:
1374 unlock_page_cgroup(pc);
1375}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001376
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001377/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001378 * size of first charge trial. "32" comes from vmscan.c's magic value.
1379 * TODO: maybe necessary to use big numbers in big irons.
1380 */
1381#define CHARGE_SIZE (32 * PAGE_SIZE)
1382struct memcg_stock_pcp {
1383 struct mem_cgroup *cached; /* this never be root cgroup */
1384 int charge;
1385 struct work_struct work;
1386};
1387static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1388static atomic_t memcg_drain_count;
1389
1390/*
1391 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1392 * from local stock and true is returned. If the stock is 0 or charges from a
1393 * cgroup which is not current target, returns false. This stock will be
1394 * refilled.
1395 */
1396static bool consume_stock(struct mem_cgroup *mem)
1397{
1398 struct memcg_stock_pcp *stock;
1399 bool ret = true;
1400
1401 stock = &get_cpu_var(memcg_stock);
1402 if (mem == stock->cached && stock->charge)
1403 stock->charge -= PAGE_SIZE;
1404 else /* need to call res_counter_charge */
1405 ret = false;
1406 put_cpu_var(memcg_stock);
1407 return ret;
1408}
1409
1410/*
1411 * Returns stocks cached in percpu to res_counter and reset cached information.
1412 */
1413static void drain_stock(struct memcg_stock_pcp *stock)
1414{
1415 struct mem_cgroup *old = stock->cached;
1416
1417 if (stock->charge) {
1418 res_counter_uncharge(&old->res, stock->charge);
1419 if (do_swap_account)
1420 res_counter_uncharge(&old->memsw, stock->charge);
1421 }
1422 stock->cached = NULL;
1423 stock->charge = 0;
1424}
1425
1426/*
1427 * This must be called under preempt disabled or must be called by
1428 * a thread which is pinned to local cpu.
1429 */
1430static void drain_local_stock(struct work_struct *dummy)
1431{
1432 struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1433 drain_stock(stock);
1434}
1435
1436/*
1437 * Cache charges(val) which is from res_counter, to local per_cpu area.
1438 * This will be consumed by consumt_stock() function, later.
1439 */
1440static void refill_stock(struct mem_cgroup *mem, int val)
1441{
1442 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1443
1444 if (stock->cached != mem) { /* reset if necessary */
1445 drain_stock(stock);
1446 stock->cached = mem;
1447 }
1448 stock->charge += val;
1449 put_cpu_var(memcg_stock);
1450}
1451
1452/*
1453 * Tries to drain stocked charges in other cpus. This function is asynchronous
1454 * and just put a work per cpu for draining localy on each cpu. Caller can
1455 * expects some charges will be back to res_counter later but cannot wait for
1456 * it.
1457 */
1458static void drain_all_stock_async(void)
1459{
1460 int cpu;
1461 /* This function is for scheduling "drain" in asynchronous way.
1462 * The result of "drain" is not directly handled by callers. Then,
1463 * if someone is calling drain, we don't have to call drain more.
1464 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1465 * there is a race. We just do loose check here.
1466 */
1467 if (atomic_read(&memcg_drain_count))
1468 return;
1469 /* Notify other cpus that system-wide "drain" is running */
1470 atomic_inc(&memcg_drain_count);
1471 get_online_cpus();
1472 for_each_online_cpu(cpu) {
1473 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1474 schedule_work_on(cpu, &stock->work);
1475 }
1476 put_online_cpus();
1477 atomic_dec(&memcg_drain_count);
1478 /* We don't wait for flush_work */
1479}
1480
1481/* This is a synchronous drain interface. */
1482static void drain_all_stock_sync(void)
1483{
1484 /* called when force_empty is called */
1485 atomic_inc(&memcg_drain_count);
1486 schedule_on_each_cpu(drain_local_stock);
1487 atomic_dec(&memcg_drain_count);
1488}
1489
1490static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1491 unsigned long action,
1492 void *hcpu)
1493{
1494 int cpu = (unsigned long)hcpu;
1495 struct memcg_stock_pcp *stock;
1496
1497 if (action != CPU_DEAD)
1498 return NOTIFY_OK;
1499 stock = &per_cpu(memcg_stock, cpu);
1500 drain_stock(stock);
1501 return NOTIFY_OK;
1502}
1503
1504/*
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001505 * Unlike exported interface, "oom" parameter is added. if oom==true,
1506 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001507 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001508static int __mem_cgroup_try_charge(struct mm_struct *mm,
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08001509 gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001510{
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07001511 struct mem_cgroup *mem, *mem_over_limit;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001512 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07001513 struct res_counter *fail_res;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001514 int csize = CHARGE_SIZE;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001515
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001516 /*
1517 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1518 * in system level. So, allow to go ahead dying process in addition to
1519 * MEMDIE process.
1520 */
1521 if (unlikely(test_thread_flag(TIF_MEMDIE)
1522 || fatal_signal_pending(current)))
1523 goto bypass;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001524
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001525 /*
Hugh Dickins3be91272008-02-07 00:14:19 -08001526 * We always charge the cgroup the mm_struct belongs to.
1527 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001528 * thread group leader migrates. It's possible that mm is not
1529 * set, if so charge the init_mm (happens for pagecache usage).
1530 */
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001531 mem = *memcg;
1532 if (likely(!mem)) {
1533 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001534 *memcg = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001535 } else {
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001536 css_get(&mem->css);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001537 }
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001538 if (unlikely(!mem))
1539 return 0;
1540
Nikanth Karthikesan46f7e602009-05-28 14:34:41 -07001541 VM_BUG_ON(css_is_removed(&mem->css));
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001542 if (mem_cgroup_is_root(mem))
1543 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001544
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001545 while (1) {
Balbir Singh0c3e73e2009-09-23 15:56:42 -07001546 int ret = 0;
Balbir Singh75822b42009-09-23 15:56:38 -07001547 unsigned long flags = 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001548
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001549 if (consume_stock(mem))
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08001550 goto done;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001551
1552 ret = res_counter_charge(&mem->res, csize, &fail_res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001553 if (likely(!ret)) {
1554 if (!do_swap_account)
1555 break;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001556 ret = res_counter_charge(&mem->memsw, csize, &fail_res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001557 if (likely(!ret))
1558 break;
1559 /* mem+swap counter fails */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001560 res_counter_uncharge(&mem->res, csize);
Balbir Singh75822b42009-09-23 15:56:38 -07001561 flags |= MEM_CGROUP_RECLAIM_NOSWAP;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001562 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1563 memsw);
1564 } else
1565 /* mem counter fails */
1566 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1567 res);
1568
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001569 /* reduce request size and retry */
1570 if (csize > PAGE_SIZE) {
1571 csize = PAGE_SIZE;
1572 continue;
1573 }
Hugh Dickins3be91272008-02-07 00:14:19 -08001574 if (!(gfp_mask & __GFP_WAIT))
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001575 goto nomem;
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001576
Balbir Singh4e416952009-09-23 15:56:39 -07001577 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1578 gfp_mask, flags);
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -08001579 if (ret)
1580 continue;
Balbir Singh66e17072008-02-07 00:13:56 -08001581
1582 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001583 * try_to_free_mem_cgroup_pages() might not give us a full
1584 * picture of reclaim. Some pages are reclaimed and might be
1585 * moved to swap cache or just unmapped from the cgroup.
1586 * Check the limit again to see if the reclaim reduced the
1587 * current usage of the cgroup before giving up
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001588 *
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001589 */
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -08001590 if (mem_cgroup_check_under_limit(mem_over_limit))
1591 continue;
Hugh Dickins3be91272008-02-07 00:14:19 -08001592
Daisuke Nishimura8033b972010-03-10 15:22:16 -08001593 /* try to avoid oom while someone is moving charge */
1594 if (mc.moving_task && current != mc.moving_task) {
1595 struct mem_cgroup *from, *to;
1596 bool do_continue = false;
1597 /*
1598 * There is a small race that "from" or "to" can be
1599 * freed by rmdir, so we use css_tryget().
1600 */
1601 rcu_read_lock();
1602 from = mc.from;
1603 to = mc.to;
1604 if (from && css_tryget(&from->css)) {
1605 if (mem_over_limit->use_hierarchy)
1606 do_continue = css_is_ancestor(
1607 &from->css,
1608 &mem_over_limit->css);
1609 else
1610 do_continue = (from == mem_over_limit);
1611 css_put(&from->css);
1612 }
1613 if (!do_continue && to && css_tryget(&to->css)) {
1614 if (mem_over_limit->use_hierarchy)
1615 do_continue = css_is_ancestor(
1616 &to->css,
1617 &mem_over_limit->css);
1618 else
1619 do_continue = (to == mem_over_limit);
1620 css_put(&to->css);
1621 }
1622 rcu_read_unlock();
1623 if (do_continue) {
1624 DEFINE_WAIT(wait);
1625 prepare_to_wait(&mc.waitq, &wait,
1626 TASK_INTERRUPTIBLE);
1627 /* moving charge context might have finished. */
1628 if (mc.moving_task)
1629 schedule();
1630 finish_wait(&mc.waitq, &wait);
1631 continue;
1632 }
1633 }
1634
Hugh Dickins3be91272008-02-07 00:14:19 -08001635 if (!nr_retries--) {
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001636 if (!oom)
1637 goto nomem;
1638 if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) {
1639 nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1640 continue;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001641 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001642 /* When we reach here, current task is dying .*/
1643 css_put(&mem->css);
1644 goto bypass;
Balbir Singh66e17072008-02-07 00:13:56 -08001645 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001646 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001647 if (csize > PAGE_SIZE)
1648 refill_stock(mem, csize - PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07001649done:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001650 return 0;
1651nomem:
1652 css_put(&mem->css);
1653 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001654bypass:
1655 *memcg = NULL;
1656 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001657}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001658
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001659/*
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001660 * Somemtimes we have to undo a charge we got by try_charge().
1661 * This function is for that and do uncharge, put css's refcnt.
1662 * gotten by try_charge().
1663 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001664static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
1665 unsigned long count)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001666{
1667 if (!mem_cgroup_is_root(mem)) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001668 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001669 if (do_swap_account)
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001670 res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
1671 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
1672 WARN_ON_ONCE(count > INT_MAX);
1673 __css_put(&mem->css, (int)count);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001674 }
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001675 /* we don't need css_put for root */
1676}
1677
1678static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
1679{
1680 __mem_cgroup_cancel_charge(mem, 1);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001681}
1682
1683/*
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001684 * A helper function to get mem_cgroup from ID. must be called under
1685 * rcu_read_lock(). The caller must check css_is_removed() or some if
1686 * it's concern. (dropping refcnt from swap can be called against removed
1687 * memcg.)
1688 */
1689static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1690{
1691 struct cgroup_subsys_state *css;
1692
1693 /* ID 0 is unused ID */
1694 if (!id)
1695 return NULL;
1696 css = css_lookup(&mem_cgroup_subsys, id);
1697 if (!css)
1698 return NULL;
1699 return container_of(css, struct mem_cgroup, css);
1700}
1701
Wu Fengguange42d9d52009-12-16 12:19:59 +01001702struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001703{
Wu Fengguange42d9d52009-12-16 12:19:59 +01001704 struct mem_cgroup *mem = NULL;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001705 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001706 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001707 swp_entry_t ent;
1708
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001709 VM_BUG_ON(!PageLocked(page));
1710
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001711 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07001712 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001713 if (PageCgroupUsed(pc)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001714 mem = pc->mem_cgroup;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001715 if (mem && !css_tryget(&mem->css))
1716 mem = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01001717 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001718 ent.val = page_private(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001719 id = lookup_swap_cgroup(ent);
1720 rcu_read_lock();
1721 mem = mem_cgroup_lookup(id);
1722 if (mem && !css_tryget(&mem->css))
1723 mem = NULL;
1724 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001725 }
Daisuke Nishimurac0bd3f62009-04-30 15:08:11 -07001726 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001727 return mem;
1728}
1729
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001730/*
Daisuke Nishimuraa5e924f2009-01-07 18:08:28 -08001731 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001732 * USED state. If already USED, uncharge and return.
1733 */
1734
1735static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1736 struct page_cgroup *pc,
1737 enum charge_type ctype)
1738{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001739 /* try_charge() can return NULL to *memcg, taking care of it. */
1740 if (!mem)
1741 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001742
1743 lock_page_cgroup(pc);
1744 if (unlikely(PageCgroupUsed(pc))) {
1745 unlock_page_cgroup(pc);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08001746 mem_cgroup_cancel_charge(mem);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001747 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001748 }
Balbir Singh4b3bde42009-09-23 15:56:32 -07001749
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001750 pc->mem_cgroup = mem;
KAMEZAWA Hiroyuki261fb612009-09-23 15:56:33 -07001751 /*
1752 * We access a page_cgroup asynchronously without lock_page_cgroup().
1753 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1754 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1755 * before USED bit, we need memory barrier here.
1756 * See mem_cgroup_add_lru_list(), etc.
1757 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001758 smp_wmb();
Balbir Singh4b3bde42009-09-23 15:56:32 -07001759 switch (ctype) {
1760 case MEM_CGROUP_CHARGE_TYPE_CACHE:
1761 case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1762 SetPageCgroupCache(pc);
1763 SetPageCgroupUsed(pc);
1764 break;
1765 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1766 ClearPageCgroupCache(pc);
1767 SetPageCgroupUsed(pc);
1768 break;
1769 default:
1770 break;
1771 }
Hugh Dickins3be91272008-02-07 00:14:19 -08001772
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001773 mem_cgroup_charge_statistics(mem, pc, true);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001774
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001775 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08001776 /*
1777 * "charge_statistics" updated event counter. Then, check it.
1778 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1779 * if they exceeds softlimit.
1780 */
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001781 memcg_check_events(mem, pc->page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001782}
1783
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001784/**
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001785 * __mem_cgroup_move_account - move account of the page
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001786 * @pc: page_cgroup of the page.
1787 * @from: mem_cgroup which the page is moved from.
1788 * @to: mem_cgroup which the page is moved to. @from != @to.
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001789 * @uncharge: whether we should call uncharge and css_put against @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001790 *
1791 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001792 * - page is not on LRU (isolate_page() is useful.)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001793 * - the pc is locked, used, and ->mem_cgroup points to @from.
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001794 *
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001795 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1796 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1797 * true, this function does "uncharge" from old cgroup, but it doesn't if
1798 * @uncharge is false, so a caller should do "uncharge".
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001799 */
1800
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001801static void __mem_cgroup_move_account(struct page_cgroup *pc,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001802 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001803{
Balbir Singhd69b0422009-06-17 16:26:34 -07001804 struct page *page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001805
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001806 VM_BUG_ON(from == to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001807 VM_BUG_ON(PageLRU(pc->page));
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001808 VM_BUG_ON(!PageCgroupLocked(pc));
1809 VM_BUG_ON(!PageCgroupUsed(pc));
1810 VM_BUG_ON(pc->mem_cgroup != from);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001811
Balbir Singhd69b0422009-06-17 16:26:34 -07001812 page = pc->page;
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08001813 if (page_mapped(page) && !PageAnon(page)) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08001814 /* Update mapped_file data for mem_cgroup */
1815 preempt_disable();
1816 __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1817 __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1818 preempt_enable();
Balbir Singhd69b0422009-06-17 16:26:34 -07001819 }
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001820 mem_cgroup_charge_statistics(from, pc, false);
1821 if (uncharge)
1822 /* This is not "cancel", but cancel_charge does all we need. */
1823 mem_cgroup_cancel_charge(from);
Balbir Singhd69b0422009-06-17 16:26:34 -07001824
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001825 /* caller should have done css_get */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001826 pc->mem_cgroup = to;
1827 mem_cgroup_charge_statistics(to, pc, true);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001828 /*
1829 * We charges against "to" which may not have any tasks. Then, "to"
1830 * can be under rmdir(). But in current implementation, caller of
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08001831 * this function is just force_empty() and move charge, so it's
1832 * garanteed that "to" is never removed. So, we don't check rmdir
1833 * status here.
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001834 */
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001835}
1836
1837/*
1838 * check whether the @pc is valid for moving account and call
1839 * __mem_cgroup_move_account()
1840 */
1841static int mem_cgroup_move_account(struct page_cgroup *pc,
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001842 struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001843{
1844 int ret = -EINVAL;
1845 lock_page_cgroup(pc);
1846 if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001847 __mem_cgroup_move_account(pc, from, to, uncharge);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001848 ret = 0;
1849 }
1850 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001851 /*
1852 * check events
1853 */
1854 memcg_check_events(to, pc->page);
1855 memcg_check_events(from, pc->page);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001856 return ret;
1857}
1858
1859/*
1860 * move charges to its parent.
1861 */
1862
1863static int mem_cgroup_move_parent(struct page_cgroup *pc,
1864 struct mem_cgroup *child,
1865 gfp_t gfp_mask)
1866{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001867 struct page *page = pc->page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001868 struct cgroup *cg = child->css.cgroup;
1869 struct cgroup *pcg = cg->parent;
1870 struct mem_cgroup *parent;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001871 int ret;
1872
1873 /* Is ROOT ? */
1874 if (!pcg)
1875 return -EINVAL;
1876
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001877 ret = -EBUSY;
1878 if (!get_page_unless_zero(page))
1879 goto out;
1880 if (isolate_lru_page(page))
1881 goto put;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001882
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001883 parent = mem_cgroup_from_cont(pcg);
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08001884 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001885 if (ret || !parent)
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001886 goto put_back;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001887
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08001888 ret = mem_cgroup_move_account(pc, child, parent, true);
1889 if (ret)
1890 mem_cgroup_cancel_charge(parent);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001891put_back:
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001892 putback_lru_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001893put:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001894 put_page(page);
Daisuke Nishimura57f9fd72009-12-15 16:47:11 -08001895out:
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001896 return ret;
1897}
1898
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001899/*
1900 * Charge the memory controller for page usage.
1901 * Return
1902 * 0 if the charge was successful
1903 * < 0 if the cgroup is over its limit
1904 */
1905static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1906 gfp_t gfp_mask, enum charge_type ctype,
1907 struct mem_cgroup *memcg)
1908{
1909 struct mem_cgroup *mem;
1910 struct page_cgroup *pc;
1911 int ret;
1912
1913 pc = lookup_page_cgroup(page);
1914 /* can happen at boot */
1915 if (unlikely(!pc))
1916 return 0;
1917 prefetchw(pc);
1918
1919 mem = memcg;
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08001920 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001921 if (ret || !mem)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001922 return ret;
1923
1924 __mem_cgroup_commit_charge(mem, pc, ctype);
1925 return 0;
1926}
1927
1928int mem_cgroup_newpage_charge(struct page *page,
1929 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001930{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001931 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001932 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001933 if (PageCompound(page))
1934 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001935 /*
1936 * If already mapped, we don't have to account.
1937 * If page cache, page->mapping has address_space.
1938 * But page->mapping may have out-of-use anon_vma pointer,
1939 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1940 * is NULL.
1941 */
1942 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1943 return 0;
1944 if (unlikely(!mm))
1945 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001946 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001947 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001948}
1949
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001950static void
1951__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1952 enum charge_type ctype);
1953
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001954int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1955 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08001956{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001957 struct mem_cgroup *mem = NULL;
1958 int ret;
1959
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001960 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001961 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001962 if (PageCompound(page))
1963 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001964 /*
1965 * Corner case handling. This is called from add_to_page_cache()
1966 * in usual. But some FS (shmem) precharges this page before calling it
1967 * and call add_to_page_cache() with GFP_NOWAIT.
1968 *
1969 * For GFP_NOWAIT case, the page may be pre-charged before calling
1970 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1971 * charge twice. (It works but has to pay a bit larger cost.)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001972 * And when the page is SwapCache, it should take swap information
1973 * into account. This is under lock_page() now.
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001974 */
1975 if (!(gfp_mask & __GFP_WAIT)) {
1976 struct page_cgroup *pc;
1977
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001978
1979 pc = lookup_page_cgroup(page);
1980 if (!pc)
1981 return 0;
1982 lock_page_cgroup(pc);
1983 if (PageCgroupUsed(pc)) {
1984 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001985 return 0;
1986 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001987 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001988 }
1989
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001990 if (unlikely(!mm && !mem))
Balbir Singh8697d332008-02-07 00:13:59 -08001991 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001992
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001993 if (page_is_file_cache(page))
1994 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001995 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001996
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001997 /* shmem */
1998 if (PageSwapCache(page)) {
1999 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2000 if (!ret)
2001 __mem_cgroup_commit_charge_swapin(page, mem,
2002 MEM_CGROUP_CHARGE_TYPE_SHMEM);
2003 } else
2004 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2005 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002006
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002007 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002008}
2009
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002010/*
2011 * While swap-in, try_charge -> commit or cancel, the page is locked.
2012 * And when try_charge() successfully returns, one refcnt to memcg without
Uwe Kleine-König21ae2952009-10-07 15:21:09 +02002013 * struct page_cgroup is acquired. This refcnt will be consumed by
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002014 * "commit()" or removed by "cancel()"
2015 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002016int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2017 struct page *page,
2018 gfp_t mask, struct mem_cgroup **ptr)
2019{
2020 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002021 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002022
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002023 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002024 return 0;
2025
2026 if (!do_swap_account)
2027 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002028 /*
2029 * A racing thread's fault, or swapoff, may have already updated
Hugh Dickins407f9c82009-12-14 17:59:30 -08002030 * the pte, and even removed page from swap cache: in those cases
2031 * do_swap_page()'s pte_same() test will fail; but there's also a
2032 * KSM case which does need to charge the page.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002033 */
2034 if (!PageSwapCache(page))
Hugh Dickins407f9c82009-12-14 17:59:30 -08002035 goto charge_cur_mm;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002036 mem = try_get_mem_cgroup_from_page(page);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002037 if (!mem)
2038 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002039 *ptr = mem;
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002040 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08002041 /* drop extra refcnt from tryget */
2042 css_put(&mem->css);
2043 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002044charge_cur_mm:
2045 if (unlikely(!mm))
2046 mm = &init_mm;
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002047 return __mem_cgroup_try_charge(mm, mask, ptr, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002048}
2049
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002050static void
2051__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2052 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002053{
2054 struct page_cgroup *pc;
2055
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002056 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002057 return;
2058 if (!ptr)
2059 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002060 cgroup_exclude_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002061 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002062 mem_cgroup_lru_del_before_commit_swapcache(page);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002063 __mem_cgroup_commit_charge(ptr, pc, ctype);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002064 mem_cgroup_lru_add_after_commit_swapcache(page);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002065 /*
2066 * Now swap is on-memory. This means this page may be
2067 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002068 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2069 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2070 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002071 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002072 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002073 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002074 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002075 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002076
2077 id = swap_cgroup_record(ent, 0);
2078 rcu_read_lock();
2079 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002080 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002081 /*
2082 * This recorded memcg can be obsolete one. So, avoid
2083 * calling css_tryget
2084 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002085 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07002086 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002087 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002088 mem_cgroup_put(memcg);
2089 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002090 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002091 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002092 /*
2093 * At swapin, we may charge account against cgroup which has no tasks.
2094 * So, rmdir()->pre_destroy() can be called while we do this charge.
2095 * In that case, we need to call pre_destroy() again. check it here.
2096 */
2097 cgroup_release_and_wakeup_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002098}
2099
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07002100void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2101{
2102 __mem_cgroup_commit_charge_swapin(page, ptr,
2103 MEM_CGROUP_CHARGE_TYPE_MAPPED);
2104}
2105
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002106void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2107{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002108 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002109 return;
2110 if (!mem)
2111 return;
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002112 mem_cgroup_cancel_charge(mem);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002113}
2114
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002115static void
2116__do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
2117{
2118 struct memcg_batch_info *batch = NULL;
2119 bool uncharge_memsw = true;
2120 /* If swapout, usage of swap doesn't decrease */
2121 if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2122 uncharge_memsw = false;
2123 /*
2124 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2125 * In those cases, all pages freed continously can be expected to be in
2126 * the same cgroup and we have chance to coalesce uncharges.
2127 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2128 * because we want to do uncharge as soon as possible.
2129 */
2130 if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
2131 goto direct_uncharge;
2132
2133 batch = &current->memcg_batch;
2134 /*
2135 * In usual, we do css_get() when we remember memcg pointer.
2136 * But in this case, we keep res->usage until end of a series of
2137 * uncharges. Then, it's ok to ignore memcg's refcnt.
2138 */
2139 if (!batch->memcg)
2140 batch->memcg = mem;
2141 /*
2142 * In typical case, batch->memcg == mem. This means we can
2143 * merge a series of uncharges to an uncharge of res_counter.
2144 * If not, we uncharge res_counter ony by one.
2145 */
2146 if (batch->memcg != mem)
2147 goto direct_uncharge;
2148 /* remember freed charge and uncharge it later */
2149 batch->bytes += PAGE_SIZE;
2150 if (uncharge_memsw)
2151 batch->memsw_bytes += PAGE_SIZE;
2152 return;
2153direct_uncharge:
2154 res_counter_uncharge(&mem->res, PAGE_SIZE);
2155 if (uncharge_memsw)
2156 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2157 return;
2158}
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002159
Balbir Singh8697d332008-02-07 00:13:59 -08002160/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002161 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002162 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002163static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002164__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002165{
Hugh Dickins82895462008-03-04 14:29:08 -08002166 struct page_cgroup *pc;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002167 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08002168 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002169
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002170 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002171 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07002172
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002173 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002174 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002175
Balbir Singh8697d332008-02-07 00:13:59 -08002176 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08002177 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08002178 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002179 pc = lookup_page_cgroup(page);
2180 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002181 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002182
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002183 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002184
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002185 mem = pc->mem_cgroup;
2186
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002187 if (!PageCgroupUsed(pc))
2188 goto unlock_out;
2189
2190 switch (ctype) {
2191 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002192 case MEM_CGROUP_CHARGE_TYPE_DROP:
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002193 if (page_mapped(page))
2194 goto unlock_out;
2195 break;
2196 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2197 if (!PageAnon(page)) { /* Shared memory */
2198 if (page->mapping && !page_is_file_cache(page))
2199 goto unlock_out;
2200 } else if (page_mapped(page)) /* Anon */
2201 goto unlock_out;
2202 break;
2203 default:
2204 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002205 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002206
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002207 if (!mem_cgroup_is_root(mem))
2208 __do_uncharge(mem, ctype);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002209 if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2210 mem_cgroup_swap_statistics(mem, true);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002211 mem_cgroup_charge_statistics(mem, pc, false);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002212
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002213 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08002214 /*
2215 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2216 * freed from LRU. This is safe because uncharged page is expected not
2217 * to be reused (freed soon). Exception is SwapCache, it's handled by
2218 * special functions.
2219 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08002220
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002221 mz = page_cgroup_zoneinfo(pc);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002222 unlock_page_cgroup(pc);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08002223
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08002224 memcg_check_events(mem, page);
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08002225 /* at swapout, this memcg will be accessed to record to swap */
2226 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2227 css_put(&mem->css);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002228
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002229 return mem;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002230
2231unlock_out:
2232 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002233 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08002234}
2235
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002236void mem_cgroup_uncharge_page(struct page *page)
2237{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002238 /* early check. */
2239 if (page_mapped(page))
2240 return;
2241 if (page->mapping && !PageAnon(page))
2242 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002243 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2244}
2245
2246void mem_cgroup_uncharge_cache_page(struct page *page)
2247{
2248 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07002249 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002250 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2251}
2252
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08002253/*
2254 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2255 * In that cases, pages are freed continuously and we can expect pages
2256 * are in the same memcg. All these calls itself limits the number of
2257 * pages freed at once, then uncharge_start/end() is called properly.
2258 * This may be called prural(2) times in a context,
2259 */
2260
2261void mem_cgroup_uncharge_start(void)
2262{
2263 current->memcg_batch.do_batch++;
2264 /* We can do nest. */
2265 if (current->memcg_batch.do_batch == 1) {
2266 current->memcg_batch.memcg = NULL;
2267 current->memcg_batch.bytes = 0;
2268 current->memcg_batch.memsw_bytes = 0;
2269 }
2270}
2271
2272void mem_cgroup_uncharge_end(void)
2273{
2274 struct memcg_batch_info *batch = &current->memcg_batch;
2275
2276 if (!batch->do_batch)
2277 return;
2278
2279 batch->do_batch--;
2280 if (batch->do_batch) /* If stacked, do nothing. */
2281 return;
2282
2283 if (!batch->memcg)
2284 return;
2285 /*
2286 * This "batch->memcg" is valid without any css_get/put etc...
2287 * bacause we hide charges behind us.
2288 */
2289 if (batch->bytes)
2290 res_counter_uncharge(&batch->memcg->res, batch->bytes);
2291 if (batch->memsw_bytes)
2292 res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2293 /* forget this pointer (for sanity check) */
2294 batch->memcg = NULL;
2295}
2296
Daisuke Nishimurae767e052009-05-28 14:34:28 -07002297#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002298/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07002299 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002300 * memcg information is recorded to swap_cgroup of "ent"
2301 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002302void
2303mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002304{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002305 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002306 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002307
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002308 if (!swapout) /* this was a swap cache but the swap is unused ! */
2309 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2310
2311 memcg = __mem_cgroup_uncharge_common(page, ctype);
2312
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002313 /* record memcg information */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002314 if (do_swap_account && swapout && memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002315 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002316 mem_cgroup_get(memcg);
2317 }
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07002318 if (swapout && memcg)
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08002319 css_put(&memcg->css);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002320}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07002321#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002322
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002323#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2324/*
2325 * called from swap_entry_free(). remove record in swap_cgroup and
2326 * uncharge "memsw" account.
2327 */
2328void mem_cgroup_uncharge_swap(swp_entry_t ent)
2329{
2330 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002331 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002332
2333 if (!do_swap_account)
2334 return;
2335
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002336 id = swap_cgroup_record(ent, 0);
2337 rcu_read_lock();
2338 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002339 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002340 /*
2341 * We uncharge this because swap is freed.
2342 * This memcg can be obsolete one. We avoid calling css_tryget
2343 */
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002344 if (!mem_cgroup_is_root(memcg))
KAMEZAWA Hiroyuki4e649152009-10-01 15:44:11 -07002345 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002346 mem_cgroup_swap_statistics(memcg, false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002347 mem_cgroup_put(memcg);
2348 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002349 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002350}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002351
2352/**
2353 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2354 * @entry: swap entry to be moved
2355 * @from: mem_cgroup which the entry is moved from
2356 * @to: mem_cgroup which the entry is moved to
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08002357 * @need_fixup: whether we should fixup res_counters and refcounts.
Daisuke Nishimura02491442010-03-10 15:22:17 -08002358 *
2359 * It succeeds only when the swap_cgroup's record for this entry is the same
2360 * as the mem_cgroup's id of @from.
2361 *
2362 * Returns 0 on success, -EINVAL on failure.
2363 *
2364 * The caller must have charged to @to, IOW, called res_counter_charge() about
2365 * both res and memsw, and called css_get().
2366 */
2367static int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08002368 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002369{
2370 unsigned short old_id, new_id;
2371
2372 old_id = css_id(&from->css);
2373 new_id = css_id(&to->css);
2374
2375 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002376 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002377 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08002378 /*
2379 * This function is only called from task migration context now.
2380 * It postpones res_counter and refcount handling till the end
2381 * of task migration(mem_cgroup_clear_mc()) for performance
2382 * improvement. But we cannot postpone mem_cgroup_get(to)
2383 * because if the process that has been moved to @to does
2384 * swap-in, the refcount of @to might be decreased to 0.
2385 */
Daisuke Nishimura02491442010-03-10 15:22:17 -08002386 mem_cgroup_get(to);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08002387 if (need_fixup) {
2388 if (!mem_cgroup_is_root(from))
2389 res_counter_uncharge(&from->memsw, PAGE_SIZE);
2390 mem_cgroup_put(from);
2391 /*
2392 * we charged both to->res and to->memsw, so we should
2393 * uncharge to->res.
2394 */
2395 if (!mem_cgroup_is_root(to))
2396 res_counter_uncharge(&to->res, PAGE_SIZE);
2397 css_put(&to->css);
2398 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08002399 return 0;
2400 }
2401 return -EINVAL;
2402}
2403#else
2404static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08002405 struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002406{
2407 return -EINVAL;
2408}
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002409#endif
2410
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08002411/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002412 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2413 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08002414 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002415int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08002416{
2417 struct page_cgroup *pc;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002418 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002419 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08002420
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002421 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07002422 return 0;
2423
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002424 pc = lookup_page_cgroup(page);
2425 lock_page_cgroup(pc);
2426 if (PageCgroupUsed(pc)) {
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002427 mem = pc->mem_cgroup;
2428 css_get(&mem->css);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08002429 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002430 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002431
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002432 if (mem) {
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08002433 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002434 css_put(&mem->css);
2435 }
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002436 *ptr = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002437 return ret;
2438}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08002439
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002440/* remove redundant charge if migration failed*/
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002441void mem_cgroup_end_migration(struct mem_cgroup *mem,
2442 struct page *oldpage, struct page *newpage)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07002443{
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002444 struct page *target, *unused;
2445 struct page_cgroup *pc;
2446 enum charge_type ctype;
2447
2448 if (!mem)
2449 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002450 cgroup_exclude_rmdir(&mem->css);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002451 /* at migration success, oldpage->mapping is NULL. */
2452 if (oldpage->mapping) {
2453 target = oldpage;
2454 unused = NULL;
2455 } else {
2456 target = newpage;
2457 unused = oldpage;
2458 }
2459
2460 if (PageAnon(target))
2461 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2462 else if (page_is_file_cache(target))
2463 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2464 else
2465 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2466
2467 /* unused page is not on radix-tree now. */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08002468 if (unused)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002469 __mem_cgroup_uncharge_common(unused, ctype);
2470
2471 pc = lookup_page_cgroup(target);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002472 /*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002473 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
2474 * So, double-counting is effectively avoided.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07002475 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08002476 __mem_cgroup_commit_charge(mem, pc, ctype);
2477
2478 /*
2479 * Both of oldpage and newpage are still under lock_page().
2480 * Then, we don't have to care about race in radix-tree.
2481 * But we have to be careful that this page is unmapped or not.
2482 *
2483 * There is a case for !page_mapped(). At the start of
2484 * migration, oldpage was mapped. But now, it's zapped.
2485 * But we know *target* page is not freed/reused under us.
2486 * mem_cgroup_uncharge_page() does all necessary checks.
2487 */
2488 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
2489 mem_cgroup_uncharge_page(target);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07002490 /*
2491 * At migration, we may charge account against cgroup which has no tasks
2492 * So, rmdir()->pre_destroy() can be called while we do this charge.
2493 * In that case, we need to call pre_destroy() again. check it here.
2494 */
2495 cgroup_release_and_wakeup_rmdir(&mem->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08002496}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08002497
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002498/*
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07002499 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2500 * Calling hierarchical_reclaim is not enough because we should update
2501 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2502 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2503 * not from the memcg which this page would be charged to.
2504 * try_charge_swapin does all of these works properly.
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002505 */
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07002506int mem_cgroup_shmem_charge_fallback(struct page *page,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002507 struct mm_struct *mm,
2508 gfp_t gfp_mask)
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002509{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002510 struct mem_cgroup *mem = NULL;
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07002511 int ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002512
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002513 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07002514 return 0;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002515
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07002516 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2517 if (!ret)
2518 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002519
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07002520 return ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002521}
2522
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002523static DEFINE_MUTEX(set_limit_mutex);
2524
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002525static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002526 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002527{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002528 int retry_count;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002529 u64 memswlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002530 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002531 int children = mem_cgroup_count_children(memcg);
2532 u64 curusage, oldusage;
2533
2534 /*
2535 * For keeping hierarchical_reclaim simple, how long we should retry
2536 * is depends on callers. We set our retry-count to be function
2537 * of # of children which we should visit in this loop.
2538 */
2539 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2540
2541 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002542
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002543 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002544 if (signal_pending(current)) {
2545 ret = -EINTR;
2546 break;
2547 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002548 /*
2549 * Rather than hide all in some function, I do this in
2550 * open coded manner. You see what this really does.
2551 * We have to guarantee mem->res.limit < mem->memsw.limit.
2552 */
2553 mutex_lock(&set_limit_mutex);
2554 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2555 if (memswlimit < val) {
2556 ret = -EINVAL;
2557 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002558 break;
2559 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002560 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07002561 if (!ret) {
2562 if (memswlimit == val)
2563 memcg->memsw_is_minimum = true;
2564 else
2565 memcg->memsw_is_minimum = false;
2566 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002567 mutex_unlock(&set_limit_mutex);
2568
2569 if (!ret)
2570 break;
2571
Bob Liuaa20d482009-12-15 16:47:14 -08002572 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
Balbir Singh4e416952009-09-23 15:56:39 -07002573 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002574 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2575 /* Usage is reduced ? */
2576 if (curusage >= oldusage)
2577 retry_count--;
2578 else
2579 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002580 }
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002581
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002582 return ret;
2583}
2584
Li Zefan338c8432009-06-17 16:27:15 -07002585static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2586 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002587{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002588 int retry_count;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002589 u64 memlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002590 int children = mem_cgroup_count_children(memcg);
2591 int ret = -EBUSY;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002592
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002593 /* see mem_cgroup_resize_res_limit */
2594 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2595 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002596 while (retry_count) {
2597 if (signal_pending(current)) {
2598 ret = -EINTR;
2599 break;
2600 }
2601 /*
2602 * Rather than hide all in some function, I do this in
2603 * open coded manner. You see what this really does.
2604 * We have to guarantee mem->res.limit < mem->memsw.limit.
2605 */
2606 mutex_lock(&set_limit_mutex);
2607 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2608 if (memlimit > val) {
2609 ret = -EINVAL;
2610 mutex_unlock(&set_limit_mutex);
2611 break;
2612 }
2613 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07002614 if (!ret) {
2615 if (memlimit == val)
2616 memcg->memsw_is_minimum = true;
2617 else
2618 memcg->memsw_is_minimum = false;
2619 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002620 mutex_unlock(&set_limit_mutex);
2621
2622 if (!ret)
2623 break;
2624
Balbir Singh4e416952009-09-23 15:56:39 -07002625 mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
Balbir Singh75822b42009-09-23 15:56:38 -07002626 MEM_CGROUP_RECLAIM_NOSWAP |
2627 MEM_CGROUP_RECLAIM_SHRINK);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002628 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002629 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002630 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002631 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002632 else
2633 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002634 }
2635 return ret;
2636}
2637
Balbir Singh4e416952009-09-23 15:56:39 -07002638unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2639 gfp_t gfp_mask, int nid,
2640 int zid)
2641{
2642 unsigned long nr_reclaimed = 0;
2643 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2644 unsigned long reclaimed;
2645 int loop = 0;
2646 struct mem_cgroup_tree_per_zone *mctz;
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07002647 unsigned long long excess;
Balbir Singh4e416952009-09-23 15:56:39 -07002648
2649 if (order > 0)
2650 return 0;
2651
2652 mctz = soft_limit_tree_node_zone(nid, zid);
2653 /*
2654 * This loop can run a while, specially if mem_cgroup's continuously
2655 * keep exceeding their soft limit and putting the system under
2656 * pressure
2657 */
2658 do {
2659 if (next_mz)
2660 mz = next_mz;
2661 else
2662 mz = mem_cgroup_largest_soft_limit_node(mctz);
2663 if (!mz)
2664 break;
2665
2666 reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2667 gfp_mask,
2668 MEM_CGROUP_RECLAIM_SOFT);
2669 nr_reclaimed += reclaimed;
2670 spin_lock(&mctz->lock);
2671
2672 /*
2673 * If we failed to reclaim anything from this memory cgroup
2674 * it is time to move on to the next cgroup
2675 */
2676 next_mz = NULL;
2677 if (!reclaimed) {
2678 do {
2679 /*
2680 * Loop until we find yet another one.
2681 *
2682 * By the time we get the soft_limit lock
2683 * again, someone might have aded the
2684 * group back on the RB tree. Iterate to
2685 * make sure we get a different mem.
2686 * mem_cgroup_largest_soft_limit_node returns
2687 * NULL if no other cgroup is present on
2688 * the tree
2689 */
2690 next_mz =
2691 __mem_cgroup_largest_soft_limit_node(mctz);
2692 if (next_mz == mz) {
2693 css_put(&next_mz->mem->css);
2694 next_mz = NULL;
2695 } else /* next_mz == NULL or other memcg */
2696 break;
2697 } while (1);
2698 }
Balbir Singh4e416952009-09-23 15:56:39 -07002699 __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07002700 excess = res_counter_soft_limit_excess(&mz->mem->res);
Balbir Singh4e416952009-09-23 15:56:39 -07002701 /*
2702 * One school of thought says that we should not add
2703 * back the node to the tree if reclaim returns 0.
2704 * But our reclaim could return 0, simply because due
2705 * to priority we are exposing a smaller subset of
2706 * memory to reclaim from. Consider this as a longer
2707 * term TODO.
2708 */
KAMEZAWA Hiroyukief8745c2009-10-01 15:44:12 -07002709 /* If excess == 0, no tree ops */
2710 __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
Balbir Singh4e416952009-09-23 15:56:39 -07002711 spin_unlock(&mctz->lock);
2712 css_put(&mz->mem->css);
2713 loop++;
2714 /*
2715 * Could not reclaim anything and there are no more
2716 * mem cgroups to try or we seem to be looping without
2717 * reclaiming anything.
2718 */
2719 if (!nr_reclaimed &&
2720 (next_mz == NULL ||
2721 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2722 break;
2723 } while (!nr_reclaimed);
2724 if (next_mz)
2725 css_put(&next_mz->mem->css);
2726 return nr_reclaimed;
2727}
2728
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07002729/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002730 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002731 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2732 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002733static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002734 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002735{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002736 struct zone *zone;
2737 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002738 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002739 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08002740 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002741 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08002742
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002743 zone = &NODE_DATA(node)->node_zones[zid];
2744 mz = mem_cgroup_zoneinfo(mem, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07002745 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002746
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002747 loop = MEM_CGROUP_ZSTAT(mz, lru);
2748 /* give some margin against EBUSY etc...*/
2749 loop += 256;
2750 busy = NULL;
2751 while (loop--) {
2752 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002753 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002754 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002755 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002756 break;
2757 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002758 pc = list_entry(list->prev, struct page_cgroup, lru);
2759 if (busy == pc) {
2760 list_move(&pc->lru, list);
Thiago Farina648bcc72010-03-05 13:42:04 -08002761 busy = NULL;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002762 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002763 continue;
2764 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002765 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002766
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002767 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002768 if (ret == -ENOMEM)
2769 break;
2770
2771 if (ret == -EBUSY || ret == -EINVAL) {
2772 /* found lock contention or "pc" is obsolete. */
2773 busy = pc;
2774 cond_resched();
2775 } else
2776 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002777 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002778
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002779 if (!ret && !list_empty(list))
2780 return -EBUSY;
2781 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002782}
2783
2784/*
2785 * make mem_cgroup's charge to be 0 if there is no task.
2786 * This enables deleting this mem_cgroup.
2787 */
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002788static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002789{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002790 int ret;
2791 int node, zid, shrink;
2792 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002793 struct cgroup *cgrp = mem->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08002794
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002795 css_get(&mem->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002796
2797 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002798 /* should free all ? */
2799 if (free_all)
2800 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002801move_account:
Daisuke Nishimurafce66472010-01-15 17:01:30 -08002802 do {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002803 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002804 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002805 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002806 ret = -EINTR;
2807 if (signal_pending(current))
2808 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002809 /* This is for making all *used* pages to be on LRU. */
2810 lru_add_drain_all();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002811 drain_all_stock_sync();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002812 ret = 0;
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08002813 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002814 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07002815 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002816 for_each_lru(l) {
2817 ret = mem_cgroup_force_empty_list(mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002818 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002819 if (ret)
2820 break;
2821 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002822 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002823 if (ret)
2824 break;
2825 }
2826 /* it seems parent cgroup doesn't have enough mem */
2827 if (ret == -ENOMEM)
2828 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07002829 cond_resched();
Daisuke Nishimurafce66472010-01-15 17:01:30 -08002830 /* "ret" should also be checked to ensure all lists are empty. */
2831 } while (mem->res.usage > 0 || ret);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002832out:
2833 css_put(&mem->css);
2834 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002835
2836try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002837 /* returns EBUSY if there is a task or if we come here twice. */
2838 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002839 ret = -EBUSY;
2840 goto out;
2841 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002842 /* we call try-to-free pages for make this cgroup empty */
2843 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002844 /* try to free all pages in this cgroup */
2845 shrink = 1;
2846 while (nr_retries && mem->res.usage > 0) {
2847 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002848
2849 if (signal_pending(current)) {
2850 ret = -EINTR;
2851 goto out;
2852 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002853 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
2854 false, get_swappiness(mem));
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002855 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002856 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002857 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002858 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002859 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002860
2861 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002862 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002863 /* try move_account...there may be some *locked* pages. */
Daisuke Nishimurafce66472010-01-15 17:01:30 -08002864 goto move_account;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002865}
2866
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002867int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2868{
2869 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2870}
2871
2872
Balbir Singh18f59ea2009-01-07 18:08:07 -08002873static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2874{
2875 return mem_cgroup_from_cont(cont)->use_hierarchy;
2876}
2877
2878static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2879 u64 val)
2880{
2881 int retval = 0;
2882 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2883 struct cgroup *parent = cont->parent;
2884 struct mem_cgroup *parent_mem = NULL;
2885
2886 if (parent)
2887 parent_mem = mem_cgroup_from_cont(parent);
2888
2889 cgroup_lock();
2890 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002891 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002892 * in the child subtrees. If it is unset, then the change can
2893 * occur, provided the current cgroup has no children.
2894 *
2895 * For the root cgroup, parent_mem is NULL, we allow value to be
2896 * set if there are no children.
2897 */
2898 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2899 (val == 1 || val == 0)) {
2900 if (list_empty(&cont->children))
2901 mem->use_hierarchy = val;
2902 else
2903 retval = -EBUSY;
2904 } else
2905 retval = -EINVAL;
2906 cgroup_unlock();
2907
2908 return retval;
2909}
2910
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002911struct mem_cgroup_idx_data {
2912 s64 val;
2913 enum mem_cgroup_stat_index idx;
2914};
2915
2916static int
2917mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
2918{
2919 struct mem_cgroup_idx_data *d = data;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08002920 d->val += mem_cgroup_read_stat(mem, d->idx);
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002921 return 0;
2922}
2923
2924static void
2925mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
2926 enum mem_cgroup_stat_index idx, s64 *val)
2927{
2928 struct mem_cgroup_idx_data d;
2929 d.idx = idx;
2930 d.val = 0;
2931 mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
2932 *val = d.val;
2933}
2934
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08002935static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
2936{
2937 u64 idx_val, val;
2938
2939 if (!mem_cgroup_is_root(mem)) {
2940 if (!swap)
2941 return res_counter_read_u64(&mem->res, RES_USAGE);
2942 else
2943 return res_counter_read_u64(&mem->memsw, RES_USAGE);
2944 }
2945
2946 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val);
2947 val = idx_val;
2948 mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val);
2949 val += idx_val;
2950
2951 if (swap) {
2952 mem_cgroup_get_recursive_idx_stat(mem,
2953 MEM_CGROUP_STAT_SWAPOUT, &idx_val);
2954 val += idx_val;
2955 }
2956
2957 return val << PAGE_SHIFT;
2958}
2959
Paul Menage2c3daa72008-04-29 00:59:58 -07002960static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002961{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002962 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08002963 u64 val;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002964 int type, name;
2965
2966 type = MEMFILE_TYPE(cft->private);
2967 name = MEMFILE_ATTR(cft->private);
2968 switch (type) {
2969 case _MEM:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08002970 if (name == RES_USAGE)
2971 val = mem_cgroup_usage(mem, false);
2972 else
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002973 val = res_counter_read_u64(&mem->res, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002974 break;
2975 case _MEMSWAP:
Kirill A. Shutemov104f3922010-03-10 15:22:21 -08002976 if (name == RES_USAGE)
2977 val = mem_cgroup_usage(mem, true);
2978 else
Balbir Singh0c3e73e2009-09-23 15:56:42 -07002979 val = res_counter_read_u64(&mem->memsw, name);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002980 break;
2981 default:
2982 BUG();
2983 break;
2984 }
2985 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002986}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002987/*
2988 * The user of this function is...
2989 * RES_LIMIT.
2990 */
Paul Menage856c13a2008-07-25 01:47:04 -07002991static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2992 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002993{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002994 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002995 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002996 unsigned long long val;
2997 int ret;
2998
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002999 type = MEMFILE_TYPE(cft->private);
3000 name = MEMFILE_ATTR(cft->private);
3001 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003002 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003003 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3004 ret = -EINVAL;
3005 break;
3006 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003007 /* This function does all necessary parse...reuse it */
3008 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003009 if (ret)
3010 break;
3011 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003012 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003013 else
3014 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003015 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003016 case RES_SOFT_LIMIT:
3017 ret = res_counter_memparse_write_strategy(buffer, &val);
3018 if (ret)
3019 break;
3020 /*
3021 * For memsw, soft limits are hard to implement in terms
3022 * of semantics, for now, we support soft limits for
3023 * control without swap
3024 */
3025 if (type == _MEM)
3026 ret = res_counter_set_soft_limit(&memcg->res, val);
3027 else
3028 ret = -EINVAL;
3029 break;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003030 default:
3031 ret = -EINVAL; /* should be BUG() ? */
3032 break;
3033 }
3034 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003035}
3036
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003037static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3038 unsigned long long *mem_limit, unsigned long long *memsw_limit)
3039{
3040 struct cgroup *cgroup;
3041 unsigned long long min_limit, min_memsw_limit, tmp;
3042
3043 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3044 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3045 cgroup = memcg->css.cgroup;
3046 if (!memcg->use_hierarchy)
3047 goto out;
3048
3049 while (cgroup->parent) {
3050 cgroup = cgroup->parent;
3051 memcg = mem_cgroup_from_cont(cgroup);
3052 if (!memcg->use_hierarchy)
3053 break;
3054 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3055 min_limit = min(min_limit, tmp);
3056 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3057 min_memsw_limit = min(min_memsw_limit, tmp);
3058 }
3059out:
3060 *mem_limit = min_limit;
3061 *memsw_limit = min_memsw_limit;
3062 return;
3063}
3064
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003065static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003066{
3067 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003068 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003069
3070 mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003071 type = MEMFILE_TYPE(event);
3072 name = MEMFILE_ATTR(event);
3073 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003074 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003075 if (type == _MEM)
3076 res_counter_reset_max(&mem->res);
3077 else
3078 res_counter_reset_max(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003079 break;
3080 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003081 if (type == _MEM)
3082 res_counter_reset_failcnt(&mem->res);
3083 else
3084 res_counter_reset_failcnt(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003085 break;
3086 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003087
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07003088 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003089}
3090
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003091static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3092 struct cftype *cft)
3093{
3094 return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3095}
3096
Daisuke Nishimura02491442010-03-10 15:22:17 -08003097#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003098static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3099 struct cftype *cft, u64 val)
3100{
3101 struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3102
3103 if (val >= (1 << NR_MOVE_TYPE))
3104 return -EINVAL;
3105 /*
3106 * We check this value several times in both in can_attach() and
3107 * attach(), so we need cgroup lock to prevent this value from being
3108 * inconsistent.
3109 */
3110 cgroup_lock();
3111 mem->move_charge_at_immigrate = val;
3112 cgroup_unlock();
3113
3114 return 0;
3115}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003116#else
3117static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3118 struct cftype *cft, u64 val)
3119{
3120 return -ENOSYS;
3121}
3122#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003123
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003124
3125/* For read statistics */
3126enum {
3127 MCS_CACHE,
3128 MCS_RSS,
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08003129 MCS_FILE_MAPPED,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003130 MCS_PGPGIN,
3131 MCS_PGPGOUT,
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003132 MCS_SWAP,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003133 MCS_INACTIVE_ANON,
3134 MCS_ACTIVE_ANON,
3135 MCS_INACTIVE_FILE,
3136 MCS_ACTIVE_FILE,
3137 MCS_UNEVICTABLE,
3138 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003139};
3140
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003141struct mcs_total_stat {
3142 s64 stat[NR_MCS_STAT];
3143};
3144
3145struct {
3146 char *local_name;
3147 char *total_name;
3148} memcg_stat_strings[NR_MCS_STAT] = {
3149 {"cache", "total_cache"},
3150 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07003151 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003152 {"pgpgin", "total_pgpgin"},
3153 {"pgpgout", "total_pgpgout"},
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003154 {"swap", "total_swap"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003155 {"inactive_anon", "total_inactive_anon"},
3156 {"active_anon", "total_active_anon"},
3157 {"inactive_file", "total_inactive_file"},
3158 {"active_file", "total_active_file"},
3159 {"unevictable", "total_unevictable"}
3160};
3161
3162
3163static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
3164{
3165 struct mcs_total_stat *s = data;
3166 s64 val;
3167
3168 /* per cpu stat */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003169 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003170 s->stat[MCS_CACHE] += val * PAGE_SIZE;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003171 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003172 s->stat[MCS_RSS] += val * PAGE_SIZE;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003173 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -08003174 s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003175 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003176 s->stat[MCS_PGPGIN] += val;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003177 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003178 s->stat[MCS_PGPGOUT] += val;
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003179 if (do_swap_account) {
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003180 val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003181 s->stat[MCS_SWAP] += val * PAGE_SIZE;
3182 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003183
3184 /* per zone stat */
3185 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3186 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3187 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3188 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3189 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3190 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3191 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3192 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3193 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3194 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3195 return 0;
3196}
3197
3198static void
3199mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3200{
3201 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
3202}
3203
Paul Menagec64745c2008-04-29 01:00:02 -07003204static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3205 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003206{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003207 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003208 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003209 int i;
3210
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003211 memset(&mystat, 0, sizeof(mystat));
3212 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003213
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003214 for (i = 0; i < NR_MCS_STAT; i++) {
3215 if (i == MCS_SWAP && !do_swap_account)
3216 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003217 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003218 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003219
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003220 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003221 {
3222 unsigned long long limit, memsw_limit;
3223 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3224 cb->fill(cb, "hierarchical_memory_limit", limit);
3225 if (do_swap_account)
3226 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3227 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003228
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003229 memset(&mystat, 0, sizeof(mystat));
3230 mem_cgroup_get_total_stat(mem_cont, &mystat);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003231 for (i = 0; i < NR_MCS_STAT; i++) {
3232 if (i == MCS_SWAP && !do_swap_account)
3233 continue;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003234 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003235 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003236
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003237#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08003238 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003239
3240 {
3241 int nid, zid;
3242 struct mem_cgroup_per_zone *mz;
3243 unsigned long recent_rotated[2] = {0, 0};
3244 unsigned long recent_scanned[2] = {0, 0};
3245
3246 for_each_online_node(nid)
3247 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3248 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3249
3250 recent_rotated[0] +=
3251 mz->reclaim_stat.recent_rotated[0];
3252 recent_rotated[1] +=
3253 mz->reclaim_stat.recent_rotated[1];
3254 recent_scanned[0] +=
3255 mz->reclaim_stat.recent_scanned[0];
3256 recent_scanned[1] +=
3257 mz->reclaim_stat.recent_scanned[1];
3258 }
3259 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3260 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3261 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3262 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3263 }
3264#endif
3265
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003266 return 0;
3267}
3268
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003269static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3270{
3271 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3272
3273 return get_swappiness(memcg);
3274}
3275
3276static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3277 u64 val)
3278{
3279 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3280 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08003281
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003282 if (val > 100)
3283 return -EINVAL;
3284
3285 if (cgrp->parent == NULL)
3286 return -EINVAL;
3287
3288 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08003289
3290 cgroup_lock();
3291
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003292 /* If under hierarchy, only empty-root can set this value */
3293 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08003294 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3295 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003296 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08003297 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003298
3299 spin_lock(&memcg->reclaim_param_lock);
3300 memcg->swappiness = val;
3301 spin_unlock(&memcg->reclaim_param_lock);
3302
Li Zefan068b38c2009-01-15 13:51:26 -08003303 cgroup_unlock();
3304
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003305 return 0;
3306}
3307
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003308static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3309{
3310 struct mem_cgroup_threshold_ary *t;
3311 u64 usage;
3312 int i;
3313
3314 rcu_read_lock();
3315 if (!swap)
3316 t = rcu_dereference(memcg->thresholds);
3317 else
3318 t = rcu_dereference(memcg->memsw_thresholds);
3319
3320 if (!t)
3321 goto unlock;
3322
3323 usage = mem_cgroup_usage(memcg, swap);
3324
3325 /*
3326 * current_threshold points to threshold just below usage.
3327 * If it's not true, a threshold was crossed after last
3328 * call of __mem_cgroup_threshold().
3329 */
3330 i = atomic_read(&t->current_threshold);
3331
3332 /*
3333 * Iterate backward over array of thresholds starting from
3334 * current_threshold and check if a threshold is crossed.
3335 * If none of thresholds below usage is crossed, we read
3336 * only one element of the array here.
3337 */
3338 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3339 eventfd_signal(t->entries[i].eventfd, 1);
3340
3341 /* i = current_threshold + 1 */
3342 i++;
3343
3344 /*
3345 * Iterate forward over array of thresholds starting from
3346 * current_threshold+1 and check if a threshold is crossed.
3347 * If none of thresholds above usage is crossed, we read
3348 * only one element of the array here.
3349 */
3350 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3351 eventfd_signal(t->entries[i].eventfd, 1);
3352
3353 /* Update current_threshold */
3354 atomic_set(&t->current_threshold, i - 1);
3355unlock:
3356 rcu_read_unlock();
3357}
3358
3359static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3360{
3361 __mem_cgroup_threshold(memcg, false);
3362 if (do_swap_account)
3363 __mem_cgroup_threshold(memcg, true);
3364}
3365
3366static int compare_thresholds(const void *a, const void *b)
3367{
3368 const struct mem_cgroup_threshold *_a = a;
3369 const struct mem_cgroup_threshold *_b = b;
3370
3371 return _a->threshold - _b->threshold;
3372}
3373
3374static int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft,
3375 struct eventfd_ctx *eventfd, const char *args)
3376{
3377 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3378 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3379 int type = MEMFILE_TYPE(cft->private);
3380 u64 threshold, usage;
3381 int size;
3382 int i, ret;
3383
3384 ret = res_counter_memparse_write_strategy(args, &threshold);
3385 if (ret)
3386 return ret;
3387
3388 mutex_lock(&memcg->thresholds_lock);
3389 if (type == _MEM)
3390 thresholds = memcg->thresholds;
3391 else if (type == _MEMSWAP)
3392 thresholds = memcg->memsw_thresholds;
3393 else
3394 BUG();
3395
3396 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3397
3398 /* Check if a threshold crossed before adding a new one */
3399 if (thresholds)
3400 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3401
3402 if (thresholds)
3403 size = thresholds->size + 1;
3404 else
3405 size = 1;
3406
3407 /* Allocate memory for new array of thresholds */
3408 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3409 size * sizeof(struct mem_cgroup_threshold),
3410 GFP_KERNEL);
3411 if (!thresholds_new) {
3412 ret = -ENOMEM;
3413 goto unlock;
3414 }
3415 thresholds_new->size = size;
3416
3417 /* Copy thresholds (if any) to new array */
3418 if (thresholds)
3419 memcpy(thresholds_new->entries, thresholds->entries,
3420 thresholds->size *
3421 sizeof(struct mem_cgroup_threshold));
3422 /* Add new threshold */
3423 thresholds_new->entries[size - 1].eventfd = eventfd;
3424 thresholds_new->entries[size - 1].threshold = threshold;
3425
3426 /* Sort thresholds. Registering of new threshold isn't time-critical */
3427 sort(thresholds_new->entries, size,
3428 sizeof(struct mem_cgroup_threshold),
3429 compare_thresholds, NULL);
3430
3431 /* Find current threshold */
3432 atomic_set(&thresholds_new->current_threshold, -1);
3433 for (i = 0; i < size; i++) {
3434 if (thresholds_new->entries[i].threshold < usage) {
3435 /*
3436 * thresholds_new->current_threshold will not be used
3437 * until rcu_assign_pointer(), so it's safe to increment
3438 * it here.
3439 */
3440 atomic_inc(&thresholds_new->current_threshold);
3441 }
3442 }
3443
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444 if (type == _MEM)
3445 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3446 else
3447 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3448
3449 /* To be sure that nobody uses thresholds before freeing it */
3450 synchronize_rcu();
3451
3452 kfree(thresholds);
3453unlock:
3454 mutex_unlock(&memcg->thresholds_lock);
3455
3456 return ret;
3457}
3458
3459static int mem_cgroup_unregister_event(struct cgroup *cgrp, struct cftype *cft,
3460 struct eventfd_ctx *eventfd)
3461{
3462 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3463 struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
3464 int type = MEMFILE_TYPE(cft->private);
3465 u64 usage;
3466 int size = 0;
3467 int i, j, ret;
3468
3469 mutex_lock(&memcg->thresholds_lock);
3470 if (type == _MEM)
3471 thresholds = memcg->thresholds;
3472 else if (type == _MEMSWAP)
3473 thresholds = memcg->memsw_thresholds;
3474 else
3475 BUG();
3476
3477 /*
3478 * Something went wrong if we trying to unregister a threshold
3479 * if we don't have thresholds
3480 */
3481 BUG_ON(!thresholds);
3482
3483 usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3484
3485 /* Check if a threshold crossed before removing */
3486 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3487
3488 /* Calculate new number of threshold */
3489 for (i = 0; i < thresholds->size; i++) {
3490 if (thresholds->entries[i].eventfd != eventfd)
3491 size++;
3492 }
3493
3494 /* Set thresholds array to NULL if we don't have thresholds */
3495 if (!size) {
3496 thresholds_new = NULL;
3497 goto assign;
3498 }
3499
3500 /* Allocate memory for new array of thresholds */
3501 thresholds_new = kmalloc(sizeof(*thresholds_new) +
3502 size * sizeof(struct mem_cgroup_threshold),
3503 GFP_KERNEL);
3504 if (!thresholds_new) {
3505 ret = -ENOMEM;
3506 goto unlock;
3507 }
3508 thresholds_new->size = size;
3509
3510 /* Copy thresholds and find current threshold */
3511 atomic_set(&thresholds_new->current_threshold, -1);
3512 for (i = 0, j = 0; i < thresholds->size; i++) {
3513 if (thresholds->entries[i].eventfd == eventfd)
3514 continue;
3515
3516 thresholds_new->entries[j] = thresholds->entries[i];
3517 if (thresholds_new->entries[j].threshold < usage) {
3518 /*
3519 * thresholds_new->current_threshold will not be used
3520 * until rcu_assign_pointer(), so it's safe to increment
3521 * it here.
3522 */
3523 atomic_inc(&thresholds_new->current_threshold);
3524 }
3525 j++;
3526 }
3527
3528assign:
3529 if (type == _MEM)
3530 rcu_assign_pointer(memcg->thresholds, thresholds_new);
3531 else
3532 rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new);
3533
3534 /* To be sure that nobody uses thresholds before freeing it */
3535 synchronize_rcu();
3536
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003537 kfree(thresholds);
3538unlock:
3539 mutex_unlock(&memcg->thresholds_lock);
3540
3541 return ret;
3542}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003543
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003544static struct cftype mem_cgroup_files[] = {
3545 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003546 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003547 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07003548 .read_u64 = mem_cgroup_read,
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003549 .register_event = mem_cgroup_register_event,
3550 .unregister_event = mem_cgroup_unregister_event,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003551 },
3552 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003553 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003554 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003555 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003556 .read_u64 = mem_cgroup_read,
3557 },
3558 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003559 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003560 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07003561 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07003562 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003563 },
3564 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003565 .name = "soft_limit_in_bytes",
3566 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3567 .write_string = mem_cgroup_write,
3568 .read_u64 = mem_cgroup_read,
3569 },
3570 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003571 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003572 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003573 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07003574 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003575 },
Balbir Singh8697d332008-02-07 00:13:59 -08003576 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003577 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07003578 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003579 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003580 {
3581 .name = "force_empty",
3582 .trigger = mem_cgroup_force_empty_write,
3583 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003584 {
3585 .name = "use_hierarchy",
3586 .write_u64 = mem_cgroup_hierarchy_write,
3587 .read_u64 = mem_cgroup_hierarchy_read,
3588 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003589 {
3590 .name = "swappiness",
3591 .read_u64 = mem_cgroup_swappiness_read,
3592 .write_u64 = mem_cgroup_swappiness_write,
3593 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003594 {
3595 .name = "move_charge_at_immigrate",
3596 .read_u64 = mem_cgroup_move_charge_read,
3597 .write_u64 = mem_cgroup_move_charge_write,
3598 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003599};
3600
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003601#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3602static struct cftype memsw_cgroup_files[] = {
3603 {
3604 .name = "memsw.usage_in_bytes",
3605 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3606 .read_u64 = mem_cgroup_read,
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003607 .register_event = mem_cgroup_register_event,
3608 .unregister_event = mem_cgroup_unregister_event,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003609 },
3610 {
3611 .name = "memsw.max_usage_in_bytes",
3612 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3613 .trigger = mem_cgroup_reset,
3614 .read_u64 = mem_cgroup_read,
3615 },
3616 {
3617 .name = "memsw.limit_in_bytes",
3618 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3619 .write_string = mem_cgroup_write,
3620 .read_u64 = mem_cgroup_read,
3621 },
3622 {
3623 .name = "memsw.failcnt",
3624 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3625 .trigger = mem_cgroup_reset,
3626 .read_u64 = mem_cgroup_read,
3627 },
3628};
3629
3630static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3631{
3632 if (!do_swap_account)
3633 return 0;
3634 return cgroup_add_files(cont, ss, memsw_cgroup_files,
3635 ARRAY_SIZE(memsw_cgroup_files));
3636};
3637#else
3638static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3639{
3640 return 0;
3641}
3642#endif
3643
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003644static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3645{
3646 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003647 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07003648 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07003649 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003650 /*
3651 * This routine is called against possible nodes.
3652 * But it's BUG to call kmalloc() against offline node.
3653 *
3654 * TODO: this routine can waste much memory for nodes which will
3655 * never be onlined. It's better to use memory hotplug callback
3656 * function.
3657 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07003658 if (!node_state(node, N_NORMAL_MEMORY))
3659 tmp = -1;
3660 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003661 if (!pn)
3662 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003663
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003664 mem->info.nodeinfo[node] = pn;
3665 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003666
3667 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3668 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07003669 for_each_lru(l)
3670 INIT_LIST_HEAD(&mz->lists[l]);
Balbir Singhf64c3f52009-09-23 15:56:37 -07003671 mz->usage_in_excess = 0;
Balbir Singh4e416952009-09-23 15:56:39 -07003672 mz->on_tree = false;
3673 mz->mem = mem;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003674 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003675 return 0;
3676}
3677
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08003678static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3679{
3680 kfree(mem->info.nodeinfo[node]);
3681}
3682
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003683static struct mem_cgroup *mem_cgroup_alloc(void)
3684{
3685 struct mem_cgroup *mem;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003686 int size = sizeof(struct mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003687
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003688 /* Can be very big if MAX_NUMNODES is very big */
Jan Blunckc8dad2b2009-01-07 18:07:53 -08003689 if (size < PAGE_SIZE)
3690 mem = kmalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003691 else
Jan Blunckc8dad2b2009-01-07 18:07:53 -08003692 mem = vmalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003693
3694 if (mem)
Jan Blunckc8dad2b2009-01-07 18:07:53 -08003695 memset(mem, 0, size);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003696 mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
3697 if (!mem->stat) {
3698 if (size < PAGE_SIZE)
3699 kfree(mem);
3700 else
3701 vfree(mem);
3702 mem = NULL;
3703 }
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003704 return mem;
3705}
3706
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003707/*
3708 * At destroying mem_cgroup, references from swap_cgroup can remain.
3709 * (scanning all at force_empty is too costly...)
3710 *
3711 * Instead of clearing all references at force_empty, we remember
3712 * the number of reference from swap_cgroup and free mem_cgroup when
3713 * it goes down to 0.
3714 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003715 * Removal of cgroup itself succeeds regardless of refs from swap.
3716 */
3717
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08003718static void __mem_cgroup_free(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003719{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003720 int node;
3721
Balbir Singhf64c3f52009-09-23 15:56:37 -07003722 mem_cgroup_remove_from_trees(mem);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003723 free_css_id(&mem_cgroup_subsys, &mem->css);
3724
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08003725 for_each_node_state(node, N_POSSIBLE)
3726 free_mem_cgroup_per_zone_info(mem, node);
3727
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -08003728 free_percpu(mem->stat);
3729 if (sizeof(struct mem_cgroup) < PAGE_SIZE)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003730 kfree(mem);
3731 else
3732 vfree(mem);
3733}
3734
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003735static void mem_cgroup_get(struct mem_cgroup *mem)
3736{
3737 atomic_inc(&mem->refcnt);
3738}
3739
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003740static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003741{
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003742 if (atomic_sub_and_test(count, &mem->refcnt)) {
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08003743 struct mem_cgroup *parent = parent_mem_cgroup(mem);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08003744 __mem_cgroup_free(mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08003745 if (parent)
3746 mem_cgroup_put(parent);
3747 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003748}
3749
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08003750static void mem_cgroup_put(struct mem_cgroup *mem)
3751{
3752 __mem_cgroup_put(mem, 1);
3753}
3754
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08003755/*
3756 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3757 */
3758static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
3759{
3760 if (!mem->res.parent)
3761 return NULL;
3762 return mem_cgroup_from_res_counter(mem->res.parent, res);
3763}
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07003764
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08003765#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3766static void __init enable_swap_cgroup(void)
3767{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08003768 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08003769 do_swap_account = 1;
3770}
3771#else
3772static void __init enable_swap_cgroup(void)
3773{
3774}
3775#endif
3776
Balbir Singhf64c3f52009-09-23 15:56:37 -07003777static int mem_cgroup_soft_limit_tree_init(void)
3778{
3779 struct mem_cgroup_tree_per_node *rtpn;
3780 struct mem_cgroup_tree_per_zone *rtpz;
3781 int tmp, node, zone;
3782
3783 for_each_node_state(node, N_POSSIBLE) {
3784 tmp = node;
3785 if (!node_state(node, N_NORMAL_MEMORY))
3786 tmp = -1;
3787 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
3788 if (!rtpn)
3789 return 1;
3790
3791 soft_limit_tree.rb_tree_per_node[node] = rtpn;
3792
3793 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3794 rtpz = &rtpn->rb_tree_per_zone[zone];
3795 rtpz->rb_root = RB_ROOT;
3796 spin_lock_init(&rtpz->lock);
3797 }
3798 }
3799 return 0;
3800}
3801
Li Zefan0eb253e2009-01-15 13:51:25 -08003802static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003803mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
3804{
Balbir Singh28dbc4b2009-01-07 18:08:05 -08003805 struct mem_cgroup *mem, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003806 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003807 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003808
Jan Blunckc8dad2b2009-01-07 18:07:53 -08003809 mem = mem_cgroup_alloc();
3810 if (!mem)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003811 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08003812
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003813 for_each_node_state(node, N_POSSIBLE)
3814 if (alloc_mem_cgroup_per_zone_info(mem, node))
3815 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07003816
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08003817 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08003818 if (cont->parent == NULL) {
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08003819 int cpu;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08003820 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08003821 parent = NULL;
Balbir Singh4b3bde42009-09-23 15:56:32 -07003822 root_mem_cgroup = mem;
Balbir Singhf64c3f52009-09-23 15:56:37 -07003823 if (mem_cgroup_soft_limit_tree_init())
3824 goto free_out;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08003825 for_each_possible_cpu(cpu) {
3826 struct memcg_stock_pcp *stock =
3827 &per_cpu(memcg_stock, cpu);
3828 INIT_WORK(&stock->work, drain_local_stock);
3829 }
3830 hotcpu_notifier(memcg_stock_cpu_callback, 0);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003831 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08003832 parent = mem_cgroup_from_cont(cont->parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003833 mem->use_hierarchy = parent->use_hierarchy;
3834 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08003835
Balbir Singh18f59ea2009-01-07 18:08:07 -08003836 if (parent && parent->use_hierarchy) {
3837 res_counter_init(&mem->res, &parent->res);
3838 res_counter_init(&mem->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08003839 /*
3840 * We increment refcnt of the parent to ensure that we can
3841 * safely access it on res_counter_charge/uncharge.
3842 * This refcnt will be decremented when freeing this
3843 * mem_cgroup(see mem_cgroup_put).
3844 */
3845 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003846 } else {
3847 res_counter_init(&mem->res, NULL);
3848 res_counter_init(&mem->memsw, NULL);
3849 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003850 mem->last_scanned_child = 0;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -08003851 spin_lock_init(&mem->reclaim_param_lock);
Balbir Singh6d61ef42009-01-07 18:08:06 -08003852
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003853 if (parent)
3854 mem->swappiness = get_swappiness(parent);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08003855 atomic_set(&mem->refcnt, 1);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003856 mem->move_charge_at_immigrate = 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003857 mutex_init(&mem->thresholds_lock);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003858 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003859free_out:
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08003860 __mem_cgroup_free(mem);
Balbir Singh4b3bde42009-09-23 15:56:32 -07003861 root_mem_cgroup = NULL;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07003862 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003863}
3864
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003865static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08003866 struct cgroup *cont)
3867{
3868 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003869
3870 return mem_cgroup_force_empty(mem, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08003871}
3872
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003873static void mem_cgroup_destroy(struct cgroup_subsys *ss,
3874 struct cgroup *cont)
3875{
Daisuke Nishimurac268e992009-01-15 13:51:13 -08003876 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08003877
Daisuke Nishimurac268e992009-01-15 13:51:13 -08003878 mem_cgroup_put(mem);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003879}
3880
3881static int mem_cgroup_populate(struct cgroup_subsys *ss,
3882 struct cgroup *cont)
3883{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08003884 int ret;
3885
3886 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
3887 ARRAY_SIZE(mem_cgroup_files));
3888
3889 if (!ret)
3890 ret = register_memsw_files(cont, ss);
3891 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003892}
3893
Daisuke Nishimura02491442010-03-10 15:22:17 -08003894#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003895/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003896#define PRECHARGE_COUNT_AT_ONCE 256
3897static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003898{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003899 int ret = 0;
3900 int batch_count = PRECHARGE_COUNT_AT_ONCE;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003901 struct mem_cgroup *mem = mc.to;
3902
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003903 if (mem_cgroup_is_root(mem)) {
3904 mc.precharge += count;
3905 /* we don't need css_get for root */
3906 return ret;
3907 }
3908 /* try to charge at once */
3909 if (count > 1) {
3910 struct res_counter *dummy;
3911 /*
3912 * "mem" cannot be under rmdir() because we've already checked
3913 * by cgroup_lock_live_cgroup() that it is not removed and we
3914 * are still under the same cgroup_mutex. So we can postpone
3915 * css_get().
3916 */
3917 if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
3918 goto one_by_one;
3919 if (do_swap_account && res_counter_charge(&mem->memsw,
3920 PAGE_SIZE * count, &dummy)) {
3921 res_counter_uncharge(&mem->res, PAGE_SIZE * count);
3922 goto one_by_one;
3923 }
3924 mc.precharge += count;
3925 VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
3926 WARN_ON_ONCE(count > INT_MAX);
3927 __css_get(&mem->css, (int)count);
3928 return ret;
3929 }
3930one_by_one:
3931 /* fall back to one by one charge */
3932 while (count--) {
3933 if (signal_pending(current)) {
3934 ret = -EINTR;
3935 break;
3936 }
3937 if (!batch_count--) {
3938 batch_count = PRECHARGE_COUNT_AT_ONCE;
3939 cond_resched();
3940 }
KAMEZAWA Hiroyuki430e48632010-03-10 15:22:30 -08003941 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08003942 if (ret || !mem)
3943 /* mem_cgroup_clear_mc() will do uncharge later */
3944 return -ENOMEM;
3945 mc.precharge++;
3946 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003947 return ret;
3948}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003949#else /* !CONFIG_MMU */
3950static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
3951 struct cgroup *cgroup,
3952 struct task_struct *p,
3953 bool threadgroup)
3954{
3955 return 0;
3956}
3957static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
3958 struct cgroup *cgroup,
3959 struct task_struct *p,
3960 bool threadgroup)
3961{
3962}
3963static void mem_cgroup_move_task(struct cgroup_subsys *ss,
3964 struct cgroup *cont,
3965 struct cgroup *old_cont,
3966 struct task_struct *p,
3967 bool threadgroup)
3968{
3969}
3970#endif
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003971
3972/**
3973 * is_target_pte_for_mc - check a pte whether it is valid for move charge
3974 * @vma: the vma the pte to be checked belongs
3975 * @addr: the address corresponding to the pte to be checked
3976 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08003977 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003978 *
3979 * Returns
3980 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
3981 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
3982 * move charge. if @target is not NULL, the page is stored in target->page
3983 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08003984 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
3985 * target for charge migration. if @target is not NULL, the entry is stored
3986 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003987 *
3988 * Called with pte lock held.
3989 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003990union mc_target {
3991 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08003992 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003993};
3994
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003995enum mc_target_type {
3996 MC_TARGET_NONE, /* not used */
3997 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003998 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08003999};
4000
4001static int is_target_pte_for_mc(struct vm_area_struct *vma,
4002 unsigned long addr, pte_t ptent, union mc_target *target)
4003{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004004 struct page *page = NULL;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004005 struct page_cgroup *pc;
4006 int ret = 0;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004007 swp_entry_t ent = { .val = 0 };
4008 int usage_count = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004009 bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON,
4010 &mc.to->move_charge_at_immigrate);
4011
Daisuke Nishimura02491442010-03-10 15:22:17 -08004012 if (!pte_present(ptent)) {
4013 /* TODO: handle swap of shmes/tmpfs */
4014 if (pte_none(ptent) || pte_file(ptent))
4015 return 0;
4016 else if (is_swap_pte(ptent)) {
4017 ent = pte_to_swp_entry(ptent);
4018 if (!move_anon || non_swap_entry(ent))
4019 return 0;
4020 usage_count = mem_cgroup_count_swap_user(ent, &page);
4021 }
4022 } else {
4023 page = vm_normal_page(vma, addr, ptent);
4024 if (!page || !page_mapped(page))
4025 return 0;
4026 /*
4027 * TODO: We don't move charges of file(including shmem/tmpfs)
4028 * pages for now.
4029 */
4030 if (!move_anon || !PageAnon(page))
4031 return 0;
4032 if (!get_page_unless_zero(page))
4033 return 0;
4034 usage_count = page_mapcount(page);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004035 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004036 if (usage_count > 1) {
4037 /*
4038 * TODO: We don't move charges of shared(used by multiple
4039 * processes) pages for now.
4040 */
4041 if (page)
4042 put_page(page);
4043 return 0;
4044 }
4045 if (page) {
4046 pc = lookup_page_cgroup(page);
4047 /*
4048 * Do only loose check w/o page_cgroup lock.
4049 * mem_cgroup_move_account() checks the pc is valid or not under
4050 * the lock.
4051 */
4052 if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
4053 ret = MC_TARGET_PAGE;
4054 if (target)
4055 target->page = page;
4056 }
4057 if (!ret || !target)
4058 put_page(page);
4059 }
4060 /* throught */
4061 if (ent.val && do_swap_account && !ret &&
4062 css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
4063 ret = MC_TARGET_SWAP;
4064 if (target)
4065 target->ent = ent;
4066 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004067 return ret;
4068}
4069
4070static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4071 unsigned long addr, unsigned long end,
4072 struct mm_walk *walk)
4073{
4074 struct vm_area_struct *vma = walk->private;
4075 pte_t *pte;
4076 spinlock_t *ptl;
4077
4078 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4079 for (; addr != end; pte++, addr += PAGE_SIZE)
4080 if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4081 mc.precharge++; /* increment precharge temporarily */
4082 pte_unmap_unlock(pte - 1, ptl);
4083 cond_resched();
4084
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004085 return 0;
4086}
4087
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004088static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4089{
4090 unsigned long precharge;
4091 struct vm_area_struct *vma;
4092
4093 down_read(&mm->mmap_sem);
4094 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4095 struct mm_walk mem_cgroup_count_precharge_walk = {
4096 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4097 .mm = mm,
4098 .private = vma,
4099 };
4100 if (is_vm_hugetlb_page(vma))
4101 continue;
4102 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4103 if (vma->vm_flags & VM_SHARED)
4104 continue;
4105 walk_page_range(vma->vm_start, vma->vm_end,
4106 &mem_cgroup_count_precharge_walk);
4107 }
4108 up_read(&mm->mmap_sem);
4109
4110 precharge = mc.precharge;
4111 mc.precharge = 0;
4112
4113 return precharge;
4114}
4115
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004116static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4117{
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004118 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm));
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004119}
4120
4121static void mem_cgroup_clear_mc(void)
4122{
4123 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004124 if (mc.precharge) {
4125 __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4126 mc.precharge = 0;
4127 }
4128 /*
4129 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4130 * we must uncharge here.
4131 */
4132 if (mc.moved_charge) {
4133 __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4134 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004135 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004136 /* we must fixup refcnts and charges */
4137 if (mc.moved_swap) {
4138 WARN_ON_ONCE(mc.moved_swap > INT_MAX);
4139 /* uncharge swap account from the old cgroup */
4140 if (!mem_cgroup_is_root(mc.from))
4141 res_counter_uncharge(&mc.from->memsw,
4142 PAGE_SIZE * mc.moved_swap);
4143 __mem_cgroup_put(mc.from, mc.moved_swap);
4144
4145 if (!mem_cgroup_is_root(mc.to)) {
4146 /*
4147 * we charged both to->res and to->memsw, so we should
4148 * uncharge to->res.
4149 */
4150 res_counter_uncharge(&mc.to->res,
4151 PAGE_SIZE * mc.moved_swap);
4152 VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags));
4153 __css_put(&mc.to->css, mc.moved_swap);
4154 }
4155 /* we've already done mem_cgroup_get(mc.to) */
4156
4157 mc.moved_swap = 0;
4158 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004159 mc.from = NULL;
4160 mc.to = NULL;
Daisuke Nishimura8033b972010-03-10 15:22:16 -08004161 mc.moving_task = NULL;
4162 wake_up_all(&mc.waitq);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004163}
4164
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004165static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4166 struct cgroup *cgroup,
4167 struct task_struct *p,
4168 bool threadgroup)
4169{
4170 int ret = 0;
4171 struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4172
4173 if (mem->move_charge_at_immigrate) {
4174 struct mm_struct *mm;
4175 struct mem_cgroup *from = mem_cgroup_from_task(p);
4176
4177 VM_BUG_ON(from == mem);
4178
4179 mm = get_task_mm(p);
4180 if (!mm)
4181 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004182 /* We move charges only when we move a owner of the mm */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004183 if (mm->owner == p) {
4184 VM_BUG_ON(mc.from);
4185 VM_BUG_ON(mc.to);
4186 VM_BUG_ON(mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004187 VM_BUG_ON(mc.moved_charge);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004188 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura8033b972010-03-10 15:22:16 -08004189 VM_BUG_ON(mc.moving_task);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004190 mc.from = from;
4191 mc.to = mem;
4192 mc.precharge = 0;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004193 mc.moved_charge = 0;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004194 mc.moved_swap = 0;
Daisuke Nishimura8033b972010-03-10 15:22:16 -08004195 mc.moving_task = current;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004196
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004197 ret = mem_cgroup_precharge_mc(mm);
4198 if (ret)
4199 mem_cgroup_clear_mc();
4200 }
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004201 mmput(mm);
4202 }
4203 return ret;
4204}
4205
4206static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4207 struct cgroup *cgroup,
4208 struct task_struct *p,
4209 bool threadgroup)
4210{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004211 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004212}
4213
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004214static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4215 unsigned long addr, unsigned long end,
4216 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004217{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004218 int ret = 0;
4219 struct vm_area_struct *vma = walk->private;
4220 pte_t *pte;
4221 spinlock_t *ptl;
4222
4223retry:
4224 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4225 for (; addr != end; addr += PAGE_SIZE) {
4226 pte_t ptent = *(pte++);
4227 union mc_target target;
4228 int type;
4229 struct page *page;
4230 struct page_cgroup *pc;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004231 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004232
4233 if (!mc.precharge)
4234 break;
4235
4236 type = is_target_pte_for_mc(vma, addr, ptent, &target);
4237 switch (type) {
4238 case MC_TARGET_PAGE:
4239 page = target.page;
4240 if (isolate_lru_page(page))
4241 goto put;
4242 pc = lookup_page_cgroup(page);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004243 if (!mem_cgroup_move_account(pc,
4244 mc.from, mc.to, false)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004245 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004246 /* we uncharge from mc.from later. */
4247 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004248 }
4249 putback_lru_page(page);
4250put: /* is_target_pte_for_mc() gets the page */
4251 put_page(page);
4252 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004253 case MC_TARGET_SWAP:
4254 ent = target.ent;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004255 if (!mem_cgroup_move_swap_account(ent,
4256 mc.from, mc.to, false)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004257 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004258 /* we fixup refcnts and charges later. */
4259 mc.moved_swap++;
4260 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004261 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004262 default:
4263 break;
4264 }
4265 }
4266 pte_unmap_unlock(pte - 1, ptl);
4267 cond_resched();
4268
4269 if (addr != end) {
4270 /*
4271 * We have consumed all precharges we got in can_attach().
4272 * We try charge one by one, but don't do any additional
4273 * charges to mc.to if we have failed in charge once in attach()
4274 * phase.
4275 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004276 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004277 if (!ret)
4278 goto retry;
4279 }
4280
4281 return ret;
4282}
4283
4284static void mem_cgroup_move_charge(struct mm_struct *mm)
4285{
4286 struct vm_area_struct *vma;
4287
4288 lru_add_drain_all();
4289 down_read(&mm->mmap_sem);
4290 for (vma = mm->mmap; vma; vma = vma->vm_next) {
4291 int ret;
4292 struct mm_walk mem_cgroup_move_charge_walk = {
4293 .pmd_entry = mem_cgroup_move_charge_pte_range,
4294 .mm = mm,
4295 .private = vma,
4296 };
4297 if (is_vm_hugetlb_page(vma))
4298 continue;
4299 /* TODO: We don't move charges of shmem/tmpfs pages for now. */
4300 if (vma->vm_flags & VM_SHARED)
4301 continue;
4302 ret = walk_page_range(vma->vm_start, vma->vm_end,
4303 &mem_cgroup_move_charge_walk);
4304 if (ret)
4305 /*
4306 * means we have consumed all precharges and failed in
4307 * doing additional charge. Just abandon here.
4308 */
4309 break;
4310 }
4311 up_read(&mm->mmap_sem);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004312}
4313
Balbir Singh67e465a2008-02-07 00:13:54 -08004314static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4315 struct cgroup *cont,
4316 struct cgroup *old_cont,
Ben Blumbe367d02009-09-23 15:56:31 -07004317 struct task_struct *p,
4318 bool threadgroup)
Balbir Singh67e465a2008-02-07 00:13:54 -08004319{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004320 struct mm_struct *mm;
4321
4322 if (!mc.to)
4323 /* no need to move charge */
4324 return;
4325
4326 mm = get_task_mm(p);
4327 if (mm) {
4328 mem_cgroup_move_charge(mm);
4329 mmput(mm);
4330 }
4331 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08004332}
4333
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004334struct cgroup_subsys mem_cgroup_subsys = {
4335 .name = "memory",
4336 .subsys_id = mem_cgroup_subsys_id,
4337 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004338 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004339 .destroy = mem_cgroup_destroy,
4340 .populate = mem_cgroup_populate,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004341 .can_attach = mem_cgroup_can_attach,
4342 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08004343 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004344 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004345 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004346};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004347
4348#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4349
4350static int __init disable_swap_account(char *s)
4351{
4352 really_do_swap_account = 0;
4353 return 1;
4354}
4355__setup("noswapaccount", disable_swap_account);
4356#endif