blob: fd4529d86de59cbfc7093a433752e9b9b604dc93 [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 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -08009 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
18 */
19
20#include <linux/res_counter.h>
21#include <linux/memcontrol.h>
22#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080023#include <linux/mm.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080024#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080025#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080026#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080027#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080028#include <linux/bit_spinlock.h>
29#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070030#include <linux/limits.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080031#include <linux/mutex.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070032#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080033#include <linux/swap.h>
34#include <linux/spinlock.h>
35#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080036#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070037#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070038#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070039#include <linux/page_cgroup.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080040#include "internal.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080041
Balbir Singh8697d332008-02-07 00:13:59 -080042#include <asm/uaccess.h>
43
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070044struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070045#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080046
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080047#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
Li Zefan338c8432009-06-17 16:27:15 -070048/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080049int do_swap_account __read_mostly;
50static int really_do_swap_account __initdata = 1; /* for remember boot option*/
51#else
52#define do_swap_account (0)
53#endif
54
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -080055static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080056
Balbir Singh8cdea7c2008-02-07 00:13:50 -080057/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080058 * Statistics for memory cgroup.
59 */
60enum mem_cgroup_stat_index {
61 /*
62 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
63 */
64 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
Balbir Singhd69b0422009-06-17 16:26:34 -070065 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
66 MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */
Balaji Rao55e462b2008-05-01 04:35:12 -070067 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
68 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080069
70 MEM_CGROUP_STAT_NSTATS,
71};
72
73struct mem_cgroup_stat_cpu {
74 s64 count[MEM_CGROUP_STAT_NSTATS];
75} ____cacheline_aligned_in_smp;
76
77struct mem_cgroup_stat {
Jan Blunckc8dad2bb2009-01-07 18:07:53 -080078 struct mem_cgroup_stat_cpu cpustat[0];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080079};
80
81/*
82 * For accounting under irq disable, no need for increment preempt count.
83 */
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070084static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080085 enum mem_cgroup_stat_index idx, int val)
86{
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070087 stat->count[idx] += val;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080088}
89
90static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
91 enum mem_cgroup_stat_index idx)
92{
93 int cpu;
94 s64 ret = 0;
95 for_each_possible_cpu(cpu)
96 ret += stat->cpustat[cpu].count[idx];
97 return ret;
98}
99
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700100static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
101{
102 s64 ret;
103
104 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
105 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
106 return ret;
107}
108
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800109/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800110 * per-zone information in memory controller.
111 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800112struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800113 /*
114 * spin_lock to protect the per cgroup LRU
115 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700116 struct list_head lists[NR_LRU_LISTS];
117 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800118
119 struct zone_reclaim_stat reclaim_stat;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800120};
121/* Macro for accessing counter */
122#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
123
124struct mem_cgroup_per_node {
125 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
126};
127
128struct mem_cgroup_lru_info {
129 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
130};
131
132/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800133 * The memory controller data structure. The memory controller controls both
134 * page cache and RSS per cgroup. We would eventually like to provide
135 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
136 * to help the administrator determine what knobs to tune.
137 *
138 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800139 * we hit the water mark. May be even add a low water mark, such that
140 * no reclaim occurs from a cgroup at it's low water mark, this is
141 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800142 */
143struct mem_cgroup {
144 struct cgroup_subsys_state css;
145 /*
146 * the counter to account for memory usage
147 */
148 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800149 /*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800150 * the counter to account for mem+swap usage.
151 */
152 struct res_counter memsw;
153 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800154 * Per cgroup active and inactive list, similar to the
155 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800156 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800157 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800158
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800159 /*
160 protect against reclaim related member.
161 */
162 spinlock_t reclaim_param_lock;
163
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800164 int prev_priority; /* for recording reclaim priority */
Balbir Singh6d61ef42009-01-07 18:08:06 -0800165
166 /*
167 * While reclaiming in a hiearchy, we cache the last child we
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700168 * reclaimed from.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800169 */
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700170 int last_scanned_child;
Balbir Singh18f59ea2009-01-07 18:08:07 -0800171 /*
172 * Should the accounting and control be hierarchical, per subtree?
173 */
174 bool use_hierarchy;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800175 unsigned long last_oom_jiffies;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800176 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800177
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800178 unsigned int swappiness;
179
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700180 /* set when res.limit == memsw.limit */
181 bool memsw_is_minimum;
182
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800183 /*
Jan Blunckc8dad2bb2009-01-07 18:07:53 -0800184 * statistics. This must be placed at the end of memcg.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800185 */
186 struct mem_cgroup_stat stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800187};
188
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800189enum charge_type {
190 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
191 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700192 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700193 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800194 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700195 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700196 NR_CHARGE_TYPE,
197};
198
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700199/* only for here (for easy reading.) */
200#define PCGF_CACHE (1UL << PCG_CACHE)
201#define PCGF_USED (1UL << PCG_USED)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700202#define PCGF_LOCK (1UL << PCG_LOCK)
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700203static const unsigned long
204pcg_default_flags[NR_CHARGE_TYPE] = {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800205 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
206 PCGF_USED | PCGF_LOCK, /* Anon */
207 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700208 0, /* FORCE */
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800209};
210
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800211/* for encoding cft->private value on file */
212#define _MEM (0)
213#define _MEMSWAP (1)
214#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
215#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
216#define MEMFILE_ATTR(val) ((val) & 0xffff)
217
218static void mem_cgroup_get(struct mem_cgroup *mem);
219static void mem_cgroup_put(struct mem_cgroup *mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -0800220static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800221
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700222static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
223 struct page_cgroup *pc,
224 bool charge)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800225{
226 int val = (charge)? 1 : -1;
227 struct mem_cgroup_stat *stat = &mem->stat;
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700228 struct mem_cgroup_stat_cpu *cpustat;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800229 int cpu = get_cpu();
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800230
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800231 cpustat = &stat->cpustat[cpu];
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700232 if (PageCgroupCache(pc))
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700233 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800234 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700235 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
Balaji Rao55e462b2008-05-01 04:35:12 -0700236
237 if (charge)
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700238 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700239 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
240 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700241 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700242 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800243 put_cpu();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800244}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800245
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800246static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800247mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
248{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800249 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
250}
251
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800252static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800253page_cgroup_zoneinfo(struct page_cgroup *pc)
254{
255 struct mem_cgroup *mem = pc->mem_cgroup;
256 int nid = page_cgroup_nid(pc);
257 int zid = page_cgroup_zid(pc);
258
KOSAKI Motohiro54992762009-01-07 18:08:18 -0800259 if (!mem)
260 return NULL;
261
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800262 return mem_cgroup_zoneinfo(mem, nid, zid);
263}
264
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700265static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
Christoph Lameterb69408e2008-10-18 20:26:14 -0700266 enum lru_list idx)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800267{
268 int nid, zid;
269 struct mem_cgroup_per_zone *mz;
270 u64 total = 0;
271
272 for_each_online_node(nid)
273 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
274 mz = mem_cgroup_zoneinfo(mem, nid, zid);
275 total += MEM_CGROUP_ZSTAT(mz, idx);
276 }
277 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800278}
279
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800280static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800281{
282 return container_of(cgroup_subsys_state(cont,
283 mem_cgroup_subsys_id), struct mem_cgroup,
284 css);
285}
286
Balbir Singhcf475ad2008-04-29 01:00:16 -0700287struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800288{
Balbir Singh31a78f22008-09-28 23:09:31 +0100289 /*
290 * mm_update_next_owner() may clear mm->owner to NULL
291 * if it races with swapoff, page migration, etc.
292 * So this can be called with p == NULL.
293 */
294 if (unlikely(!p))
295 return NULL;
296
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800297 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
298 struct mem_cgroup, css);
299}
300
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800301static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
302{
303 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700304
305 if (!mm)
306 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800307 /*
308 * Because we have no locks, mm->owner's may be being moved to other
309 * cgroup. We use css_tryget() here even if this looks
310 * pessimistic (rather than adding locks here).
311 */
312 rcu_read_lock();
313 do {
314 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
315 if (unlikely(!mem))
316 break;
317 } while (!css_tryget(&mem->css));
318 rcu_read_unlock();
319 return mem;
320}
321
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700322/*
323 * Call callback function against all cgroup under hierarchy tree.
324 */
325static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
326 int (*func)(struct mem_cgroup *, void *))
327{
328 int found, ret, nextid;
329 struct cgroup_subsys_state *css;
330 struct mem_cgroup *mem;
331
332 if (!root->use_hierarchy)
333 return (*func)(root, data);
334
335 nextid = 1;
336 do {
337 ret = 0;
338 mem = NULL;
339
340 rcu_read_lock();
341 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
342 &found);
343 if (css && css_tryget(css))
344 mem = container_of(css, struct mem_cgroup, css);
345 rcu_read_unlock();
346
347 if (mem) {
348 ret = (*func)(mem, data);
349 css_put(&mem->css);
350 }
351 nextid = found + 1;
352 } while (!ret && css);
353
354 return ret;
355}
356
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800357/*
358 * Following LRU functions are allowed to be used without PCG_LOCK.
359 * Operations are called by routine of global LRU independently from memcg.
360 * What we have to take care of here is validness of pc->mem_cgroup.
361 *
362 * Changes to pc->mem_cgroup happens when
363 * 1. charge
364 * 2. moving account
365 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
366 * It is added to LRU before charge.
367 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
368 * When moving account, the page is not on LRU. It's isolated.
369 */
370
371void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800372{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800373 struct page_cgroup *pc;
374 struct mem_cgroup *mem;
375 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700376
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800377 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800378 return;
379 pc = lookup_page_cgroup(page);
380 /* can happen while we handle swapcache. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800381 if (list_empty(&pc->lru) || !pc->mem_cgroup)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800382 return;
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800383 /*
384 * We don't check PCG_USED bit. It's cleared when the "page" is finally
385 * removed from global LRU.
386 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800387 mz = page_cgroup_zoneinfo(pc);
388 mem = pc->mem_cgroup;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700389 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800390 list_del_init(&pc->lru);
391 return;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800392}
393
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800394void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800395{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800396 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800397}
398
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800399void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -0800400{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800401 struct mem_cgroup_per_zone *mz;
402 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800403
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800404 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700405 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700406
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800407 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800408 /*
409 * Used bit is set without atomic ops but after smp_wmb().
410 * For making pc->mem_cgroup visible, insert smp_rmb() here.
411 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800412 smp_rmb();
413 /* unused page is not rotated. */
414 if (!PageCgroupUsed(pc))
415 return;
416 mz = page_cgroup_zoneinfo(pc);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700417 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -0800418}
419
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800420void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
421{
422 struct page_cgroup *pc;
423 struct mem_cgroup_per_zone *mz;
424
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800425 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800426 return;
427 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800428 /*
429 * Used bit is set without atomic ops but after smp_wmb().
430 * For making pc->mem_cgroup visible, insert smp_rmb() here.
431 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800432 smp_rmb();
433 if (!PageCgroupUsed(pc))
434 return;
435
436 mz = page_cgroup_zoneinfo(pc);
437 MEM_CGROUP_ZSTAT(mz, lru) += 1;
438 list_add(&pc->lru, &mz->lists[lru]);
439}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800440
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800441/*
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800442 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
443 * lru because the page may.be reused after it's fully uncharged (because of
444 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
445 * it again. This function is only used to charge SwapCache. It's done under
446 * lock_page and expected that zone->lru_lock is never held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800447 */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800448static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800449{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800450 unsigned long flags;
451 struct zone *zone = page_zone(page);
452 struct page_cgroup *pc = lookup_page_cgroup(page);
453
454 spin_lock_irqsave(&zone->lru_lock, flags);
455 /*
456 * Forget old LRU when this page_cgroup is *not* used. This Used bit
457 * is guarded by lock_page() because the page is SwapCache.
458 */
459 if (!PageCgroupUsed(pc))
460 mem_cgroup_del_lru_list(page, page_lru(page));
461 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800462}
463
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800464static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
465{
466 unsigned long flags;
467 struct zone *zone = page_zone(page);
468 struct page_cgroup *pc = lookup_page_cgroup(page);
469
470 spin_lock_irqsave(&zone->lru_lock, flags);
471 /* link when the page is linked to LRU but page_cgroup isn't */
472 if (PageLRU(page) && list_empty(&pc->lru))
473 mem_cgroup_add_lru_list(page, page_lru(page));
474 spin_unlock_irqrestore(&zone->lru_lock, flags);
475}
476
477
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800478void mem_cgroup_move_lists(struct page *page,
479 enum lru_list from, enum lru_list to)
480{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800481 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800482 return;
483 mem_cgroup_del_lru_list(page, from);
484 mem_cgroup_add_lru_list(page, to);
485}
486
David Rientjes4c4a2212008-02-07 00:14:06 -0800487int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
488{
489 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700490 struct mem_cgroup *curr = NULL;
David Rientjes4c4a2212008-02-07 00:14:06 -0800491
492 task_lock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700493 rcu_read_lock();
494 curr = try_get_mem_cgroup_from_mm(task->mm);
495 rcu_read_unlock();
David Rientjes4c4a2212008-02-07 00:14:06 -0800496 task_unlock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700497 if (!curr)
498 return 0;
499 if (curr->use_hierarchy)
500 ret = css_is_ancestor(&curr->css, &mem->css);
501 else
502 ret = (curr == mem);
503 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -0800504 return ret;
505}
506
Balbir Singh66e17072008-02-07 00:13:56 -0800507/*
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800508 * prev_priority control...this will be used in memory reclaim path.
509 */
510int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
511{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800512 int prev_priority;
513
514 spin_lock(&mem->reclaim_param_lock);
515 prev_priority = mem->prev_priority;
516 spin_unlock(&mem->reclaim_param_lock);
517
518 return prev_priority;
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800519}
520
521void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
522{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800523 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800524 if (priority < mem->prev_priority)
525 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800526 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800527}
528
529void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
530{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800531 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800532 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800533 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800534}
535
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800536static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800537{
538 unsigned long active;
539 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800540 unsigned long gb;
541 unsigned long inactive_ratio;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800542
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700543 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
544 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800545
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800546 gb = (inactive + active) >> (30 - PAGE_SHIFT);
547 if (gb)
548 inactive_ratio = int_sqrt(10 * gb);
549 else
550 inactive_ratio = 1;
551
552 if (present_pages) {
553 present_pages[0] = inactive;
554 present_pages[1] = active;
555 }
556
557 return inactive_ratio;
558}
559
560int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
561{
562 unsigned long active;
563 unsigned long inactive;
564 unsigned long present_pages[2];
565 unsigned long inactive_ratio;
566
567 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
568
569 inactive = present_pages[0];
570 active = present_pages[1];
571
572 if (inactive * inactive_ratio < active)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800573 return 1;
574
575 return 0;
576}
577
Rik van Riel56e49d22009-06-16 15:32:28 -0700578int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
579{
580 unsigned long active;
581 unsigned long inactive;
582
583 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
584 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
585
586 return (active > inactive);
587}
588
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800589unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
590 struct zone *zone,
591 enum lru_list lru)
592{
593 int nid = zone->zone_pgdat->node_id;
594 int zid = zone_idx(zone);
595 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
596
597 return MEM_CGROUP_ZSTAT(mz, lru);
598}
599
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800600struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
601 struct zone *zone)
602{
603 int nid = zone->zone_pgdat->node_id;
604 int zid = zone_idx(zone);
605 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
606
607 return &mz->reclaim_stat;
608}
609
610struct zone_reclaim_stat *
611mem_cgroup_get_reclaim_stat_from_page(struct page *page)
612{
613 struct page_cgroup *pc;
614 struct mem_cgroup_per_zone *mz;
615
616 if (mem_cgroup_disabled())
617 return NULL;
618
619 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800620 /*
621 * Used bit is set without atomic ops but after smp_wmb().
622 * For making pc->mem_cgroup visible, insert smp_rmb() here.
623 */
624 smp_rmb();
625 if (!PageCgroupUsed(pc))
626 return NULL;
627
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800628 mz = page_cgroup_zoneinfo(pc);
629 if (!mz)
630 return NULL;
631
632 return &mz->reclaim_stat;
633}
634
Balbir Singh66e17072008-02-07 00:13:56 -0800635unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
636 struct list_head *dst,
637 unsigned long *scanned, int order,
638 int mode, struct zone *z,
639 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700640 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800641{
642 unsigned long nr_taken = 0;
643 struct page *page;
644 unsigned long scan;
645 LIST_HEAD(pc_list);
646 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800647 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800648 int nid = z->zone_pgdat->node_id;
649 int zid = zone_idx(z);
650 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700651 int lru = LRU_FILE * !!file + !!active;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700652 int ret;
Balbir Singh66e17072008-02-07 00:13:56 -0800653
Balbir Singhcf475ad2008-04-29 01:00:16 -0700654 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800655 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700656 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -0800657
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800658 scan = 0;
659 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800660 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800661 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800662
663 page = pc->page;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700664 if (unlikely(!PageCgroupUsed(pc)))
665 continue;
Hugh Dickins436c65412008-02-07 00:14:12 -0800666 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800667 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800668
Hugh Dickins436c65412008-02-07 00:14:12 -0800669 scan++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700670 ret = __isolate_lru_page(page, mode, file);
671 switch (ret) {
672 case 0:
Balbir Singh66e17072008-02-07 00:13:56 -0800673 list_move(&page->lru, dst);
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700674 mem_cgroup_del_lru(page);
Balbir Singh66e17072008-02-07 00:13:56 -0800675 nr_taken++;
KAMEZAWA Hiroyuki2ffebca2009-06-17 16:27:21 -0700676 break;
677 case -EBUSY:
678 /* we don't affect global LRU but rotate in our LRU */
679 mem_cgroup_rotate_lru_list(page, page_lru(page));
680 break;
681 default:
682 break;
Balbir Singh66e17072008-02-07 00:13:56 -0800683 }
684 }
685
Balbir Singh66e17072008-02-07 00:13:56 -0800686 *scanned = scan;
687 return nr_taken;
688}
689
Balbir Singh6d61ef42009-01-07 18:08:06 -0800690#define mem_cgroup_from_res_counter(counter, member) \
691 container_of(counter, struct mem_cgroup, member)
692
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800693static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
694{
695 if (do_swap_account) {
696 if (res_counter_check_under_limit(&mem->res) &&
697 res_counter_check_under_limit(&mem->memsw))
698 return true;
699 } else
700 if (res_counter_check_under_limit(&mem->res))
701 return true;
702 return false;
703}
704
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800705static unsigned int get_swappiness(struct mem_cgroup *memcg)
706{
707 struct cgroup *cgrp = memcg->css.cgroup;
708 unsigned int swappiness;
709
710 /* root ? */
711 if (cgrp->parent == NULL)
712 return vm_swappiness;
713
714 spin_lock(&memcg->reclaim_param_lock);
715 swappiness = memcg->swappiness;
716 spin_unlock(&memcg->reclaim_param_lock);
717
718 return swappiness;
719}
720
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700721static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
722{
723 int *val = data;
724 (*val)++;
725 return 0;
726}
Balbir Singhe2224322009-04-02 16:57:39 -0700727
728/**
729 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
730 * @memcg: The memory cgroup that went over limit
731 * @p: Task that is going to be killed
732 *
733 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
734 * enabled
735 */
736void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
737{
738 struct cgroup *task_cgrp;
739 struct cgroup *mem_cgrp;
740 /*
741 * Need a buffer in BSS, can't rely on allocations. The code relies
742 * on the assumption that OOM is serialized for memory controller.
743 * If this assumption is broken, revisit this code.
744 */
745 static char memcg_name[PATH_MAX];
746 int ret;
747
748 if (!memcg)
749 return;
750
751
752 rcu_read_lock();
753
754 mem_cgrp = memcg->css.cgroup;
755 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
756
757 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
758 if (ret < 0) {
759 /*
760 * Unfortunately, we are unable to convert to a useful name
761 * But we'll still print out the usage information
762 */
763 rcu_read_unlock();
764 goto done;
765 }
766 rcu_read_unlock();
767
768 printk(KERN_INFO "Task in %s killed", memcg_name);
769
770 rcu_read_lock();
771 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
772 if (ret < 0) {
773 rcu_read_unlock();
774 goto done;
775 }
776 rcu_read_unlock();
777
778 /*
779 * Continues from above, so we don't need an KERN_ level
780 */
781 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
782done:
783
784 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
785 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
786 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
787 res_counter_read_u64(&memcg->res, RES_FAILCNT));
788 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
789 "failcnt %llu\n",
790 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
791 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
792 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
793}
794
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700795/*
796 * This function returns the number of memcg under hierarchy tree. Returns
797 * 1(self count) if no children.
798 */
799static int mem_cgroup_count_children(struct mem_cgroup *mem)
800{
801 int num = 0;
802 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
803 return num;
804}
805
Balbir Singh6d61ef42009-01-07 18:08:06 -0800806/*
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700807 * Visit the first child (need not be the first child as per the ordering
808 * of the cgroup list, since we track last_scanned_child) of @mem and use
809 * that to reclaim free pages from.
810 */
811static struct mem_cgroup *
812mem_cgroup_select_victim(struct mem_cgroup *root_mem)
813{
814 struct mem_cgroup *ret = NULL;
815 struct cgroup_subsys_state *css;
816 int nextid, found;
817
818 if (!root_mem->use_hierarchy) {
819 css_get(&root_mem->css);
820 ret = root_mem;
821 }
822
823 while (!ret) {
824 rcu_read_lock();
825 nextid = root_mem->last_scanned_child + 1;
826 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
827 &found);
828 if (css && css_tryget(css))
829 ret = container_of(css, struct mem_cgroup, css);
830
831 rcu_read_unlock();
832 /* Updates scanning parameter */
833 spin_lock(&root_mem->reclaim_param_lock);
834 if (!css) {
835 /* this means start scan from ID:1 */
836 root_mem->last_scanned_child = 0;
837 } else
838 root_mem->last_scanned_child = found;
839 spin_unlock(&root_mem->reclaim_param_lock);
840 }
841
842 return ret;
843}
844
845/*
846 * Scan the hierarchy if needed to reclaim memory. We remember the last child
847 * we reclaimed from, so that we don't end up penalizing one child extensively
848 * based on its position in the children list.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800849 *
850 * root_mem is the original ancestor that we've been reclaim from.
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700851 *
852 * We give up and return to the caller when we visit root_mem twice.
853 * (other groups can be removed while we're walking....)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700854 *
855 * If shrink==true, for avoiding to free too much, this returns immedieately.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800856 */
857static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700858 gfp_t gfp_mask, bool noswap, bool shrink)
Balbir Singh6d61ef42009-01-07 18:08:06 -0800859{
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700860 struct mem_cgroup *victim;
861 int ret, total = 0;
862 int loop = 0;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800863
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -0700864 /* If memsw_is_minimum==1, swap-out is of-no-use. */
865 if (root_mem->memsw_is_minimum)
866 noswap = true;
867
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700868 while (loop < 2) {
869 victim = mem_cgroup_select_victim(root_mem);
870 if (victim == root_mem)
871 loop++;
872 if (!mem_cgroup_local_usage(&victim->stat)) {
873 /* this cgroup's local usage == 0 */
874 css_put(&victim->css);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800875 continue;
876 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700877 /* we use swappiness of local cgroup */
878 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
879 get_swappiness(victim));
880 css_put(&victim->css);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700881 /*
882 * At shrinking usage, we can't check we should stop here or
883 * reclaim more. It's depends on callers. last_scanned_child
884 * will work enough for keeping fairness under tree.
885 */
886 if (shrink)
887 return ret;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700888 total += ret;
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800889 if (mem_cgroup_check_under_limit(root_mem))
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700890 return 1 + total;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800891 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700892 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800893}
894
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800895bool mem_cgroup_oom_called(struct task_struct *task)
896{
897 bool ret = false;
898 struct mem_cgroup *mem;
899 struct mm_struct *mm;
900
901 rcu_read_lock();
902 mm = task->mm;
903 if (!mm)
904 mm = &init_mm;
905 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
906 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
907 ret = true;
908 rcu_read_unlock();
909 return ret;
910}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700911
912static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
913{
914 mem->last_oom_jiffies = jiffies;
915 return 0;
916}
917
918static void record_last_oom(struct mem_cgroup *mem)
919{
920 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
921}
922
Balbir Singhd69b0422009-06-17 16:26:34 -0700923/*
924 * Currently used to update mapped file statistics, but the routine can be
925 * generalized to update other statistics as well.
926 */
927void mem_cgroup_update_mapped_file_stat(struct page *page, int val)
928{
929 struct mem_cgroup *mem;
930 struct mem_cgroup_stat *stat;
931 struct mem_cgroup_stat_cpu *cpustat;
932 int cpu;
933 struct page_cgroup *pc;
934
935 if (!page_is_file_cache(page))
936 return;
937
938 pc = lookup_page_cgroup(page);
939 if (unlikely(!pc))
940 return;
941
942 lock_page_cgroup(pc);
943 mem = pc->mem_cgroup;
944 if (!mem)
945 goto done;
946
947 if (!PageCgroupUsed(pc))
948 goto done;
949
950 /*
951 * Preemption is already disabled, we don't need get_cpu()
952 */
953 cpu = smp_processor_id();
954 stat = &mem->stat;
955 cpustat = &stat->cpustat[cpu];
956
957 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val);
958done:
959 unlock_page_cgroup(pc);
960}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700961
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800962/*
963 * Unlike exported interface, "oom" parameter is added. if oom==true,
964 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800965 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800966static int __mem_cgroup_try_charge(struct mm_struct *mm,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800967 gfp_t gfp_mask, struct mem_cgroup **memcg,
968 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800969{
Balbir Singh6d61ef42009-01-07 18:08:06 -0800970 struct mem_cgroup *mem, *mem_over_limit;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800971 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800972 struct res_counter *fail_res;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800973
974 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
975 /* Don't account this! */
976 *memcg = NULL;
977 return 0;
978 }
979
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800980 /*
Hugh Dickins3be912772008-02-07 00:14:19 -0800981 * We always charge the cgroup the mm_struct belongs to.
982 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800983 * thread group leader migrates. It's possible that mm is not
984 * set, if so charge the init_mm (happens for pagecache usage).
985 */
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800986 mem = *memcg;
987 if (likely(!mem)) {
988 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800989 *memcg = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700990 } else {
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800991 css_get(&mem->css);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700992 }
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800993 if (unlikely(!mem))
994 return 0;
995
Nikanth Karthikesan46f7e602009-05-28 14:34:41 -0700996 VM_BUG_ON(css_is_removed(&mem->css));
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800997
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800998 while (1) {
999 int ret;
1000 bool noswap = false;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001001
Balbir Singh28dbc4b2009-01-07 18:08:05 -08001002 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001003 if (likely(!ret)) {
1004 if (!do_swap_account)
1005 break;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08001006 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
1007 &fail_res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001008 if (likely(!ret))
1009 break;
1010 /* mem+swap counter fails */
1011 res_counter_uncharge(&mem->res, PAGE_SIZE);
1012 noswap = true;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001013 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1014 memsw);
1015 } else
1016 /* mem counter fails */
1017 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1018 res);
1019
Hugh Dickins3be912772008-02-07 00:14:19 -08001020 if (!(gfp_mask & __GFP_WAIT))
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001021 goto nomem;
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001022
Balbir Singh6d61ef42009-01-07 18:08:06 -08001023 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001024 noswap, false);
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -08001025 if (ret)
1026 continue;
Balbir Singh66e17072008-02-07 00:13:56 -08001027
1028 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001029 * try_to_free_mem_cgroup_pages() might not give us a full
1030 * picture of reclaim. Some pages are reclaimed and might be
1031 * moved to swap cache or just unmapped from the cgroup.
1032 * Check the limit again to see if the reclaim reduced the
1033 * current usage of the cgroup before giving up
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001034 *
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001035 */
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -08001036 if (mem_cgroup_check_under_limit(mem_over_limit))
1037 continue;
Hugh Dickins3be912772008-02-07 00:14:19 -08001038
1039 if (!nr_retries--) {
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001040 if (oom) {
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08001041 mutex_lock(&memcg_tasklist);
KAMEZAWA Hiroyuki88700752009-01-07 18:08:09 -08001042 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08001043 mutex_unlock(&memcg_tasklist);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001044 record_last_oom(mem_over_limit);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001045 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001046 goto nomem;
Balbir Singh66e17072008-02-07 00:13:56 -08001047 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001048 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001049 return 0;
1050nomem:
1051 css_put(&mem->css);
1052 return -ENOMEM;
1053}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001054
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001055
1056/*
1057 * A helper function to get mem_cgroup from ID. must be called under
1058 * rcu_read_lock(). The caller must check css_is_removed() or some if
1059 * it's concern. (dropping refcnt from swap can be called against removed
1060 * memcg.)
1061 */
1062static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1063{
1064 struct cgroup_subsys_state *css;
1065
1066 /* ID 0 is unused ID */
1067 if (!id)
1068 return NULL;
1069 css = css_lookup(&mem_cgroup_subsys, id);
1070 if (!css)
1071 return NULL;
1072 return container_of(css, struct mem_cgroup, css);
1073}
1074
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001075static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1076{
1077 struct mem_cgroup *mem;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001078 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001079 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001080 swp_entry_t ent;
1081
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001082 VM_BUG_ON(!PageLocked(page));
1083
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001084 if (!PageSwapCache(page))
1085 return NULL;
1086
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001087 pc = lookup_page_cgroup(page);
Daisuke Nishimurac0bd3f632009-04-30 15:08:11 -07001088 lock_page_cgroup(pc);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001089 if (PageCgroupUsed(pc)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001090 mem = pc->mem_cgroup;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001091 if (mem && !css_tryget(&mem->css))
1092 mem = NULL;
1093 } else {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001094 ent.val = page_private(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001095 id = lookup_swap_cgroup(ent);
1096 rcu_read_lock();
1097 mem = mem_cgroup_lookup(id);
1098 if (mem && !css_tryget(&mem->css))
1099 mem = NULL;
1100 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001101 }
Daisuke Nishimurac0bd3f632009-04-30 15:08:11 -07001102 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001103 return mem;
1104}
1105
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001106/*
Daisuke Nishimuraa5e924f2009-01-07 18:08:28 -08001107 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001108 * USED state. If already USED, uncharge and return.
1109 */
1110
1111static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1112 struct page_cgroup *pc,
1113 enum charge_type ctype)
1114{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001115 /* try_charge() can return NULL to *memcg, taking care of it. */
1116 if (!mem)
1117 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001118
1119 lock_page_cgroup(pc);
1120 if (unlikely(PageCgroupUsed(pc))) {
1121 unlock_page_cgroup(pc);
1122 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001123 if (do_swap_account)
1124 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001125 css_put(&mem->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001126 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001127 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001128 pc->mem_cgroup = mem;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001129 smp_wmb();
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001130 pc->flags = pcg_default_flags[ctype];
Hugh Dickins3be912772008-02-07 00:14:19 -08001131
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001132 mem_cgroup_charge_statistics(mem, pc, true);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001133
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001134 unlock_page_cgroup(pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001135}
1136
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001137/**
1138 * mem_cgroup_move_account - move account of the page
1139 * @pc: page_cgroup of the page.
1140 * @from: mem_cgroup which the page is moved from.
1141 * @to: mem_cgroup which the page is moved to. @from != @to.
1142 *
1143 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001144 * - page is not on LRU (isolate_page() is useful.)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001145 *
1146 * returns 0 at success,
1147 * returns -EBUSY when lock is busy or "pc" is unstable.
1148 *
1149 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1150 * new cgroup. It should be done by a caller.
1151 */
1152
1153static int mem_cgroup_move_account(struct page_cgroup *pc,
1154 struct mem_cgroup *from, struct mem_cgroup *to)
1155{
1156 struct mem_cgroup_per_zone *from_mz, *to_mz;
1157 int nid, zid;
1158 int ret = -EBUSY;
Balbir Singhd69b0422009-06-17 16:26:34 -07001159 struct page *page;
1160 int cpu;
1161 struct mem_cgroup_stat *stat;
1162 struct mem_cgroup_stat_cpu *cpustat;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001163
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001164 VM_BUG_ON(from == to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001165 VM_BUG_ON(PageLRU(pc->page));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001166
1167 nid = page_cgroup_nid(pc);
1168 zid = page_cgroup_zid(pc);
1169 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1170 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1171
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001172 if (!trylock_page_cgroup(pc))
1173 return ret;
1174
1175 if (!PageCgroupUsed(pc))
1176 goto out;
1177
1178 if (pc->mem_cgroup != from)
1179 goto out;
1180
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001181 res_counter_uncharge(&from->res, PAGE_SIZE);
1182 mem_cgroup_charge_statistics(from, pc, false);
Balbir Singhd69b0422009-06-17 16:26:34 -07001183
1184 page = pc->page;
1185 if (page_is_file_cache(page) && page_mapped(page)) {
1186 cpu = smp_processor_id();
1187 /* Update mapped_file data for mem_cgroup "from" */
1188 stat = &from->stat;
1189 cpustat = &stat->cpustat[cpu];
1190 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1191 -1);
1192
1193 /* Update mapped_file data for mem_cgroup "to" */
1194 stat = &to->stat;
1195 cpustat = &stat->cpustat[cpu];
1196 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE,
1197 1);
1198 }
1199
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001200 if (do_swap_account)
1201 res_counter_uncharge(&from->memsw, PAGE_SIZE);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001202 css_put(&from->css);
1203
1204 css_get(&to->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001205 pc->mem_cgroup = to;
1206 mem_cgroup_charge_statistics(to, pc, true);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001207 ret = 0;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001208out:
1209 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001210 /*
1211 * We charges against "to" which may not have any tasks. Then, "to"
1212 * can be under rmdir(). But in current implementation, caller of
1213 * this function is just force_empty() and it's garanteed that
1214 * "to" is never removed. So, we don't check rmdir status here.
1215 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001216 return ret;
1217}
1218
1219/*
1220 * move charges to its parent.
1221 */
1222
1223static int mem_cgroup_move_parent(struct page_cgroup *pc,
1224 struct mem_cgroup *child,
1225 gfp_t gfp_mask)
1226{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001227 struct page *page = pc->page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001228 struct cgroup *cg = child->css.cgroup;
1229 struct cgroup *pcg = cg->parent;
1230 struct mem_cgroup *parent;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001231 int ret;
1232
1233 /* Is ROOT ? */
1234 if (!pcg)
1235 return -EINVAL;
1236
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001237
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001238 parent = mem_cgroup_from_cont(pcg);
1239
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001240
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001241 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001242 if (ret || !parent)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001243 return ret;
1244
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001245 if (!get_page_unless_zero(page)) {
1246 ret = -EBUSY;
1247 goto uncharge;
1248 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001249
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001250 ret = isolate_lru_page(page);
1251
1252 if (ret)
1253 goto cancel;
1254
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001255 ret = mem_cgroup_move_account(pc, child, parent);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001256
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001257 putback_lru_page(page);
1258 if (!ret) {
1259 put_page(page);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001260 /* drop extra refcnt by try_charge() */
1261 css_put(&parent->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001262 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001263 }
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001264
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001265cancel:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001266 put_page(page);
1267uncharge:
1268 /* drop extra refcnt by try_charge() */
1269 css_put(&parent->css);
1270 /* uncharge if move fails */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001271 res_counter_uncharge(&parent->res, PAGE_SIZE);
1272 if (do_swap_account)
1273 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001274 return ret;
1275}
1276
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001277/*
1278 * Charge the memory controller for page usage.
1279 * Return
1280 * 0 if the charge was successful
1281 * < 0 if the cgroup is over its limit
1282 */
1283static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1284 gfp_t gfp_mask, enum charge_type ctype,
1285 struct mem_cgroup *memcg)
1286{
1287 struct mem_cgroup *mem;
1288 struct page_cgroup *pc;
1289 int ret;
1290
1291 pc = lookup_page_cgroup(page);
1292 /* can happen at boot */
1293 if (unlikely(!pc))
1294 return 0;
1295 prefetchw(pc);
1296
1297 mem = memcg;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001298 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001299 if (ret || !mem)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001300 return ret;
1301
1302 __mem_cgroup_commit_charge(mem, pc, ctype);
1303 return 0;
1304}
1305
1306int mem_cgroup_newpage_charge(struct page *page,
1307 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001308{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001309 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001310 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001311 if (PageCompound(page))
1312 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001313 /*
1314 * If already mapped, we don't have to account.
1315 * If page cache, page->mapping has address_space.
1316 * But page->mapping may have out-of-use anon_vma pointer,
1317 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1318 * is NULL.
1319 */
1320 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1321 return 0;
1322 if (unlikely(!mm))
1323 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001324 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001325 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001326}
1327
Daisuke Nishimura83aae4c72009-04-02 16:57:48 -07001328static void
1329__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1330 enum charge_type ctype);
1331
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001332int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1333 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08001334{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001335 struct mem_cgroup *mem = NULL;
1336 int ret;
1337
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001338 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001339 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001340 if (PageCompound(page))
1341 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001342 /*
1343 * Corner case handling. This is called from add_to_page_cache()
1344 * in usual. But some FS (shmem) precharges this page before calling it
1345 * and call add_to_page_cache() with GFP_NOWAIT.
1346 *
1347 * For GFP_NOWAIT case, the page may be pre-charged before calling
1348 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1349 * charge twice. (It works but has to pay a bit larger cost.)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001350 * And when the page is SwapCache, it should take swap information
1351 * into account. This is under lock_page() now.
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001352 */
1353 if (!(gfp_mask & __GFP_WAIT)) {
1354 struct page_cgroup *pc;
1355
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001356
1357 pc = lookup_page_cgroup(page);
1358 if (!pc)
1359 return 0;
1360 lock_page_cgroup(pc);
1361 if (PageCgroupUsed(pc)) {
1362 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001363 return 0;
1364 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001365 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001366 }
1367
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001368 if (unlikely(!mm && !mem))
Balbir Singh8697d332008-02-07 00:13:59 -08001369 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001370
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001371 if (page_is_file_cache(page))
1372 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001373 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001374
Daisuke Nishimura83aae4c72009-04-02 16:57:48 -07001375 /* shmem */
1376 if (PageSwapCache(page)) {
1377 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1378 if (!ret)
1379 __mem_cgroup_commit_charge_swapin(page, mem,
1380 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1381 } else
1382 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1383 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001384
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001385 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001386}
1387
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001388/*
1389 * While swap-in, try_charge -> commit or cancel, the page is locked.
1390 * And when try_charge() successfully returns, one refcnt to memcg without
1391 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1392 * "commit()" or removed by "cancel()"
1393 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001394int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1395 struct page *page,
1396 gfp_t mask, struct mem_cgroup **ptr)
1397{
1398 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001399 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001400
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001401 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001402 return 0;
1403
1404 if (!do_swap_account)
1405 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001406 /*
1407 * A racing thread's fault, or swapoff, may have already updated
1408 * the pte, and even removed page from swap cache: return success
1409 * to go on to do_swap_page()'s pte_same() test, which should fail.
1410 */
1411 if (!PageSwapCache(page))
1412 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001413 mem = try_get_mem_cgroup_from_swapcache(page);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001414 if (!mem)
1415 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001416 *ptr = mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001417 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1418 /* drop extra refcnt from tryget */
1419 css_put(&mem->css);
1420 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001421charge_cur_mm:
1422 if (unlikely(!mm))
1423 mm = &init_mm;
1424 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1425}
1426
Daisuke Nishimura83aae4c72009-04-02 16:57:48 -07001427static void
1428__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1429 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001430{
1431 struct page_cgroup *pc;
1432
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001433 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001434 return;
1435 if (!ptr)
1436 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001437 cgroup_exclude_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001438 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001439 mem_cgroup_lru_del_before_commit_swapcache(page);
Daisuke Nishimura83aae4c72009-04-02 16:57:48 -07001440 __mem_cgroup_commit_charge(ptr, pc, ctype);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001441 mem_cgroup_lru_add_after_commit_swapcache(page);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001442 /*
1443 * Now swap is on-memory. This means this page may be
1444 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001445 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1446 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1447 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001448 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001449 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001450 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001451 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001452 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001453
1454 id = swap_cgroup_record(ent, 0);
1455 rcu_read_lock();
1456 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001457 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001458 /*
1459 * This recorded memcg can be obsolete one. So, avoid
1460 * calling css_tryget
1461 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001462 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1463 mem_cgroup_put(memcg);
1464 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001465 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001466 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001467 /*
1468 * At swapin, we may charge account against cgroup which has no tasks.
1469 * So, rmdir()->pre_destroy() can be called while we do this charge.
1470 * In that case, we need to call pre_destroy() again. check it here.
1471 */
1472 cgroup_release_and_wakeup_rmdir(&ptr->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001473}
1474
Daisuke Nishimura83aae4c72009-04-02 16:57:48 -07001475void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1476{
1477 __mem_cgroup_commit_charge_swapin(page, ptr,
1478 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1479}
1480
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001481void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1482{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001483 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001484 return;
1485 if (!mem)
1486 return;
1487 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001488 if (do_swap_account)
1489 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001490 css_put(&mem->css);
1491}
1492
1493
Balbir Singh8697d332008-02-07 00:13:59 -08001494/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001495 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001496 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001497static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001498__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001499{
Hugh Dickins8289546e2008-03-04 14:29:08 -08001500 struct page_cgroup *pc;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001501 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001502 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001503
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001504 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001505 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07001506
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001507 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001508 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001509
Balbir Singh8697d332008-02-07 00:13:59 -08001510 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08001511 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08001512 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001513 pc = lookup_page_cgroup(page);
1514 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001515 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001516
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001517 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001518
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001519 mem = pc->mem_cgroup;
1520
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001521 if (!PageCgroupUsed(pc))
1522 goto unlock_out;
1523
1524 switch (ctype) {
1525 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001526 case MEM_CGROUP_CHARGE_TYPE_DROP:
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001527 if (page_mapped(page))
1528 goto unlock_out;
1529 break;
1530 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1531 if (!PageAnon(page)) { /* Shared memory */
1532 if (page->mapping && !page_is_file_cache(page))
1533 goto unlock_out;
1534 } else if (page_mapped(page)) /* Anon */
1535 goto unlock_out;
1536 break;
1537 default:
1538 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001539 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001540
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001541 res_counter_uncharge(&mem->res, PAGE_SIZE);
1542 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1543 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001544 mem_cgroup_charge_statistics(mem, pc, false);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001545
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001546 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001547 /*
1548 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1549 * freed from LRU. This is safe because uncharged page is expected not
1550 * to be reused (freed soon). Exception is SwapCache, it's handled by
1551 * special functions.
1552 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001553
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001554 mz = page_cgroup_zoneinfo(pc);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001555 unlock_page_cgroup(pc);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001556
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001557 /* at swapout, this memcg will be accessed to record to swap */
1558 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1559 css_put(&mem->css);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001560
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001561 return mem;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001562
1563unlock_out:
1564 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001565 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08001566}
1567
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001568void mem_cgroup_uncharge_page(struct page *page)
1569{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001570 /* early check. */
1571 if (page_mapped(page))
1572 return;
1573 if (page->mapping && !PageAnon(page))
1574 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001575 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1576}
1577
1578void mem_cgroup_uncharge_cache_page(struct page *page)
1579{
1580 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07001581 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001582 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1583}
1584
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001585#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001586/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001587 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001588 * memcg information is recorded to swap_cgroup of "ent"
1589 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001590void
1591mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001592{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001593 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001594 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001595
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001596 if (!swapout) /* this was a swap cache but the swap is unused ! */
1597 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
1598
1599 memcg = __mem_cgroup_uncharge_common(page, ctype);
1600
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001601 /* record memcg information */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001602 if (do_swap_account && swapout && memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001603 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001604 mem_cgroup_get(memcg);
1605 }
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001606 if (swapout && memcg)
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001607 css_put(&memcg->css);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001608}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001609#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001610
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001611#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1612/*
1613 * called from swap_entry_free(). remove record in swap_cgroup and
1614 * uncharge "memsw" account.
1615 */
1616void mem_cgroup_uncharge_swap(swp_entry_t ent)
1617{
1618 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001619 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001620
1621 if (!do_swap_account)
1622 return;
1623
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001624 id = swap_cgroup_record(ent, 0);
1625 rcu_read_lock();
1626 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001627 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001628 /*
1629 * We uncharge this because swap is freed.
1630 * This memcg can be obsolete one. We avoid calling css_tryget
1631 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001632 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1633 mem_cgroup_put(memcg);
1634 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001635 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001636}
1637#endif
1638
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001639/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001640 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1641 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001642 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001643int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001644{
1645 struct page_cgroup *pc;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001646 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001647 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001648
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001649 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07001650 return 0;
1651
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001652 pc = lookup_page_cgroup(page);
1653 lock_page_cgroup(pc);
1654 if (PageCgroupUsed(pc)) {
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001655 mem = pc->mem_cgroup;
1656 css_get(&mem->css);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001657 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001658 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001659
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001660 if (mem) {
Daisuke Nishimura3bb4edf2009-01-07 18:08:28 -08001661 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001662 css_put(&mem->css);
1663 }
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001664 *ptr = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001665 return ret;
1666}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001667
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001668/* remove redundant charge if migration failed*/
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001669void mem_cgroup_end_migration(struct mem_cgroup *mem,
1670 struct page *oldpage, struct page *newpage)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001671{
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001672 struct page *target, *unused;
1673 struct page_cgroup *pc;
1674 enum charge_type ctype;
1675
1676 if (!mem)
1677 return;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001678 cgroup_exclude_rmdir(&mem->css);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001679 /* at migration success, oldpage->mapping is NULL. */
1680 if (oldpage->mapping) {
1681 target = oldpage;
1682 unused = NULL;
1683 } else {
1684 target = newpage;
1685 unused = oldpage;
1686 }
1687
1688 if (PageAnon(target))
1689 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1690 else if (page_is_file_cache(target))
1691 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1692 else
1693 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1694
1695 /* unused page is not on radix-tree now. */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001696 if (unused)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001697 __mem_cgroup_uncharge_common(unused, ctype);
1698
1699 pc = lookup_page_cgroup(target);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001700 /*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001701 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1702 * So, double-counting is effectively avoided.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001703 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001704 __mem_cgroup_commit_charge(mem, pc, ctype);
1705
1706 /*
1707 * Both of oldpage and newpage are still under lock_page().
1708 * Then, we don't have to care about race in radix-tree.
1709 * But we have to be careful that this page is unmapped or not.
1710 *
1711 * There is a case for !page_mapped(). At the start of
1712 * migration, oldpage was mapped. But now, it's zapped.
1713 * But we know *target* page is not freed/reused under us.
1714 * mem_cgroup_uncharge_page() does all necessary checks.
1715 */
1716 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1717 mem_cgroup_uncharge_page(target);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001718 /*
1719 * At migration, we may charge account against cgroup which has no tasks
1720 * So, rmdir()->pre_destroy() can be called while we do this charge.
1721 * In that case, we need to call pre_destroy() again. check it here.
1722 */
1723 cgroup_release_and_wakeup_rmdir(&mem->css);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001724}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001725
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001726/*
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001727 * A call to try to shrink memory usage on charge failure at shmem's swapin.
1728 * Calling hierarchical_reclaim is not enough because we should update
1729 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
1730 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
1731 * not from the memcg which this page would be charged to.
1732 * try_charge_swapin does all of these works properly.
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001733 */
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001734int mem_cgroup_shmem_charge_fallback(struct page *page,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001735 struct mm_struct *mm,
1736 gfp_t gfp_mask)
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001737{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001738 struct mem_cgroup *mem = NULL;
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001739 int ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001740
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001741 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001742 return 0;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001743
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001744 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1745 if (!ret)
1746 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001747
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001748 return ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001749}
1750
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001751static DEFINE_MUTEX(set_limit_mutex);
1752
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08001753static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001754 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001755{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001756 int retry_count;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001757 int progress;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001758 u64 memswlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001759 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001760 int children = mem_cgroup_count_children(memcg);
1761 u64 curusage, oldusage;
1762
1763 /*
1764 * For keeping hierarchical_reclaim simple, how long we should retry
1765 * is depends on callers. We set our retry-count to be function
1766 * of # of children which we should visit in this loop.
1767 */
1768 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1769
1770 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001771
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001772 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001773 if (signal_pending(current)) {
1774 ret = -EINTR;
1775 break;
1776 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001777 /*
1778 * Rather than hide all in some function, I do this in
1779 * open coded manner. You see what this really does.
1780 * We have to guarantee mem->res.limit < mem->memsw.limit.
1781 */
1782 mutex_lock(&set_limit_mutex);
1783 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1784 if (memswlimit < val) {
1785 ret = -EINVAL;
1786 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001787 break;
1788 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001789 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001790 if (!ret) {
1791 if (memswlimit == val)
1792 memcg->memsw_is_minimum = true;
1793 else
1794 memcg->memsw_is_minimum = false;
1795 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001796 mutex_unlock(&set_limit_mutex);
1797
1798 if (!ret)
1799 break;
1800
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001801 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001802 false, true);
1803 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1804 /* Usage is reduced ? */
1805 if (curusage >= oldusage)
1806 retry_count--;
1807 else
1808 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001809 }
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001810
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001811 return ret;
1812}
1813
Li Zefan338c8432009-06-17 16:27:15 -07001814static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1815 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001816{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001817 int retry_count;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001818 u64 memlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001819 int children = mem_cgroup_count_children(memcg);
1820 int ret = -EBUSY;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001821
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001822 /* see mem_cgroup_resize_res_limit */
1823 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1824 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001825 while (retry_count) {
1826 if (signal_pending(current)) {
1827 ret = -EINTR;
1828 break;
1829 }
1830 /*
1831 * Rather than hide all in some function, I do this in
1832 * open coded manner. You see what this really does.
1833 * We have to guarantee mem->res.limit < mem->memsw.limit.
1834 */
1835 mutex_lock(&set_limit_mutex);
1836 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1837 if (memlimit > val) {
1838 ret = -EINVAL;
1839 mutex_unlock(&set_limit_mutex);
1840 break;
1841 }
1842 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001843 if (!ret) {
1844 if (memlimit == val)
1845 memcg->memsw_is_minimum = true;
1846 else
1847 memcg->memsw_is_minimum = false;
1848 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001849 mutex_unlock(&set_limit_mutex);
1850
1851 if (!ret)
1852 break;
1853
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001854 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001855 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001856 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001857 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001858 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001859 else
1860 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001861 }
1862 return ret;
1863}
1864
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001865/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001866 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001867 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1868 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001869static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001870 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001871{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001872 struct zone *zone;
1873 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001874 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001875 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001876 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001877 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001878
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001879 zone = &NODE_DATA(node)->node_zones[zid];
1880 mz = mem_cgroup_zoneinfo(mem, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001881 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001882
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001883 loop = MEM_CGROUP_ZSTAT(mz, lru);
1884 /* give some margin against EBUSY etc...*/
1885 loop += 256;
1886 busy = NULL;
1887 while (loop--) {
1888 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001889 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001890 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001891 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001892 break;
1893 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001894 pc = list_entry(list->prev, struct page_cgroup, lru);
1895 if (busy == pc) {
1896 list_move(&pc->lru, list);
1897 busy = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001898 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001899 continue;
1900 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001901 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001902
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001903 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001904 if (ret == -ENOMEM)
1905 break;
1906
1907 if (ret == -EBUSY || ret == -EINVAL) {
1908 /* found lock contention or "pc" is obsolete. */
1909 busy = pc;
1910 cond_resched();
1911 } else
1912 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001913 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001914
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001915 if (!ret && !list_empty(list))
1916 return -EBUSY;
1917 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001918}
1919
1920/*
1921 * make mem_cgroup's charge to be 0 if there is no task.
1922 * This enables deleting this mem_cgroup.
1923 */
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001924static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001925{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001926 int ret;
1927 int node, zid, shrink;
1928 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001929 struct cgroup *cgrp = mem->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001930
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001931 css_get(&mem->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001932
1933 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001934 /* should free all ? */
1935 if (free_all)
1936 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001937move_account:
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001938 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001939 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001940 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001941 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001942 ret = -EINTR;
1943 if (signal_pending(current))
1944 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001945 /* This is for making all *used* pages to be on LRU. */
1946 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001947 ret = 0;
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08001948 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001949 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001950 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001951 for_each_lru(l) {
1952 ret = mem_cgroup_force_empty_list(mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001953 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001954 if (ret)
1955 break;
1956 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001957 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001958 if (ret)
1959 break;
1960 }
1961 /* it seems parent cgroup doesn't have enough mem */
1962 if (ret == -ENOMEM)
1963 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001964 cond_resched();
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001965 }
1966 ret = 0;
1967out:
1968 css_put(&mem->css);
1969 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001970
1971try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001972 /* returns EBUSY if there is a task or if we come here twice. */
1973 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001974 ret = -EBUSY;
1975 goto out;
1976 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001977 /* we call try-to-free pages for make this cgroup empty */
1978 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001979 /* try to free all pages in this cgroup */
1980 shrink = 1;
1981 while (nr_retries && mem->res.usage > 0) {
1982 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001983
1984 if (signal_pending(current)) {
1985 ret = -EINTR;
1986 goto out;
1987 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001988 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1989 false, get_swappiness(mem));
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001990 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001991 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001992 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001993 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001994 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001995
1996 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001997 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001998 /* try move_account...there may be some *locked* pages. */
1999 if (mem->res.usage)
2000 goto move_account;
2001 ret = 0;
2002 goto out;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002003}
2004
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002005int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
2006{
2007 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
2008}
2009
2010
Balbir Singh18f59ea2009-01-07 18:08:07 -08002011static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
2012{
2013 return mem_cgroup_from_cont(cont)->use_hierarchy;
2014}
2015
2016static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2017 u64 val)
2018{
2019 int retval = 0;
2020 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2021 struct cgroup *parent = cont->parent;
2022 struct mem_cgroup *parent_mem = NULL;
2023
2024 if (parent)
2025 parent_mem = mem_cgroup_from_cont(parent);
2026
2027 cgroup_lock();
2028 /*
2029 * If parent's use_hiearchy is set, we can't make any modifications
2030 * in the child subtrees. If it is unset, then the change can
2031 * occur, provided the current cgroup has no children.
2032 *
2033 * For the root cgroup, parent_mem is NULL, we allow value to be
2034 * set if there are no children.
2035 */
2036 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2037 (val == 1 || val == 0)) {
2038 if (list_empty(&cont->children))
2039 mem->use_hierarchy = val;
2040 else
2041 retval = -EBUSY;
2042 } else
2043 retval = -EINVAL;
2044 cgroup_unlock();
2045
2046 return retval;
2047}
2048
Paul Menage2c3daa72008-04-29 00:59:58 -07002049static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002050{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002051 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2052 u64 val = 0;
2053 int type, name;
2054
2055 type = MEMFILE_TYPE(cft->private);
2056 name = MEMFILE_ATTR(cft->private);
2057 switch (type) {
2058 case _MEM:
2059 val = res_counter_read_u64(&mem->res, name);
2060 break;
2061 case _MEMSWAP:
Li Zefan338c8432009-06-17 16:27:15 -07002062 val = res_counter_read_u64(&mem->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002063 break;
2064 default:
2065 BUG();
2066 break;
2067 }
2068 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002069}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002070/*
2071 * The user of this function is...
2072 * RES_LIMIT.
2073 */
Paul Menage856c13a2008-07-25 01:47:04 -07002074static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2075 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002076{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002077 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002078 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002079 unsigned long long val;
2080 int ret;
2081
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002082 type = MEMFILE_TYPE(cft->private);
2083 name = MEMFILE_ATTR(cft->private);
2084 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002085 case RES_LIMIT:
2086 /* This function does all necessary parse...reuse it */
2087 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002088 if (ret)
2089 break;
2090 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002091 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002092 else
2093 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002094 break;
2095 default:
2096 ret = -EINVAL; /* should be BUG() ? */
2097 break;
2098 }
2099 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002100}
2101
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08002102static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2103 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2104{
2105 struct cgroup *cgroup;
2106 unsigned long long min_limit, min_memsw_limit, tmp;
2107
2108 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2109 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2110 cgroup = memcg->css.cgroup;
2111 if (!memcg->use_hierarchy)
2112 goto out;
2113
2114 while (cgroup->parent) {
2115 cgroup = cgroup->parent;
2116 memcg = mem_cgroup_from_cont(cgroup);
2117 if (!memcg->use_hierarchy)
2118 break;
2119 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2120 min_limit = min(min_limit, tmp);
2121 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2122 min_memsw_limit = min(min_memsw_limit, tmp);
2123 }
2124out:
2125 *mem_limit = min_limit;
2126 *memsw_limit = min_memsw_limit;
2127 return;
2128}
2129
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002130static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002131{
2132 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002133 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002134
2135 mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002136 type = MEMFILE_TYPE(event);
2137 name = MEMFILE_ATTR(event);
2138 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002139 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002140 if (type == _MEM)
2141 res_counter_reset_max(&mem->res);
2142 else
2143 res_counter_reset_max(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002144 break;
2145 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002146 if (type == _MEM)
2147 res_counter_reset_failcnt(&mem->res);
2148 else
2149 res_counter_reset_failcnt(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002150 break;
2151 }
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07002152 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002153}
2154
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002155
2156/* For read statistics */
2157enum {
2158 MCS_CACHE,
2159 MCS_RSS,
Balbir Singhd69b0422009-06-17 16:26:34 -07002160 MCS_MAPPED_FILE,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002161 MCS_PGPGIN,
2162 MCS_PGPGOUT,
2163 MCS_INACTIVE_ANON,
2164 MCS_ACTIVE_ANON,
2165 MCS_INACTIVE_FILE,
2166 MCS_ACTIVE_FILE,
2167 MCS_UNEVICTABLE,
2168 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002169};
2170
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002171struct mcs_total_stat {
2172 s64 stat[NR_MCS_STAT];
2173};
2174
2175struct {
2176 char *local_name;
2177 char *total_name;
2178} memcg_stat_strings[NR_MCS_STAT] = {
2179 {"cache", "total_cache"},
2180 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07002181 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002182 {"pgpgin", "total_pgpgin"},
2183 {"pgpgout", "total_pgpgout"},
2184 {"inactive_anon", "total_inactive_anon"},
2185 {"active_anon", "total_active_anon"},
2186 {"inactive_file", "total_inactive_file"},
2187 {"active_file", "total_active_file"},
2188 {"unevictable", "total_unevictable"}
2189};
2190
2191
2192static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2193{
2194 struct mcs_total_stat *s = data;
2195 s64 val;
2196
2197 /* per cpu stat */
2198 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2199 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2200 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2201 s->stat[MCS_RSS] += val * PAGE_SIZE;
Balbir Singhd69b0422009-06-17 16:26:34 -07002202 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE);
2203 s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002204 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2205 s->stat[MCS_PGPGIN] += val;
2206 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2207 s->stat[MCS_PGPGOUT] += val;
2208
2209 /* per zone stat */
2210 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2211 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2212 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2213 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2214 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2215 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2216 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2217 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2218 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2219 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2220 return 0;
2221}
2222
2223static void
2224mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2225{
2226 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2227}
2228
Paul Menagec64745c2008-04-29 01:00:02 -07002229static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2230 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002231{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002232 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002233 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002234 int i;
2235
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002236 memset(&mystat, 0, sizeof(mystat));
2237 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002238
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002239 for (i = 0; i < NR_MCS_STAT; i++)
2240 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002241
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002242 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08002243 {
2244 unsigned long long limit, memsw_limit;
2245 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2246 cb->fill(cb, "hierarchical_memory_limit", limit);
2247 if (do_swap_account)
2248 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2249 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002250
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002251 memset(&mystat, 0, sizeof(mystat));
2252 mem_cgroup_get_total_stat(mem_cont, &mystat);
2253 for (i = 0; i < NR_MCS_STAT; i++)
2254 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2255
2256
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002257#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08002258 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002259
2260 {
2261 int nid, zid;
2262 struct mem_cgroup_per_zone *mz;
2263 unsigned long recent_rotated[2] = {0, 0};
2264 unsigned long recent_scanned[2] = {0, 0};
2265
2266 for_each_online_node(nid)
2267 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2268 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2269
2270 recent_rotated[0] +=
2271 mz->reclaim_stat.recent_rotated[0];
2272 recent_rotated[1] +=
2273 mz->reclaim_stat.recent_rotated[1];
2274 recent_scanned[0] +=
2275 mz->reclaim_stat.recent_scanned[0];
2276 recent_scanned[1] +=
2277 mz->reclaim_stat.recent_scanned[1];
2278 }
2279 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2280 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2281 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2282 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2283 }
2284#endif
2285
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002286 return 0;
2287}
2288
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002289static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2290{
2291 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2292
2293 return get_swappiness(memcg);
2294}
2295
2296static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2297 u64 val)
2298{
2299 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2300 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08002301
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002302 if (val > 100)
2303 return -EINVAL;
2304
2305 if (cgrp->parent == NULL)
2306 return -EINVAL;
2307
2308 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08002309
2310 cgroup_lock();
2311
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002312 /* If under hierarchy, only empty-root can set this value */
2313 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08002314 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2315 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002316 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08002317 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002318
2319 spin_lock(&memcg->reclaim_param_lock);
2320 memcg->swappiness = val;
2321 spin_unlock(&memcg->reclaim_param_lock);
2322
Li Zefan068b38c2009-01-15 13:51:26 -08002323 cgroup_unlock();
2324
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002325 return 0;
2326}
2327
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002328
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002329static struct cftype mem_cgroup_files[] = {
2330 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002331 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002332 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07002333 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002334 },
2335 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002336 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002337 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002338 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002339 .read_u64 = mem_cgroup_read,
2340 },
2341 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002342 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002343 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07002344 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07002345 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002346 },
2347 {
2348 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002349 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002350 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07002351 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002352 },
Balbir Singh8697d332008-02-07 00:13:59 -08002353 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002354 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07002355 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002356 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002357 {
2358 .name = "force_empty",
2359 .trigger = mem_cgroup_force_empty_write,
2360 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08002361 {
2362 .name = "use_hierarchy",
2363 .write_u64 = mem_cgroup_hierarchy_write,
2364 .read_u64 = mem_cgroup_hierarchy_read,
2365 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002366 {
2367 .name = "swappiness",
2368 .read_u64 = mem_cgroup_swappiness_read,
2369 .write_u64 = mem_cgroup_swappiness_write,
2370 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002371};
2372
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002373#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2374static struct cftype memsw_cgroup_files[] = {
2375 {
2376 .name = "memsw.usage_in_bytes",
2377 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2378 .read_u64 = mem_cgroup_read,
2379 },
2380 {
2381 .name = "memsw.max_usage_in_bytes",
2382 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2383 .trigger = mem_cgroup_reset,
2384 .read_u64 = mem_cgroup_read,
2385 },
2386 {
2387 .name = "memsw.limit_in_bytes",
2388 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2389 .write_string = mem_cgroup_write,
2390 .read_u64 = mem_cgroup_read,
2391 },
2392 {
2393 .name = "memsw.failcnt",
2394 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2395 .trigger = mem_cgroup_reset,
2396 .read_u64 = mem_cgroup_read,
2397 },
2398};
2399
2400static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2401{
2402 if (!do_swap_account)
2403 return 0;
2404 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2405 ARRAY_SIZE(memsw_cgroup_files));
2406};
2407#else
2408static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2409{
2410 return 0;
2411}
2412#endif
2413
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002414static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2415{
2416 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002417 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002418 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002419 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002420 /*
2421 * This routine is called against possible nodes.
2422 * But it's BUG to call kmalloc() against offline node.
2423 *
2424 * TODO: this routine can waste much memory for nodes which will
2425 * never be onlined. It's better to use memory hotplug callback
2426 * function.
2427 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002428 if (!node_state(node, N_NORMAL_MEMORY))
2429 tmp = -1;
2430 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002431 if (!pn)
2432 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002433
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002434 mem->info.nodeinfo[node] = pn;
2435 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002436
2437 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2438 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07002439 for_each_lru(l)
2440 INIT_LIST_HEAD(&mz->lists[l]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002441 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002442 return 0;
2443}
2444
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002445static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2446{
2447 kfree(mem->info.nodeinfo[node]);
2448}
2449
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002450static int mem_cgroup_size(void)
2451{
2452 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2453 return sizeof(struct mem_cgroup) + cpustat_size;
2454}
2455
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002456static struct mem_cgroup *mem_cgroup_alloc(void)
2457{
2458 struct mem_cgroup *mem;
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002459 int size = mem_cgroup_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002460
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002461 if (size < PAGE_SIZE)
2462 mem = kmalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002463 else
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002464 mem = vmalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002465
2466 if (mem)
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002467 memset(mem, 0, size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002468 return mem;
2469}
2470
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002471/*
2472 * At destroying mem_cgroup, references from swap_cgroup can remain.
2473 * (scanning all at force_empty is too costly...)
2474 *
2475 * Instead of clearing all references at force_empty, we remember
2476 * the number of reference from swap_cgroup and free mem_cgroup when
2477 * it goes down to 0.
2478 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002479 * Removal of cgroup itself succeeds regardless of refs from swap.
2480 */
2481
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002482static void __mem_cgroup_free(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002483{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002484 int node;
2485
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002486 free_css_id(&mem_cgroup_subsys, &mem->css);
2487
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002488 for_each_node_state(node, N_POSSIBLE)
2489 free_mem_cgroup_per_zone_info(mem, node);
2490
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002491 if (mem_cgroup_size() < PAGE_SIZE)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002492 kfree(mem);
2493 else
2494 vfree(mem);
2495}
2496
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002497static void mem_cgroup_get(struct mem_cgroup *mem)
2498{
2499 atomic_inc(&mem->refcnt);
2500}
2501
2502static void mem_cgroup_put(struct mem_cgroup *mem)
2503{
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002504 if (atomic_dec_and_test(&mem->refcnt)) {
2505 struct mem_cgroup *parent = parent_mem_cgroup(mem);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002506 __mem_cgroup_free(mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002507 if (parent)
2508 mem_cgroup_put(parent);
2509 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002510}
2511
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002512/*
2513 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2514 */
2515static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2516{
2517 if (!mem->res.parent)
2518 return NULL;
2519 return mem_cgroup_from_res_counter(mem->res.parent, res);
2520}
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002521
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002522#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2523static void __init enable_swap_cgroup(void)
2524{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002525 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002526 do_swap_account = 1;
2527}
2528#else
2529static void __init enable_swap_cgroup(void)
2530{
2531}
2532#endif
2533
Li Zefan0eb253e2009-01-15 13:51:25 -08002534static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002535mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2536{
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002537 struct mem_cgroup *mem, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002538 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002539 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002540
Jan Blunckc8dad2bb2009-01-07 18:07:53 -08002541 mem = mem_cgroup_alloc();
2542 if (!mem)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002543 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08002544
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002545 for_each_node_state(node, N_POSSIBLE)
2546 if (alloc_mem_cgroup_per_zone_info(mem, node))
2547 goto free_out;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002548 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002549 if (cont->parent == NULL) {
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002550 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002551 parent = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002552 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002553 parent = mem_cgroup_from_cont(cont->parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002554 mem->use_hierarchy = parent->use_hierarchy;
2555 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002556
Balbir Singh18f59ea2009-01-07 18:08:07 -08002557 if (parent && parent->use_hierarchy) {
2558 res_counter_init(&mem->res, &parent->res);
2559 res_counter_init(&mem->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002560 /*
2561 * We increment refcnt of the parent to ensure that we can
2562 * safely access it on res_counter_charge/uncharge.
2563 * This refcnt will be decremented when freeing this
2564 * mem_cgroup(see mem_cgroup_put).
2565 */
2566 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002567 } else {
2568 res_counter_init(&mem->res, NULL);
2569 res_counter_init(&mem->memsw, NULL);
2570 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002571 mem->last_scanned_child = 0;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -08002572 spin_lock_init(&mem->reclaim_param_lock);
Balbir Singh6d61ef42009-01-07 18:08:06 -08002573
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002574 if (parent)
2575 mem->swappiness = get_swappiness(parent);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002576 atomic_set(&mem->refcnt, 1);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002577 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002578free_out:
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002579 __mem_cgroup_free(mem);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002580 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002581}
2582
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002583static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002584 struct cgroup *cont)
2585{
2586 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002587
2588 return mem_cgroup_force_empty(mem, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002589}
2590
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002591static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2592 struct cgroup *cont)
2593{
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002594 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002595
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002596 mem_cgroup_put(mem);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002597}
2598
2599static int mem_cgroup_populate(struct cgroup_subsys *ss,
2600 struct cgroup *cont)
2601{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002602 int ret;
2603
2604 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2605 ARRAY_SIZE(mem_cgroup_files));
2606
2607 if (!ret)
2608 ret = register_memsw_files(cont, ss);
2609 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002610}
2611
Balbir Singh67e465a2008-02-07 00:13:54 -08002612static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2613 struct cgroup *cont,
2614 struct cgroup *old_cont,
2615 struct task_struct *p)
2616{
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002617 mutex_lock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002618 /*
Nikanth Karthikesanf9717d22009-01-07 18:08:11 -08002619 * FIXME: It's better to move charges of this process from old
2620 * memcg to new memcg. But it's just on TODO-List now.
Balbir Singh67e465a2008-02-07 00:13:54 -08002621 */
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002622 mutex_unlock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002623}
2624
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002625struct cgroup_subsys mem_cgroup_subsys = {
2626 .name = "memory",
2627 .subsys_id = mem_cgroup_subsys_id,
2628 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002629 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002630 .destroy = mem_cgroup_destroy,
2631 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08002632 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002633 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002634 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002635};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002636
2637#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2638
2639static int __init disable_swap_account(char *s)
2640{
2641 really_do_swap_account = 0;
2642 return 1;
2643}
2644__setup("noswapaccount", disable_swap_account);
2645#endif