blob: e2fa20dadf408e2d38e6b5a4a61ed4fb9febd3a1 [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 Blunckc8dad2b2009-01-07 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 Blunckc8dad2b2009-01-07 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 Dickins3be91272008-02-07 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 Dickins3be91272008-02-07 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 Dickins3be91272008-02-07 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 Nishimurac0bd3f62009-04-30 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 Nishimurac0bd3f62009-04-30 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 Dickins3be91272008-02-07 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);
1210 return ret;
1211}
1212
1213/*
1214 * move charges to its parent.
1215 */
1216
1217static int mem_cgroup_move_parent(struct page_cgroup *pc,
1218 struct mem_cgroup *child,
1219 gfp_t gfp_mask)
1220{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001221 struct page *page = pc->page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001222 struct cgroup *cg = child->css.cgroup;
1223 struct cgroup *pcg = cg->parent;
1224 struct mem_cgroup *parent;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001225 int ret;
1226
1227 /* Is ROOT ? */
1228 if (!pcg)
1229 return -EINVAL;
1230
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001231
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001232 parent = mem_cgroup_from_cont(pcg);
1233
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001234
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001235 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001236 if (ret || !parent)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001237 return ret;
1238
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001239 if (!get_page_unless_zero(page)) {
1240 ret = -EBUSY;
1241 goto uncharge;
1242 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001243
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001244 ret = isolate_lru_page(page);
1245
1246 if (ret)
1247 goto cancel;
1248
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001249 ret = mem_cgroup_move_account(pc, child, parent);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001250
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001251 putback_lru_page(page);
1252 if (!ret) {
1253 put_page(page);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001254 /* drop extra refcnt by try_charge() */
1255 css_put(&parent->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001256 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001257 }
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001258
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001259cancel:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001260 put_page(page);
1261uncharge:
1262 /* drop extra refcnt by try_charge() */
1263 css_put(&parent->css);
1264 /* uncharge if move fails */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001265 res_counter_uncharge(&parent->res, PAGE_SIZE);
1266 if (do_swap_account)
1267 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001268 return ret;
1269}
1270
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001271/*
1272 * Charge the memory controller for page usage.
1273 * Return
1274 * 0 if the charge was successful
1275 * < 0 if the cgroup is over its limit
1276 */
1277static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1278 gfp_t gfp_mask, enum charge_type ctype,
1279 struct mem_cgroup *memcg)
1280{
1281 struct mem_cgroup *mem;
1282 struct page_cgroup *pc;
1283 int ret;
1284
1285 pc = lookup_page_cgroup(page);
1286 /* can happen at boot */
1287 if (unlikely(!pc))
1288 return 0;
1289 prefetchw(pc);
1290
1291 mem = memcg;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001292 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001293 if (ret || !mem)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001294 return ret;
1295
1296 __mem_cgroup_commit_charge(mem, pc, ctype);
1297 return 0;
1298}
1299
1300int mem_cgroup_newpage_charge(struct page *page,
1301 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001302{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001303 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001304 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001305 if (PageCompound(page))
1306 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001307 /*
1308 * If already mapped, we don't have to account.
1309 * If page cache, page->mapping has address_space.
1310 * But page->mapping may have out-of-use anon_vma pointer,
1311 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1312 * is NULL.
1313 */
1314 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1315 return 0;
1316 if (unlikely(!mm))
1317 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001318 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001319 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001320}
1321
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001322static void
1323__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1324 enum charge_type ctype);
1325
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001326int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1327 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08001328{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001329 struct mem_cgroup *mem = NULL;
1330 int ret;
1331
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001332 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001333 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001334 if (PageCompound(page))
1335 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001336 /*
1337 * Corner case handling. This is called from add_to_page_cache()
1338 * in usual. But some FS (shmem) precharges this page before calling it
1339 * and call add_to_page_cache() with GFP_NOWAIT.
1340 *
1341 * For GFP_NOWAIT case, the page may be pre-charged before calling
1342 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1343 * charge twice. (It works but has to pay a bit larger cost.)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001344 * And when the page is SwapCache, it should take swap information
1345 * into account. This is under lock_page() now.
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001346 */
1347 if (!(gfp_mask & __GFP_WAIT)) {
1348 struct page_cgroup *pc;
1349
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001350
1351 pc = lookup_page_cgroup(page);
1352 if (!pc)
1353 return 0;
1354 lock_page_cgroup(pc);
1355 if (PageCgroupUsed(pc)) {
1356 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001357 return 0;
1358 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001359 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001360 }
1361
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001362 if (unlikely(!mm && !mem))
Balbir Singh8697d332008-02-07 00:13:59 -08001363 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001364
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001365 if (page_is_file_cache(page))
1366 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001367 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001368
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001369 /* shmem */
1370 if (PageSwapCache(page)) {
1371 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1372 if (!ret)
1373 __mem_cgroup_commit_charge_swapin(page, mem,
1374 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1375 } else
1376 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1377 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001378
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001379 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001380}
1381
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001382/*
1383 * While swap-in, try_charge -> commit or cancel, the page is locked.
1384 * And when try_charge() successfully returns, one refcnt to memcg without
1385 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1386 * "commit()" or removed by "cancel()"
1387 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001388int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1389 struct page *page,
1390 gfp_t mask, struct mem_cgroup **ptr)
1391{
1392 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001393 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001394
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001395 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001396 return 0;
1397
1398 if (!do_swap_account)
1399 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001400 /*
1401 * A racing thread's fault, or swapoff, may have already updated
1402 * the pte, and even removed page from swap cache: return success
1403 * to go on to do_swap_page()'s pte_same() test, which should fail.
1404 */
1405 if (!PageSwapCache(page))
1406 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001407 mem = try_get_mem_cgroup_from_swapcache(page);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001408 if (!mem)
1409 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001410 *ptr = mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001411 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1412 /* drop extra refcnt from tryget */
1413 css_put(&mem->css);
1414 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001415charge_cur_mm:
1416 if (unlikely(!mm))
1417 mm = &init_mm;
1418 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1419}
1420
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001421static void
1422__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1423 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001424{
1425 struct page_cgroup *pc;
1426
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001427 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001428 return;
1429 if (!ptr)
1430 return;
1431 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001432 mem_cgroup_lru_del_before_commit_swapcache(page);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001433 __mem_cgroup_commit_charge(ptr, pc, ctype);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001434 mem_cgroup_lru_add_after_commit_swapcache(page);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001435 /*
1436 * Now swap is on-memory. This means this page may be
1437 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001438 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1439 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1440 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001441 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001442 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001443 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001444 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001445 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001446
1447 id = swap_cgroup_record(ent, 0);
1448 rcu_read_lock();
1449 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001450 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001451 /*
1452 * This recorded memcg can be obsolete one. So, avoid
1453 * calling css_tryget
1454 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001455 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1456 mem_cgroup_put(memcg);
1457 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001458 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001459 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001460 /* add this page(page_cgroup) to the LRU we want. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001461
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001462}
1463
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001464void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1465{
1466 __mem_cgroup_commit_charge_swapin(page, ptr,
1467 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1468}
1469
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001470void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1471{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001472 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001473 return;
1474 if (!mem)
1475 return;
1476 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001477 if (do_swap_account)
1478 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001479 css_put(&mem->css);
1480}
1481
1482
Balbir Singh8697d332008-02-07 00:13:59 -08001483/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001484 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001485 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001486static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001487__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001488{
Hugh Dickins82895462008-03-04 14:29:08 -08001489 struct page_cgroup *pc;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001490 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001491 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001492
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001493 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001494 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07001495
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001496 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001497 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001498
Balbir Singh8697d332008-02-07 00:13:59 -08001499 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08001500 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08001501 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001502 pc = lookup_page_cgroup(page);
1503 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001504 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001505
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001506 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001507
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001508 mem = pc->mem_cgroup;
1509
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001510 if (!PageCgroupUsed(pc))
1511 goto unlock_out;
1512
1513 switch (ctype) {
1514 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001515 case MEM_CGROUP_CHARGE_TYPE_DROP:
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001516 if (page_mapped(page))
1517 goto unlock_out;
1518 break;
1519 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1520 if (!PageAnon(page)) { /* Shared memory */
1521 if (page->mapping && !page_is_file_cache(page))
1522 goto unlock_out;
1523 } else if (page_mapped(page)) /* Anon */
1524 goto unlock_out;
1525 break;
1526 default:
1527 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001528 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001529
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001530 res_counter_uncharge(&mem->res, PAGE_SIZE);
1531 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1532 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001533 mem_cgroup_charge_statistics(mem, pc, false);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001534
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001535 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001536 /*
1537 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1538 * freed from LRU. This is safe because uncharged page is expected not
1539 * to be reused (freed soon). Exception is SwapCache, it's handled by
1540 * special functions.
1541 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001542
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001543 mz = page_cgroup_zoneinfo(pc);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001544 unlock_page_cgroup(pc);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001545
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001546 /* at swapout, this memcg will be accessed to record to swap */
1547 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1548 css_put(&mem->css);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001549
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001550 return mem;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001551
1552unlock_out:
1553 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001554 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08001555}
1556
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001557void mem_cgroup_uncharge_page(struct page *page)
1558{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001559 /* early check. */
1560 if (page_mapped(page))
1561 return;
1562 if (page->mapping && !PageAnon(page))
1563 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001564 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1565}
1566
1567void mem_cgroup_uncharge_cache_page(struct page *page)
1568{
1569 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07001570 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001571 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1572}
1573
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001574#ifdef CONFIG_SWAP
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001575/*
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001576 * called after __delete_from_swap_cache() and drop "page" account.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001577 * memcg information is recorded to swap_cgroup of "ent"
1578 */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001579void
1580mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001581{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001582 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001583 int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001584
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001585 if (!swapout) /* this was a swap cache but the swap is unused ! */
1586 ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
1587
1588 memcg = __mem_cgroup_uncharge_common(page, ctype);
1589
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001590 /* record memcg information */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001591 if (do_swap_account && swapout && memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001592 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001593 mem_cgroup_get(memcg);
1594 }
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -07001595 if (swapout && memcg)
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001596 css_put(&memcg->css);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001597}
Daisuke Nishimurae767e052009-05-28 14:34:28 -07001598#endif
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001599
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001600#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1601/*
1602 * called from swap_entry_free(). remove record in swap_cgroup and
1603 * uncharge "memsw" account.
1604 */
1605void mem_cgroup_uncharge_swap(swp_entry_t ent)
1606{
1607 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001608 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001609
1610 if (!do_swap_account)
1611 return;
1612
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001613 id = swap_cgroup_record(ent, 0);
1614 rcu_read_lock();
1615 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001616 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001617 /*
1618 * We uncharge this because swap is freed.
1619 * This memcg can be obsolete one. We avoid calling css_tryget
1620 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001621 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1622 mem_cgroup_put(memcg);
1623 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001624 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001625}
1626#endif
1627
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001628/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001629 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1630 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001631 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001632int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001633{
1634 struct page_cgroup *pc;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001635 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001636 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001637
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001638 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07001639 return 0;
1640
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001641 pc = lookup_page_cgroup(page);
1642 lock_page_cgroup(pc);
1643 if (PageCgroupUsed(pc)) {
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001644 mem = pc->mem_cgroup;
1645 css_get(&mem->css);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001646 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001647 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001648
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001649 if (mem) {
Daisuke Nishimura3bb4edf2009-01-07 18:08:28 -08001650 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001651 css_put(&mem->css);
1652 }
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001653 *ptr = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001654 return ret;
1655}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001656
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001657/* remove redundant charge if migration failed*/
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001658void mem_cgroup_end_migration(struct mem_cgroup *mem,
1659 struct page *oldpage, struct page *newpage)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001660{
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001661 struct page *target, *unused;
1662 struct page_cgroup *pc;
1663 enum charge_type ctype;
1664
1665 if (!mem)
1666 return;
1667
1668 /* at migration success, oldpage->mapping is NULL. */
1669 if (oldpage->mapping) {
1670 target = oldpage;
1671 unused = NULL;
1672 } else {
1673 target = newpage;
1674 unused = oldpage;
1675 }
1676
1677 if (PageAnon(target))
1678 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1679 else if (page_is_file_cache(target))
1680 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1681 else
1682 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1683
1684 /* unused page is not on radix-tree now. */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001685 if (unused)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001686 __mem_cgroup_uncharge_common(unused, ctype);
1687
1688 pc = lookup_page_cgroup(target);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001689 /*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001690 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1691 * So, double-counting is effectively avoided.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001692 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001693 __mem_cgroup_commit_charge(mem, pc, ctype);
1694
1695 /*
1696 * Both of oldpage and newpage are still under lock_page().
1697 * Then, we don't have to care about race in radix-tree.
1698 * But we have to be careful that this page is unmapped or not.
1699 *
1700 * There is a case for !page_mapped(). At the start of
1701 * migration, oldpage was mapped. But now, it's zapped.
1702 * But we know *target* page is not freed/reused under us.
1703 * mem_cgroup_uncharge_page() does all necessary checks.
1704 */
1705 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1706 mem_cgroup_uncharge_page(target);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001707}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001708
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001709/*
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001710 * A call to try to shrink memory usage on charge failure at shmem's swapin.
1711 * Calling hierarchical_reclaim is not enough because we should update
1712 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
1713 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
1714 * not from the memcg which this page would be charged to.
1715 * try_charge_swapin does all of these works properly.
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001716 */
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001717int mem_cgroup_shmem_charge_fallback(struct page *page,
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001718 struct mm_struct *mm,
1719 gfp_t gfp_mask)
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001720{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001721 struct mem_cgroup *mem = NULL;
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001722 int ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001723
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001724 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001725 return 0;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001726
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001727 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1728 if (!ret)
1729 mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001730
Daisuke Nishimuraae3abae2009-04-30 15:08:19 -07001731 return ret;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001732}
1733
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001734static DEFINE_MUTEX(set_limit_mutex);
1735
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08001736static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001737 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001738{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001739 int retry_count;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001740 int progress;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001741 u64 memswlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001742 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001743 int children = mem_cgroup_count_children(memcg);
1744 u64 curusage, oldusage;
1745
1746 /*
1747 * For keeping hierarchical_reclaim simple, how long we should retry
1748 * is depends on callers. We set our retry-count to be function
1749 * of # of children which we should visit in this loop.
1750 */
1751 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1752
1753 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001754
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001755 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001756 if (signal_pending(current)) {
1757 ret = -EINTR;
1758 break;
1759 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001760 /*
1761 * Rather than hide all in some function, I do this in
1762 * open coded manner. You see what this really does.
1763 * We have to guarantee mem->res.limit < mem->memsw.limit.
1764 */
1765 mutex_lock(&set_limit_mutex);
1766 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1767 if (memswlimit < val) {
1768 ret = -EINVAL;
1769 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001770 break;
1771 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001772 ret = res_counter_set_limit(&memcg->res, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001773 if (!ret) {
1774 if (memswlimit == val)
1775 memcg->memsw_is_minimum = true;
1776 else
1777 memcg->memsw_is_minimum = false;
1778 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001779 mutex_unlock(&set_limit_mutex);
1780
1781 if (!ret)
1782 break;
1783
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001784 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001785 false, true);
1786 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1787 /* Usage is reduced ? */
1788 if (curusage >= oldusage)
1789 retry_count--;
1790 else
1791 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001792 }
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001793
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001794 return ret;
1795}
1796
Li Zefan338c8432009-06-17 16:27:15 -07001797static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1798 unsigned long long val)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001799{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001800 int retry_count;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001801 u64 memlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001802 int children = mem_cgroup_count_children(memcg);
1803 int ret = -EBUSY;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001804
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001805 /* see mem_cgroup_resize_res_limit */
1806 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1807 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001808 while (retry_count) {
1809 if (signal_pending(current)) {
1810 ret = -EINTR;
1811 break;
1812 }
1813 /*
1814 * Rather than hide all in some function, I do this in
1815 * open coded manner. You see what this really does.
1816 * We have to guarantee mem->res.limit < mem->memsw.limit.
1817 */
1818 mutex_lock(&set_limit_mutex);
1819 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1820 if (memlimit > val) {
1821 ret = -EINVAL;
1822 mutex_unlock(&set_limit_mutex);
1823 break;
1824 }
1825 ret = res_counter_set_limit(&memcg->memsw, val);
KAMEZAWA Hiroyuki22a668d2009-06-17 16:27:19 -07001826 if (!ret) {
1827 if (memlimit == val)
1828 memcg->memsw_is_minimum = true;
1829 else
1830 memcg->memsw_is_minimum = false;
1831 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001832 mutex_unlock(&set_limit_mutex);
1833
1834 if (!ret)
1835 break;
1836
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001837 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001838 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001839 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001840 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001841 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001842 else
1843 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001844 }
1845 return ret;
1846}
1847
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001848/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001849 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001850 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1851 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001852static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001853 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001854{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001855 struct zone *zone;
1856 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001857 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001858 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001859 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001860 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001861
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001862 zone = &NODE_DATA(node)->node_zones[zid];
1863 mz = mem_cgroup_zoneinfo(mem, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001864 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001865
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001866 loop = MEM_CGROUP_ZSTAT(mz, lru);
1867 /* give some margin against EBUSY etc...*/
1868 loop += 256;
1869 busy = NULL;
1870 while (loop--) {
1871 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001872 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001873 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001874 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001875 break;
1876 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001877 pc = list_entry(list->prev, struct page_cgroup, lru);
1878 if (busy == pc) {
1879 list_move(&pc->lru, list);
1880 busy = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001881 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001882 continue;
1883 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001884 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001885
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001886 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001887 if (ret == -ENOMEM)
1888 break;
1889
1890 if (ret == -EBUSY || ret == -EINVAL) {
1891 /* found lock contention or "pc" is obsolete. */
1892 busy = pc;
1893 cond_resched();
1894 } else
1895 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001896 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001897
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001898 if (!ret && !list_empty(list))
1899 return -EBUSY;
1900 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001901}
1902
1903/*
1904 * make mem_cgroup's charge to be 0 if there is no task.
1905 * This enables deleting this mem_cgroup.
1906 */
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001907static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001908{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001909 int ret;
1910 int node, zid, shrink;
1911 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001912 struct cgroup *cgrp = mem->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001913
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001914 css_get(&mem->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001915
1916 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001917 /* should free all ? */
1918 if (free_all)
1919 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001920move_account:
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001921 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001922 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001923 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001924 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001925 ret = -EINTR;
1926 if (signal_pending(current))
1927 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001928 /* This is for making all *used* pages to be on LRU. */
1929 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001930 ret = 0;
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08001931 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001932 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001933 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001934 for_each_lru(l) {
1935 ret = mem_cgroup_force_empty_list(mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001936 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001937 if (ret)
1938 break;
1939 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001940 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001941 if (ret)
1942 break;
1943 }
1944 /* it seems parent cgroup doesn't have enough mem */
1945 if (ret == -ENOMEM)
1946 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001947 cond_resched();
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001948 }
1949 ret = 0;
1950out:
1951 css_put(&mem->css);
1952 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001953
1954try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001955 /* returns EBUSY if there is a task or if we come here twice. */
1956 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001957 ret = -EBUSY;
1958 goto out;
1959 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001960 /* we call try-to-free pages for make this cgroup empty */
1961 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001962 /* try to free all pages in this cgroup */
1963 shrink = 1;
1964 while (nr_retries && mem->res.usage > 0) {
1965 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001966
1967 if (signal_pending(current)) {
1968 ret = -EINTR;
1969 goto out;
1970 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001971 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1972 false, get_swappiness(mem));
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001973 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001974 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001975 /* maybe some writeback is necessary */
1976 congestion_wait(WRITE, HZ/10);
1977 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001978
1979 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001980 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001981 /* try move_account...there may be some *locked* pages. */
1982 if (mem->res.usage)
1983 goto move_account;
1984 ret = 0;
1985 goto out;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001986}
1987
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001988int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1989{
1990 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1991}
1992
1993
Balbir Singh18f59ea2009-01-07 18:08:07 -08001994static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1995{
1996 return mem_cgroup_from_cont(cont)->use_hierarchy;
1997}
1998
1999static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
2000 u64 val)
2001{
2002 int retval = 0;
2003 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2004 struct cgroup *parent = cont->parent;
2005 struct mem_cgroup *parent_mem = NULL;
2006
2007 if (parent)
2008 parent_mem = mem_cgroup_from_cont(parent);
2009
2010 cgroup_lock();
2011 /*
2012 * If parent's use_hiearchy is set, we can't make any modifications
2013 * in the child subtrees. If it is unset, then the change can
2014 * occur, provided the current cgroup has no children.
2015 *
2016 * For the root cgroup, parent_mem is NULL, we allow value to be
2017 * set if there are no children.
2018 */
2019 if ((!parent_mem || !parent_mem->use_hierarchy) &&
2020 (val == 1 || val == 0)) {
2021 if (list_empty(&cont->children))
2022 mem->use_hierarchy = val;
2023 else
2024 retval = -EBUSY;
2025 } else
2026 retval = -EINVAL;
2027 cgroup_unlock();
2028
2029 return retval;
2030}
2031
Paul Menage2c3daa72008-04-29 00:59:58 -07002032static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002033{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002034 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2035 u64 val = 0;
2036 int type, name;
2037
2038 type = MEMFILE_TYPE(cft->private);
2039 name = MEMFILE_ATTR(cft->private);
2040 switch (type) {
2041 case _MEM:
2042 val = res_counter_read_u64(&mem->res, name);
2043 break;
2044 case _MEMSWAP:
Li Zefan338c8432009-06-17 16:27:15 -07002045 val = res_counter_read_u64(&mem->memsw, name);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002046 break;
2047 default:
2048 BUG();
2049 break;
2050 }
2051 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002052}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002053/*
2054 * The user of this function is...
2055 * RES_LIMIT.
2056 */
Paul Menage856c13a2008-07-25 01:47:04 -07002057static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
2058 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002059{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002060 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002061 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002062 unsigned long long val;
2063 int ret;
2064
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002065 type = MEMFILE_TYPE(cft->private);
2066 name = MEMFILE_ATTR(cft->private);
2067 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002068 case RES_LIMIT:
2069 /* This function does all necessary parse...reuse it */
2070 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002071 if (ret)
2072 break;
2073 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002074 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002075 else
2076 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002077 break;
2078 default:
2079 ret = -EINVAL; /* should be BUG() ? */
2080 break;
2081 }
2082 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002083}
2084
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08002085static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
2086 unsigned long long *mem_limit, unsigned long long *memsw_limit)
2087{
2088 struct cgroup *cgroup;
2089 unsigned long long min_limit, min_memsw_limit, tmp;
2090
2091 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2092 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2093 cgroup = memcg->css.cgroup;
2094 if (!memcg->use_hierarchy)
2095 goto out;
2096
2097 while (cgroup->parent) {
2098 cgroup = cgroup->parent;
2099 memcg = mem_cgroup_from_cont(cgroup);
2100 if (!memcg->use_hierarchy)
2101 break;
2102 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2103 min_limit = min(min_limit, tmp);
2104 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2105 min_memsw_limit = min(min_memsw_limit, tmp);
2106 }
2107out:
2108 *mem_limit = min_limit;
2109 *memsw_limit = min_memsw_limit;
2110 return;
2111}
2112
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002113static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002114{
2115 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002116 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002117
2118 mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002119 type = MEMFILE_TYPE(event);
2120 name = MEMFILE_ATTR(event);
2121 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002122 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002123 if (type == _MEM)
2124 res_counter_reset_max(&mem->res);
2125 else
2126 res_counter_reset_max(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002127 break;
2128 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002129 if (type == _MEM)
2130 res_counter_reset_failcnt(&mem->res);
2131 else
2132 res_counter_reset_failcnt(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002133 break;
2134 }
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07002135 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002136}
2137
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002138
2139/* For read statistics */
2140enum {
2141 MCS_CACHE,
2142 MCS_RSS,
Balbir Singhd69b0422009-06-17 16:26:34 -07002143 MCS_MAPPED_FILE,
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002144 MCS_PGPGIN,
2145 MCS_PGPGOUT,
2146 MCS_INACTIVE_ANON,
2147 MCS_ACTIVE_ANON,
2148 MCS_INACTIVE_FILE,
2149 MCS_ACTIVE_FILE,
2150 MCS_UNEVICTABLE,
2151 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002152};
2153
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002154struct mcs_total_stat {
2155 s64 stat[NR_MCS_STAT];
2156};
2157
2158struct {
2159 char *local_name;
2160 char *total_name;
2161} memcg_stat_strings[NR_MCS_STAT] = {
2162 {"cache", "total_cache"},
2163 {"rss", "total_rss"},
Balbir Singhd69b0422009-06-17 16:26:34 -07002164 {"mapped_file", "total_mapped_file"},
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002165 {"pgpgin", "total_pgpgin"},
2166 {"pgpgout", "total_pgpgout"},
2167 {"inactive_anon", "total_inactive_anon"},
2168 {"active_anon", "total_active_anon"},
2169 {"inactive_file", "total_inactive_file"},
2170 {"active_file", "total_active_file"},
2171 {"unevictable", "total_unevictable"}
2172};
2173
2174
2175static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2176{
2177 struct mcs_total_stat *s = data;
2178 s64 val;
2179
2180 /* per cpu stat */
2181 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2182 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2183 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2184 s->stat[MCS_RSS] += val * PAGE_SIZE;
Balbir Singhd69b0422009-06-17 16:26:34 -07002185 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE);
2186 s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE;
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002187 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2188 s->stat[MCS_PGPGIN] += val;
2189 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2190 s->stat[MCS_PGPGOUT] += val;
2191
2192 /* per zone stat */
2193 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2194 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2195 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2196 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2197 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2198 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2199 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2200 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2201 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2202 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2203 return 0;
2204}
2205
2206static void
2207mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2208{
2209 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2210}
2211
Paul Menagec64745c2008-04-29 01:00:02 -07002212static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2213 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002214{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002215 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002216 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002217 int i;
2218
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002219 memset(&mystat, 0, sizeof(mystat));
2220 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002221
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002222 for (i = 0; i < NR_MCS_STAT; i++)
2223 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002224
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002225 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08002226 {
2227 unsigned long long limit, memsw_limit;
2228 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2229 cb->fill(cb, "hierarchical_memory_limit", limit);
2230 if (do_swap_account)
2231 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2232 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002233
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002234 memset(&mystat, 0, sizeof(mystat));
2235 mem_cgroup_get_total_stat(mem_cont, &mystat);
2236 for (i = 0; i < NR_MCS_STAT; i++)
2237 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2238
2239
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002240#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08002241 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002242
2243 {
2244 int nid, zid;
2245 struct mem_cgroup_per_zone *mz;
2246 unsigned long recent_rotated[2] = {0, 0};
2247 unsigned long recent_scanned[2] = {0, 0};
2248
2249 for_each_online_node(nid)
2250 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2251 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2252
2253 recent_rotated[0] +=
2254 mz->reclaim_stat.recent_rotated[0];
2255 recent_rotated[1] +=
2256 mz->reclaim_stat.recent_rotated[1];
2257 recent_scanned[0] +=
2258 mz->reclaim_stat.recent_scanned[0];
2259 recent_scanned[1] +=
2260 mz->reclaim_stat.recent_scanned[1];
2261 }
2262 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2263 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2264 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2265 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2266 }
2267#endif
2268
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002269 return 0;
2270}
2271
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002272static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2273{
2274 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2275
2276 return get_swappiness(memcg);
2277}
2278
2279static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2280 u64 val)
2281{
2282 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2283 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08002284
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002285 if (val > 100)
2286 return -EINVAL;
2287
2288 if (cgrp->parent == NULL)
2289 return -EINVAL;
2290
2291 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08002292
2293 cgroup_lock();
2294
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002295 /* If under hierarchy, only empty-root can set this value */
2296 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08002297 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2298 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002299 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08002300 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002301
2302 spin_lock(&memcg->reclaim_param_lock);
2303 memcg->swappiness = val;
2304 spin_unlock(&memcg->reclaim_param_lock);
2305
Li Zefan068b38c2009-01-15 13:51:26 -08002306 cgroup_unlock();
2307
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002308 return 0;
2309}
2310
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002311
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002312static struct cftype mem_cgroup_files[] = {
2313 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002314 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002315 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07002316 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002317 },
2318 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002319 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002320 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002321 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002322 .read_u64 = mem_cgroup_read,
2323 },
2324 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002325 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002326 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07002327 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07002328 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002329 },
2330 {
2331 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002332 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002333 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07002334 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002335 },
Balbir Singh8697d332008-02-07 00:13:59 -08002336 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002337 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07002338 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002339 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002340 {
2341 .name = "force_empty",
2342 .trigger = mem_cgroup_force_empty_write,
2343 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08002344 {
2345 .name = "use_hierarchy",
2346 .write_u64 = mem_cgroup_hierarchy_write,
2347 .read_u64 = mem_cgroup_hierarchy_read,
2348 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002349 {
2350 .name = "swappiness",
2351 .read_u64 = mem_cgroup_swappiness_read,
2352 .write_u64 = mem_cgroup_swappiness_write,
2353 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002354};
2355
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002356#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2357static struct cftype memsw_cgroup_files[] = {
2358 {
2359 .name = "memsw.usage_in_bytes",
2360 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2361 .read_u64 = mem_cgroup_read,
2362 },
2363 {
2364 .name = "memsw.max_usage_in_bytes",
2365 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2366 .trigger = mem_cgroup_reset,
2367 .read_u64 = mem_cgroup_read,
2368 },
2369 {
2370 .name = "memsw.limit_in_bytes",
2371 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2372 .write_string = mem_cgroup_write,
2373 .read_u64 = mem_cgroup_read,
2374 },
2375 {
2376 .name = "memsw.failcnt",
2377 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2378 .trigger = mem_cgroup_reset,
2379 .read_u64 = mem_cgroup_read,
2380 },
2381};
2382
2383static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2384{
2385 if (!do_swap_account)
2386 return 0;
2387 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2388 ARRAY_SIZE(memsw_cgroup_files));
2389};
2390#else
2391static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2392{
2393 return 0;
2394}
2395#endif
2396
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002397static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2398{
2399 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002400 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002401 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002402 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002403 /*
2404 * This routine is called against possible nodes.
2405 * But it's BUG to call kmalloc() against offline node.
2406 *
2407 * TODO: this routine can waste much memory for nodes which will
2408 * never be onlined. It's better to use memory hotplug callback
2409 * function.
2410 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002411 if (!node_state(node, N_NORMAL_MEMORY))
2412 tmp = -1;
2413 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002414 if (!pn)
2415 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002416
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002417 mem->info.nodeinfo[node] = pn;
2418 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002419
2420 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2421 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07002422 for_each_lru(l)
2423 INIT_LIST_HEAD(&mz->lists[l]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002424 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002425 return 0;
2426}
2427
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002428static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2429{
2430 kfree(mem->info.nodeinfo[node]);
2431}
2432
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002433static int mem_cgroup_size(void)
2434{
2435 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2436 return sizeof(struct mem_cgroup) + cpustat_size;
2437}
2438
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002439static struct mem_cgroup *mem_cgroup_alloc(void)
2440{
2441 struct mem_cgroup *mem;
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002442 int size = mem_cgroup_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002443
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002444 if (size < PAGE_SIZE)
2445 mem = kmalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002446 else
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002447 mem = vmalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002448
2449 if (mem)
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002450 memset(mem, 0, size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002451 return mem;
2452}
2453
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002454/*
2455 * At destroying mem_cgroup, references from swap_cgroup can remain.
2456 * (scanning all at force_empty is too costly...)
2457 *
2458 * Instead of clearing all references at force_empty, we remember
2459 * the number of reference from swap_cgroup and free mem_cgroup when
2460 * it goes down to 0.
2461 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002462 * Removal of cgroup itself succeeds regardless of refs from swap.
2463 */
2464
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002465static void __mem_cgroup_free(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002466{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002467 int node;
2468
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002469 free_css_id(&mem_cgroup_subsys, &mem->css);
2470
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002471 for_each_node_state(node, N_POSSIBLE)
2472 free_mem_cgroup_per_zone_info(mem, node);
2473
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002474 if (mem_cgroup_size() < PAGE_SIZE)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002475 kfree(mem);
2476 else
2477 vfree(mem);
2478}
2479
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002480static void mem_cgroup_get(struct mem_cgroup *mem)
2481{
2482 atomic_inc(&mem->refcnt);
2483}
2484
2485static void mem_cgroup_put(struct mem_cgroup *mem)
2486{
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002487 if (atomic_dec_and_test(&mem->refcnt)) {
2488 struct mem_cgroup *parent = parent_mem_cgroup(mem);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002489 __mem_cgroup_free(mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002490 if (parent)
2491 mem_cgroup_put(parent);
2492 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002493}
2494
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002495/*
2496 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2497 */
2498static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2499{
2500 if (!mem->res.parent)
2501 return NULL;
2502 return mem_cgroup_from_res_counter(mem->res.parent, res);
2503}
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002504
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002505#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2506static void __init enable_swap_cgroup(void)
2507{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002508 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002509 do_swap_account = 1;
2510}
2511#else
2512static void __init enable_swap_cgroup(void)
2513{
2514}
2515#endif
2516
Li Zefan0eb253e2009-01-15 13:51:25 -08002517static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002518mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2519{
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002520 struct mem_cgroup *mem, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002521 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002522 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002523
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002524 mem = mem_cgroup_alloc();
2525 if (!mem)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002526 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08002527
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002528 for_each_node_state(node, N_POSSIBLE)
2529 if (alloc_mem_cgroup_per_zone_info(mem, node))
2530 goto free_out;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002531 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002532 if (cont->parent == NULL) {
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002533 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002534 parent = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002535 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002536 parent = mem_cgroup_from_cont(cont->parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002537 mem->use_hierarchy = parent->use_hierarchy;
2538 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002539
Balbir Singh18f59ea2009-01-07 18:08:07 -08002540 if (parent && parent->use_hierarchy) {
2541 res_counter_init(&mem->res, &parent->res);
2542 res_counter_init(&mem->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002543 /*
2544 * We increment refcnt of the parent to ensure that we can
2545 * safely access it on res_counter_charge/uncharge.
2546 * This refcnt will be decremented when freeing this
2547 * mem_cgroup(see mem_cgroup_put).
2548 */
2549 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002550 } else {
2551 res_counter_init(&mem->res, NULL);
2552 res_counter_init(&mem->memsw, NULL);
2553 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002554 mem->last_scanned_child = 0;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -08002555 spin_lock_init(&mem->reclaim_param_lock);
Balbir Singh6d61ef42009-01-07 18:08:06 -08002556
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002557 if (parent)
2558 mem->swappiness = get_swappiness(parent);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002559 atomic_set(&mem->refcnt, 1);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002560 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002561free_out:
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002562 __mem_cgroup_free(mem);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002563 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002564}
2565
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002566static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002567 struct cgroup *cont)
2568{
2569 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002570
2571 return mem_cgroup_force_empty(mem, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002572}
2573
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002574static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2575 struct cgroup *cont)
2576{
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002577 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002578
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002579 mem_cgroup_put(mem);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002580}
2581
2582static int mem_cgroup_populate(struct cgroup_subsys *ss,
2583 struct cgroup *cont)
2584{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002585 int ret;
2586
2587 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2588 ARRAY_SIZE(mem_cgroup_files));
2589
2590 if (!ret)
2591 ret = register_memsw_files(cont, ss);
2592 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002593}
2594
Balbir Singh67e465a2008-02-07 00:13:54 -08002595static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2596 struct cgroup *cont,
2597 struct cgroup *old_cont,
2598 struct task_struct *p)
2599{
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002600 mutex_lock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002601 /*
Nikanth Karthikesanf9717d22009-01-07 18:08:11 -08002602 * FIXME: It's better to move charges of this process from old
2603 * memcg to new memcg. But it's just on TODO-List now.
Balbir Singh67e465a2008-02-07 00:13:54 -08002604 */
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002605 mutex_unlock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002606}
2607
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002608struct cgroup_subsys mem_cgroup_subsys = {
2609 .name = "memory",
2610 .subsys_id = mem_cgroup_subsys_id,
2611 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002612 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002613 .destroy = mem_cgroup_destroy,
2614 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08002615 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002616 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002617 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002618};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002619
2620#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2621
2622static int __init disable_swap_account(char *s)
2623{
2624 really_do_swap_account = 0;
2625 return 1;
2626}
2627__setup("noswapaccount", disable_swap_account);
2628#endif