blob: e44fb0fbb80e0890923385d3e97aff8dfa9d9a1b [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
48/* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
49int 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 */
65 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
Balaji Rao55e462b2008-05-01 04:35:12 -070066 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
67 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080068
69 MEM_CGROUP_STAT_NSTATS,
70};
71
72struct mem_cgroup_stat_cpu {
73 s64 count[MEM_CGROUP_STAT_NSTATS];
74} ____cacheline_aligned_in_smp;
75
76struct mem_cgroup_stat {
Jan Blunckc8dad2b2009-01-07 18:07:53 -080077 struct mem_cgroup_stat_cpu cpustat[0];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080078};
79
80/*
81 * For accounting under irq disable, no need for increment preempt count.
82 */
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070083static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080084 enum mem_cgroup_stat_index idx, int val)
85{
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070086 stat->count[idx] += val;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080087}
88
89static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
90 enum mem_cgroup_stat_index idx)
91{
92 int cpu;
93 s64 ret = 0;
94 for_each_possible_cpu(cpu)
95 ret += stat->cpustat[cpu].count[idx];
96 return ret;
97}
98
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -070099static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat)
100{
101 s64 ret;
102
103 ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE);
104 ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS);
105 return ret;
106}
107
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800108/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800109 * per-zone information in memory controller.
110 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800111struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800112 /*
113 * spin_lock to protect the per cgroup LRU
114 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700115 struct list_head lists[NR_LRU_LISTS];
116 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800117
118 struct zone_reclaim_stat reclaim_stat;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800119};
120/* Macro for accessing counter */
121#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
122
123struct mem_cgroup_per_node {
124 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
125};
126
127struct mem_cgroup_lru_info {
128 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
129};
130
131/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800132 * The memory controller data structure. The memory controller controls both
133 * page cache and RSS per cgroup. We would eventually like to provide
134 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
135 * to help the administrator determine what knobs to tune.
136 *
137 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800138 * we hit the water mark. May be even add a low water mark, such that
139 * no reclaim occurs from a cgroup at it's low water mark, this is
140 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800141 */
142struct mem_cgroup {
143 struct cgroup_subsys_state css;
144 /*
145 * the counter to account for memory usage
146 */
147 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800148 /*
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800149 * the counter to account for mem+swap usage.
150 */
151 struct res_counter memsw;
152 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800153 * Per cgroup active and inactive list, similar to the
154 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800155 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800156 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800157
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800158 /*
159 protect against reclaim related member.
160 */
161 spinlock_t reclaim_param_lock;
162
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800163 int prev_priority; /* for recording reclaim priority */
Balbir Singh6d61ef42009-01-07 18:08:06 -0800164
165 /*
166 * While reclaiming in a hiearchy, we cache the last child we
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700167 * reclaimed from.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800168 */
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700169 int last_scanned_child;
Balbir Singh18f59ea2009-01-07 18:08:07 -0800170 /*
171 * Should the accounting and control be hierarchical, per subtree?
172 */
173 bool use_hierarchy;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800174 unsigned long last_oom_jiffies;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800175 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800176
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800177 unsigned int swappiness;
178
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800179 /*
Jan Blunckc8dad2b2009-01-07 18:07:53 -0800180 * statistics. This must be placed at the end of memcg.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800181 */
182 struct mem_cgroup_stat stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800183};
184
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800185enum charge_type {
186 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
187 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700188 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700189 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800190 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700191 NR_CHARGE_TYPE,
192};
193
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700194/* only for here (for easy reading.) */
195#define PCGF_CACHE (1UL << PCG_CACHE)
196#define PCGF_USED (1UL << PCG_USED)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700197#define PCGF_LOCK (1UL << PCG_LOCK)
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700198static const unsigned long
199pcg_default_flags[NR_CHARGE_TYPE] = {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800200 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
201 PCGF_USED | PCGF_LOCK, /* Anon */
202 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700203 0, /* FORCE */
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800204};
205
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800206/* for encoding cft->private value on file */
207#define _MEM (0)
208#define _MEMSWAP (1)
209#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
210#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
211#define MEMFILE_ATTR(val) ((val) & 0xffff)
212
213static void mem_cgroup_get(struct mem_cgroup *mem);
214static void mem_cgroup_put(struct mem_cgroup *mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -0800215static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800216
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700217static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
218 struct page_cgroup *pc,
219 bool charge)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800220{
221 int val = (charge)? 1 : -1;
222 struct mem_cgroup_stat *stat = &mem->stat;
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700223 struct mem_cgroup_stat_cpu *cpustat;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800224 int cpu = get_cpu();
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800225
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800226 cpustat = &stat->cpustat[cpu];
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700227 if (PageCgroupCache(pc))
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700228 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800229 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700230 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
Balaji Rao55e462b2008-05-01 04:35:12 -0700231
232 if (charge)
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700233 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700234 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
235 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700236 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700237 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800238 put_cpu();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800239}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800240
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800241static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800242mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
243{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800244 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
245}
246
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800247static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800248page_cgroup_zoneinfo(struct page_cgroup *pc)
249{
250 struct mem_cgroup *mem = pc->mem_cgroup;
251 int nid = page_cgroup_nid(pc);
252 int zid = page_cgroup_zid(pc);
253
KOSAKI Motohiro54992762009-01-07 18:08:18 -0800254 if (!mem)
255 return NULL;
256
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800257 return mem_cgroup_zoneinfo(mem, nid, zid);
258}
259
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700260static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
Christoph Lameterb69408e2008-10-18 20:26:14 -0700261 enum lru_list idx)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800262{
263 int nid, zid;
264 struct mem_cgroup_per_zone *mz;
265 u64 total = 0;
266
267 for_each_online_node(nid)
268 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
269 mz = mem_cgroup_zoneinfo(mem, nid, zid);
270 total += MEM_CGROUP_ZSTAT(mz, idx);
271 }
272 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800273}
274
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800275static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800276{
277 return container_of(cgroup_subsys_state(cont,
278 mem_cgroup_subsys_id), struct mem_cgroup,
279 css);
280}
281
Balbir Singhcf475ad2008-04-29 01:00:16 -0700282struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800283{
Balbir Singh31a78f22008-09-28 23:09:31 +0100284 /*
285 * mm_update_next_owner() may clear mm->owner to NULL
286 * if it races with swapoff, page migration, etc.
287 * So this can be called with p == NULL.
288 */
289 if (unlikely(!p))
290 return NULL;
291
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800292 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
293 struct mem_cgroup, css);
294}
295
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800296static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
297{
298 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700299
300 if (!mm)
301 return NULL;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800302 /*
303 * Because we have no locks, mm->owner's may be being moved to other
304 * cgroup. We use css_tryget() here even if this looks
305 * pessimistic (rather than adding locks here).
306 */
307 rcu_read_lock();
308 do {
309 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
310 if (unlikely(!mem))
311 break;
312 } while (!css_tryget(&mem->css));
313 rcu_read_unlock();
314 return mem;
315}
316
317static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem)
318{
319 if (!mem)
320 return true;
321 return css_is_removed(&mem->css);
322}
323
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700324
325/*
326 * Call callback function against all cgroup under hierarchy tree.
327 */
328static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
329 int (*func)(struct mem_cgroup *, void *))
330{
331 int found, ret, nextid;
332 struct cgroup_subsys_state *css;
333 struct mem_cgroup *mem;
334
335 if (!root->use_hierarchy)
336 return (*func)(root, data);
337
338 nextid = 1;
339 do {
340 ret = 0;
341 mem = NULL;
342
343 rcu_read_lock();
344 css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
345 &found);
346 if (css && css_tryget(css))
347 mem = container_of(css, struct mem_cgroup, css);
348 rcu_read_unlock();
349
350 if (mem) {
351 ret = (*func)(mem, data);
352 css_put(&mem->css);
353 }
354 nextid = found + 1;
355 } while (!ret && css);
356
357 return ret;
358}
359
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800360/*
361 * Following LRU functions are allowed to be used without PCG_LOCK.
362 * Operations are called by routine of global LRU independently from memcg.
363 * What we have to take care of here is validness of pc->mem_cgroup.
364 *
365 * Changes to pc->mem_cgroup happens when
366 * 1. charge
367 * 2. moving account
368 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
369 * It is added to LRU before charge.
370 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
371 * When moving account, the page is not on LRU. It's isolated.
372 */
373
374void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800375{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800376 struct page_cgroup *pc;
377 struct mem_cgroup *mem;
378 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700379
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800380 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800381 return;
382 pc = lookup_page_cgroup(page);
383 /* can happen while we handle swapcache. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800384 if (list_empty(&pc->lru) || !pc->mem_cgroup)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800385 return;
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800386 /*
387 * We don't check PCG_USED bit. It's cleared when the "page" is finally
388 * removed from global LRU.
389 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800390 mz = page_cgroup_zoneinfo(pc);
391 mem = pc->mem_cgroup;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700392 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800393 list_del_init(&pc->lru);
394 return;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800395}
396
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800397void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800398{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800399 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800400}
401
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800402void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -0800403{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800404 struct mem_cgroup_per_zone *mz;
405 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800406
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800407 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700408 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700409
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800410 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800411 /*
412 * Used bit is set without atomic ops but after smp_wmb().
413 * For making pc->mem_cgroup visible, insert smp_rmb() here.
414 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800415 smp_rmb();
416 /* unused page is not rotated. */
417 if (!PageCgroupUsed(pc))
418 return;
419 mz = page_cgroup_zoneinfo(pc);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700420 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -0800421}
422
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800423void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
424{
425 struct page_cgroup *pc;
426 struct mem_cgroup_per_zone *mz;
427
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800428 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800429 return;
430 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800431 /*
432 * Used bit is set without atomic ops but after smp_wmb().
433 * For making pc->mem_cgroup visible, insert smp_rmb() here.
434 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800435 smp_rmb();
436 if (!PageCgroupUsed(pc))
437 return;
438
439 mz = page_cgroup_zoneinfo(pc);
440 MEM_CGROUP_ZSTAT(mz, lru) += 1;
441 list_add(&pc->lru, &mz->lists[lru]);
442}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800443
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800444/*
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800445 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
446 * lru because the page may.be reused after it's fully uncharged (because of
447 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
448 * it again. This function is only used to charge SwapCache. It's done under
449 * lock_page and expected that zone->lru_lock is never held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800450 */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800451static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800452{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800453 unsigned long flags;
454 struct zone *zone = page_zone(page);
455 struct page_cgroup *pc = lookup_page_cgroup(page);
456
457 spin_lock_irqsave(&zone->lru_lock, flags);
458 /*
459 * Forget old LRU when this page_cgroup is *not* used. This Used bit
460 * is guarded by lock_page() because the page is SwapCache.
461 */
462 if (!PageCgroupUsed(pc))
463 mem_cgroup_del_lru_list(page, page_lru(page));
464 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800465}
466
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800467static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
468{
469 unsigned long flags;
470 struct zone *zone = page_zone(page);
471 struct page_cgroup *pc = lookup_page_cgroup(page);
472
473 spin_lock_irqsave(&zone->lru_lock, flags);
474 /* link when the page is linked to LRU but page_cgroup isn't */
475 if (PageLRU(page) && list_empty(&pc->lru))
476 mem_cgroup_add_lru_list(page, page_lru(page));
477 spin_unlock_irqrestore(&zone->lru_lock, flags);
478}
479
480
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800481void mem_cgroup_move_lists(struct page *page,
482 enum lru_list from, enum lru_list to)
483{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -0800484 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800485 return;
486 mem_cgroup_del_lru_list(page, from);
487 mem_cgroup_add_lru_list(page, to);
488}
489
David Rientjes4c4a2212008-02-07 00:14:06 -0800490int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
491{
492 int ret;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700493 struct mem_cgroup *curr = NULL;
David Rientjes4c4a2212008-02-07 00:14:06 -0800494
495 task_lock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700496 rcu_read_lock();
497 curr = try_get_mem_cgroup_from_mm(task->mm);
498 rcu_read_unlock();
David Rientjes4c4a2212008-02-07 00:14:06 -0800499 task_unlock(task);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700500 if (!curr)
501 return 0;
502 if (curr->use_hierarchy)
503 ret = css_is_ancestor(&curr->css, &mem->css);
504 else
505 ret = (curr == mem);
506 css_put(&curr->css);
David Rientjes4c4a2212008-02-07 00:14:06 -0800507 return ret;
508}
509
Balbir Singh66e17072008-02-07 00:13:56 -0800510/*
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800511 * prev_priority control...this will be used in memory reclaim path.
512 */
513int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
514{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800515 int prev_priority;
516
517 spin_lock(&mem->reclaim_param_lock);
518 prev_priority = mem->prev_priority;
519 spin_unlock(&mem->reclaim_param_lock);
520
521 return prev_priority;
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800522}
523
524void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
525{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800526 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800527 if (priority < mem->prev_priority)
528 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800529 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800530}
531
532void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
533{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800534 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800535 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800536 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800537}
538
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800539static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800540{
541 unsigned long active;
542 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800543 unsigned long gb;
544 unsigned long inactive_ratio;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800545
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700546 inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
547 active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800548
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800549 gb = (inactive + active) >> (30 - PAGE_SHIFT);
550 if (gb)
551 inactive_ratio = int_sqrt(10 * gb);
552 else
553 inactive_ratio = 1;
554
555 if (present_pages) {
556 present_pages[0] = inactive;
557 present_pages[1] = active;
558 }
559
560 return inactive_ratio;
561}
562
563int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
564{
565 unsigned long active;
566 unsigned long inactive;
567 unsigned long present_pages[2];
568 unsigned long inactive_ratio;
569
570 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
571
572 inactive = present_pages[0];
573 active = present_pages[1];
574
575 if (inactive * inactive_ratio < active)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800576 return 1;
577
578 return 0;
579}
580
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800581unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
582 struct zone *zone,
583 enum lru_list lru)
584{
585 int nid = zone->zone_pgdat->node_id;
586 int zid = zone_idx(zone);
587 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
588
589 return MEM_CGROUP_ZSTAT(mz, lru);
590}
591
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800592struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
593 struct zone *zone)
594{
595 int nid = zone->zone_pgdat->node_id;
596 int zid = zone_idx(zone);
597 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
598
599 return &mz->reclaim_stat;
600}
601
602struct zone_reclaim_stat *
603mem_cgroup_get_reclaim_stat_from_page(struct page *page)
604{
605 struct page_cgroup *pc;
606 struct mem_cgroup_per_zone *mz;
607
608 if (mem_cgroup_disabled())
609 return NULL;
610
611 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800612 /*
613 * Used bit is set without atomic ops but after smp_wmb().
614 * For making pc->mem_cgroup visible, insert smp_rmb() here.
615 */
616 smp_rmb();
617 if (!PageCgroupUsed(pc))
618 return NULL;
619
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800620 mz = page_cgroup_zoneinfo(pc);
621 if (!mz)
622 return NULL;
623
624 return &mz->reclaim_stat;
625}
626
Balbir Singh66e17072008-02-07 00:13:56 -0800627unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
628 struct list_head *dst,
629 unsigned long *scanned, int order,
630 int mode, struct zone *z,
631 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700632 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800633{
634 unsigned long nr_taken = 0;
635 struct page *page;
636 unsigned long scan;
637 LIST_HEAD(pc_list);
638 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800639 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800640 int nid = z->zone_pgdat->node_id;
641 int zid = zone_idx(z);
642 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700643 int lru = LRU_FILE * !!file + !!active;
Balbir Singh66e17072008-02-07 00:13:56 -0800644
Balbir Singhcf475ad2008-04-29 01:00:16 -0700645 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800646 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700647 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -0800648
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800649 scan = 0;
650 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800651 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800652 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800653
654 page = pc->page;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700655 if (unlikely(!PageCgroupUsed(pc)))
656 continue;
Hugh Dickins436c65412008-02-07 00:14:12 -0800657 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800658 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800659
Hugh Dickins436c65412008-02-07 00:14:12 -0800660 scan++;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700661 if (__isolate_lru_page(page, mode, file) == 0) {
Balbir Singh66e17072008-02-07 00:13:56 -0800662 list_move(&page->lru, dst);
663 nr_taken++;
664 }
665 }
666
Balbir Singh66e17072008-02-07 00:13:56 -0800667 *scanned = scan;
668 return nr_taken;
669}
670
Balbir Singh6d61ef42009-01-07 18:08:06 -0800671#define mem_cgroup_from_res_counter(counter, member) \
672 container_of(counter, struct mem_cgroup, member)
673
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800674static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
675{
676 if (do_swap_account) {
677 if (res_counter_check_under_limit(&mem->res) &&
678 res_counter_check_under_limit(&mem->memsw))
679 return true;
680 } else
681 if (res_counter_check_under_limit(&mem->res))
682 return true;
683 return false;
684}
685
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800686static unsigned int get_swappiness(struct mem_cgroup *memcg)
687{
688 struct cgroup *cgrp = memcg->css.cgroup;
689 unsigned int swappiness;
690
691 /* root ? */
692 if (cgrp->parent == NULL)
693 return vm_swappiness;
694
695 spin_lock(&memcg->reclaim_param_lock);
696 swappiness = memcg->swappiness;
697 spin_unlock(&memcg->reclaim_param_lock);
698
699 return swappiness;
700}
701
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700702static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
703{
704 int *val = data;
705 (*val)++;
706 return 0;
707}
Balbir Singhe2224322009-04-02 16:57:39 -0700708
709/**
710 * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode.
711 * @memcg: The memory cgroup that went over limit
712 * @p: Task that is going to be killed
713 *
714 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
715 * enabled
716 */
717void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
718{
719 struct cgroup *task_cgrp;
720 struct cgroup *mem_cgrp;
721 /*
722 * Need a buffer in BSS, can't rely on allocations. The code relies
723 * on the assumption that OOM is serialized for memory controller.
724 * If this assumption is broken, revisit this code.
725 */
726 static char memcg_name[PATH_MAX];
727 int ret;
728
729 if (!memcg)
730 return;
731
732
733 rcu_read_lock();
734
735 mem_cgrp = memcg->css.cgroup;
736 task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
737
738 ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
739 if (ret < 0) {
740 /*
741 * Unfortunately, we are unable to convert to a useful name
742 * But we'll still print out the usage information
743 */
744 rcu_read_unlock();
745 goto done;
746 }
747 rcu_read_unlock();
748
749 printk(KERN_INFO "Task in %s killed", memcg_name);
750
751 rcu_read_lock();
752 ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
753 if (ret < 0) {
754 rcu_read_unlock();
755 goto done;
756 }
757 rcu_read_unlock();
758
759 /*
760 * Continues from above, so we don't need an KERN_ level
761 */
762 printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
763done:
764
765 printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
766 res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
767 res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
768 res_counter_read_u64(&memcg->res, RES_FAILCNT));
769 printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
770 "failcnt %llu\n",
771 res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
772 res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
773 res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
774}
775
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700776/*
777 * This function returns the number of memcg under hierarchy tree. Returns
778 * 1(self count) if no children.
779 */
780static int mem_cgroup_count_children(struct mem_cgroup *mem)
781{
782 int num = 0;
783 mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
784 return num;
785}
786
Balbir Singh6d61ef42009-01-07 18:08:06 -0800787/*
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700788 * Visit the first child (need not be the first child as per the ordering
789 * of the cgroup list, since we track last_scanned_child) of @mem and use
790 * that to reclaim free pages from.
791 */
792static struct mem_cgroup *
793mem_cgroup_select_victim(struct mem_cgroup *root_mem)
794{
795 struct mem_cgroup *ret = NULL;
796 struct cgroup_subsys_state *css;
797 int nextid, found;
798
799 if (!root_mem->use_hierarchy) {
800 css_get(&root_mem->css);
801 ret = root_mem;
802 }
803
804 while (!ret) {
805 rcu_read_lock();
806 nextid = root_mem->last_scanned_child + 1;
807 css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
808 &found);
809 if (css && css_tryget(css))
810 ret = container_of(css, struct mem_cgroup, css);
811
812 rcu_read_unlock();
813 /* Updates scanning parameter */
814 spin_lock(&root_mem->reclaim_param_lock);
815 if (!css) {
816 /* this means start scan from ID:1 */
817 root_mem->last_scanned_child = 0;
818 } else
819 root_mem->last_scanned_child = found;
820 spin_unlock(&root_mem->reclaim_param_lock);
821 }
822
823 return ret;
824}
825
826/*
827 * Scan the hierarchy if needed to reclaim memory. We remember the last child
828 * we reclaimed from, so that we don't end up penalizing one child extensively
829 * based on its position in the children list.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800830 *
831 * root_mem is the original ancestor that we've been reclaim from.
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700832 *
833 * We give up and return to the caller when we visit root_mem twice.
834 * (other groups can be removed while we're walking....)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700835 *
836 * If shrink==true, for avoiding to free too much, this returns immedieately.
Balbir Singh6d61ef42009-01-07 18:08:06 -0800837 */
838static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700839 gfp_t gfp_mask, bool noswap, bool shrink)
Balbir Singh6d61ef42009-01-07 18:08:06 -0800840{
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700841 struct mem_cgroup *victim;
842 int ret, total = 0;
843 int loop = 0;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800844
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700845 while (loop < 2) {
846 victim = mem_cgroup_select_victim(root_mem);
847 if (victim == root_mem)
848 loop++;
849 if (!mem_cgroup_local_usage(&victim->stat)) {
850 /* this cgroup's local usage == 0 */
851 css_put(&victim->css);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800852 continue;
853 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700854 /* we use swappiness of local cgroup */
855 ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap,
856 get_swappiness(victim));
857 css_put(&victim->css);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700858 /*
859 * At shrinking usage, we can't check we should stop here or
860 * reclaim more. It's depends on callers. last_scanned_child
861 * will work enough for keeping fairness under tree.
862 */
863 if (shrink)
864 return ret;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700865 total += ret;
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800866 if (mem_cgroup_check_under_limit(root_mem))
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700867 return 1 + total;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800868 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -0700869 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800870}
871
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800872bool mem_cgroup_oom_called(struct task_struct *task)
873{
874 bool ret = false;
875 struct mem_cgroup *mem;
876 struct mm_struct *mm;
877
878 rcu_read_lock();
879 mm = task->mm;
880 if (!mm)
881 mm = &init_mm;
882 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
883 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
884 ret = true;
885 rcu_read_unlock();
886 return ret;
887}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700888
889static int record_last_oom_cb(struct mem_cgroup *mem, void *data)
890{
891 mem->last_oom_jiffies = jiffies;
892 return 0;
893}
894
895static void record_last_oom(struct mem_cgroup *mem)
896{
897 mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb);
898}
899
900
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800901/*
902 * Unlike exported interface, "oom" parameter is added. if oom==true,
903 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800904 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800905static int __mem_cgroup_try_charge(struct mm_struct *mm,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800906 gfp_t gfp_mask, struct mem_cgroup **memcg,
907 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800908{
Balbir Singh6d61ef42009-01-07 18:08:06 -0800909 struct mem_cgroup *mem, *mem_over_limit;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800910 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800911 struct res_counter *fail_res;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800912
913 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
914 /* Don't account this! */
915 *memcg = NULL;
916 return 0;
917 }
918
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800919 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800920 * We always charge the cgroup the mm_struct belongs to.
921 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800922 * thread group leader migrates. It's possible that mm is not
923 * set, if so charge the init_mm (happens for pagecache usage).
924 */
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800925 mem = *memcg;
926 if (likely(!mem)) {
927 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800928 *memcg = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700929 } else {
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800930 css_get(&mem->css);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700931 }
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800932 if (unlikely(!mem))
933 return 0;
934
KAMEZAWA Hiroyukia8031cb2009-04-13 14:40:08 -0700935 VM_BUG_ON(!mem || mem_cgroup_is_obsolete(mem));
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800936
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800937 while (1) {
938 int ret;
939 bool noswap = false;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800940
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800941 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800942 if (likely(!ret)) {
943 if (!do_swap_account)
944 break;
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800945 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
946 &fail_res);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800947 if (likely(!ret))
948 break;
949 /* mem+swap counter fails */
950 res_counter_uncharge(&mem->res, PAGE_SIZE);
951 noswap = true;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800952 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
953 memsw);
954 } else
955 /* mem counter fails */
956 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
957 res);
958
Hugh Dickins3be91272008-02-07 00:14:19 -0800959 if (!(gfp_mask & __GFP_WAIT))
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800960 goto nomem;
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800961
Balbir Singh6d61ef42009-01-07 18:08:06 -0800962 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -0700963 noswap, false);
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800964 if (ret)
965 continue;
Balbir Singh66e17072008-02-07 00:13:56 -0800966
967 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800968 * try_to_free_mem_cgroup_pages() might not give us a full
969 * picture of reclaim. Some pages are reclaimed and might be
970 * moved to swap cache or just unmapped from the cgroup.
971 * Check the limit again to see if the reclaim reduced the
972 * current usage of the cgroup before giving up
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800973 *
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800974 */
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800975 if (mem_cgroup_check_under_limit(mem_over_limit))
976 continue;
Hugh Dickins3be91272008-02-07 00:14:19 -0800977
978 if (!nr_retries--) {
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800979 if (oom) {
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -0800980 mutex_lock(&memcg_tasklist);
KAMEZAWA Hiroyuki88700752009-01-07 18:08:09 -0800981 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -0800982 mutex_unlock(&memcg_tasklist);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700983 record_last_oom(mem_over_limit);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800984 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800985 goto nomem;
Balbir Singh66e17072008-02-07 00:13:56 -0800986 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800987 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800988 return 0;
989nomem:
990 css_put(&mem->css);
991 return -ENOMEM;
992}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800993
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -0700994
995/*
996 * A helper function to get mem_cgroup from ID. must be called under
997 * rcu_read_lock(). The caller must check css_is_removed() or some if
998 * it's concern. (dropping refcnt from swap can be called against removed
999 * memcg.)
1000 */
1001static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1002{
1003 struct cgroup_subsys_state *css;
1004
1005 /* ID 0 is unused ID */
1006 if (!id)
1007 return NULL;
1008 css = css_lookup(&mem_cgroup_subsys, id);
1009 if (!css)
1010 return NULL;
1011 return container_of(css, struct mem_cgroup, css);
1012}
1013
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001014static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
1015{
1016 struct mem_cgroup *mem;
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001017 struct page_cgroup *pc;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001018 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001019 swp_entry_t ent;
1020
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001021 VM_BUG_ON(!PageLocked(page));
1022
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001023 if (!PageSwapCache(page))
1024 return NULL;
1025
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001026 pc = lookup_page_cgroup(page);
1027 /*
1028 * Used bit of swapcache is solid under page lock.
1029 */
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001030 if (PageCgroupUsed(pc)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001031 mem = pc->mem_cgroup;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001032 if (mem && !css_tryget(&mem->css))
1033 mem = NULL;
1034 } else {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001035 ent.val = page_private(page);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001036 id = lookup_swap_cgroup(ent);
1037 rcu_read_lock();
1038 mem = mem_cgroup_lookup(id);
1039 if (mem && !css_tryget(&mem->css))
1040 mem = NULL;
1041 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07001042 }
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001043 return mem;
1044}
1045
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001046/*
Daisuke Nishimuraa5e924f2009-01-07 18:08:28 -08001047 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001048 * USED state. If already USED, uncharge and return.
1049 */
1050
1051static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1052 struct page_cgroup *pc,
1053 enum charge_type ctype)
1054{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001055 /* try_charge() can return NULL to *memcg, taking care of it. */
1056 if (!mem)
1057 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001058
1059 lock_page_cgroup(pc);
1060 if (unlikely(PageCgroupUsed(pc))) {
1061 unlock_page_cgroup(pc);
1062 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001063 if (do_swap_account)
1064 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001065 css_put(&mem->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001066 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001067 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001068 pc->mem_cgroup = mem;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001069 smp_wmb();
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001070 pc->flags = pcg_default_flags[ctype];
Hugh Dickins3be91272008-02-07 00:14:19 -08001071
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001072 mem_cgroup_charge_statistics(mem, pc, true);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001073
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001074 unlock_page_cgroup(pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001075}
1076
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001077/**
1078 * mem_cgroup_move_account - move account of the page
1079 * @pc: page_cgroup of the page.
1080 * @from: mem_cgroup which the page is moved from.
1081 * @to: mem_cgroup which the page is moved to. @from != @to.
1082 *
1083 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001084 * - page is not on LRU (isolate_page() is useful.)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001085 *
1086 * returns 0 at success,
1087 * returns -EBUSY when lock is busy or "pc" is unstable.
1088 *
1089 * This function does "uncharge" from old cgroup but doesn't do "charge" to
1090 * new cgroup. It should be done by a caller.
1091 */
1092
1093static int mem_cgroup_move_account(struct page_cgroup *pc,
1094 struct mem_cgroup *from, struct mem_cgroup *to)
1095{
1096 struct mem_cgroup_per_zone *from_mz, *to_mz;
1097 int nid, zid;
1098 int ret = -EBUSY;
1099
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001100 VM_BUG_ON(from == to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001101 VM_BUG_ON(PageLRU(pc->page));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001102
1103 nid = page_cgroup_nid(pc);
1104 zid = page_cgroup_zid(pc);
1105 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
1106 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
1107
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001108 if (!trylock_page_cgroup(pc))
1109 return ret;
1110
1111 if (!PageCgroupUsed(pc))
1112 goto out;
1113
1114 if (pc->mem_cgroup != from)
1115 goto out;
1116
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001117 res_counter_uncharge(&from->res, PAGE_SIZE);
1118 mem_cgroup_charge_statistics(from, pc, false);
1119 if (do_swap_account)
1120 res_counter_uncharge(&from->memsw, PAGE_SIZE);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001121 css_put(&from->css);
1122
1123 css_get(&to->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001124 pc->mem_cgroup = to;
1125 mem_cgroup_charge_statistics(to, pc, true);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001126 ret = 0;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001127out:
1128 unlock_page_cgroup(pc);
1129 return ret;
1130}
1131
1132/*
1133 * move charges to its parent.
1134 */
1135
1136static int mem_cgroup_move_parent(struct page_cgroup *pc,
1137 struct mem_cgroup *child,
1138 gfp_t gfp_mask)
1139{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001140 struct page *page = pc->page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001141 struct cgroup *cg = child->css.cgroup;
1142 struct cgroup *pcg = cg->parent;
1143 struct mem_cgroup *parent;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001144 int ret;
1145
1146 /* Is ROOT ? */
1147 if (!pcg)
1148 return -EINVAL;
1149
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001150
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001151 parent = mem_cgroup_from_cont(pcg);
1152
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001153
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001154 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001155 if (ret || !parent)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001156 return ret;
1157
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001158 if (!get_page_unless_zero(page)) {
1159 ret = -EBUSY;
1160 goto uncharge;
1161 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001162
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001163 ret = isolate_lru_page(page);
1164
1165 if (ret)
1166 goto cancel;
1167
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001168 ret = mem_cgroup_move_account(pc, child, parent);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001169
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001170 putback_lru_page(page);
1171 if (!ret) {
1172 put_page(page);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001173 /* drop extra refcnt by try_charge() */
1174 css_put(&parent->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001175 return 0;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001176 }
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001177
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001178cancel:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001179 put_page(page);
1180uncharge:
1181 /* drop extra refcnt by try_charge() */
1182 css_put(&parent->css);
1183 /* uncharge if move fails */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001184 res_counter_uncharge(&parent->res, PAGE_SIZE);
1185 if (do_swap_account)
1186 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001187 return ret;
1188}
1189
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001190/*
1191 * Charge the memory controller for page usage.
1192 * Return
1193 * 0 if the charge was successful
1194 * < 0 if the cgroup is over its limit
1195 */
1196static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1197 gfp_t gfp_mask, enum charge_type ctype,
1198 struct mem_cgroup *memcg)
1199{
1200 struct mem_cgroup *mem;
1201 struct page_cgroup *pc;
1202 int ret;
1203
1204 pc = lookup_page_cgroup(page);
1205 /* can happen at boot */
1206 if (unlikely(!pc))
1207 return 0;
1208 prefetchw(pc);
1209
1210 mem = memcg;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001211 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001212 if (ret || !mem)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001213 return ret;
1214
1215 __mem_cgroup_commit_charge(mem, pc, ctype);
1216 return 0;
1217}
1218
1219int mem_cgroup_newpage_charge(struct page *page,
1220 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001221{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001222 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001223 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001224 if (PageCompound(page))
1225 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001226 /*
1227 * If already mapped, we don't have to account.
1228 * If page cache, page->mapping has address_space.
1229 * But page->mapping may have out-of-use anon_vma pointer,
1230 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1231 * is NULL.
1232 */
1233 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1234 return 0;
1235 if (unlikely(!mm))
1236 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001237 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001238 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001239}
1240
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001241static void
1242__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1243 enum charge_type ctype);
1244
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001245int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1246 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08001247{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001248 struct mem_cgroup *mem = NULL;
1249 int ret;
1250
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001251 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001252 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001253 if (PageCompound(page))
1254 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001255 /*
1256 * Corner case handling. This is called from add_to_page_cache()
1257 * in usual. But some FS (shmem) precharges this page before calling it
1258 * and call add_to_page_cache() with GFP_NOWAIT.
1259 *
1260 * For GFP_NOWAIT case, the page may be pre-charged before calling
1261 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1262 * charge twice. (It works but has to pay a bit larger cost.)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001263 * And when the page is SwapCache, it should take swap information
1264 * into account. This is under lock_page() now.
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001265 */
1266 if (!(gfp_mask & __GFP_WAIT)) {
1267 struct page_cgroup *pc;
1268
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001269
1270 pc = lookup_page_cgroup(page);
1271 if (!pc)
1272 return 0;
1273 lock_page_cgroup(pc);
1274 if (PageCgroupUsed(pc)) {
1275 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001276 return 0;
1277 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001278 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001279 }
1280
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001281 if (unlikely(!mm && !mem))
Balbir Singh8697d332008-02-07 00:13:59 -08001282 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001283
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001284 if (page_is_file_cache(page))
1285 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001286 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001287
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001288 /* shmem */
1289 if (PageSwapCache(page)) {
1290 ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
1291 if (!ret)
1292 __mem_cgroup_commit_charge_swapin(page, mem,
1293 MEM_CGROUP_CHARGE_TYPE_SHMEM);
1294 } else
1295 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1296 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001297
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001298 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001299}
1300
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001301/*
1302 * While swap-in, try_charge -> commit or cancel, the page is locked.
1303 * And when try_charge() successfully returns, one refcnt to memcg without
1304 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1305 * "commit()" or removed by "cancel()"
1306 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001307int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1308 struct page *page,
1309 gfp_t mask, struct mem_cgroup **ptr)
1310{
1311 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001312 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001313
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001314 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001315 return 0;
1316
1317 if (!do_swap_account)
1318 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001319 /*
1320 * A racing thread's fault, or swapoff, may have already updated
1321 * the pte, and even removed page from swap cache: return success
1322 * to go on to do_swap_page()'s pte_same() test, which should fail.
1323 */
1324 if (!PageSwapCache(page))
1325 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001326 mem = try_get_mem_cgroup_from_swapcache(page);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001327 if (!mem)
1328 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001329 *ptr = mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001330 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1331 /* drop extra refcnt from tryget */
1332 css_put(&mem->css);
1333 return ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001334charge_cur_mm:
1335 if (unlikely(!mm))
1336 mm = &init_mm;
1337 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1338}
1339
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001340static void
1341__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
1342 enum charge_type ctype)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001343{
1344 struct page_cgroup *pc;
1345
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001346 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001347 return;
1348 if (!ptr)
1349 return;
1350 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001351 mem_cgroup_lru_del_before_commit_swapcache(page);
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001352 __mem_cgroup_commit_charge(ptr, pc, ctype);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001353 mem_cgroup_lru_add_after_commit_swapcache(page);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001354 /*
1355 * Now swap is on-memory. This means this page may be
1356 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001357 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1358 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1359 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001360 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001361 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001362 swp_entry_t ent = {.val = page_private(page)};
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001363 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001364 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001365
1366 id = swap_cgroup_record(ent, 0);
1367 rcu_read_lock();
1368 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001369 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001370 /*
1371 * This recorded memcg can be obsolete one. So, avoid
1372 * calling css_tryget
1373 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001374 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1375 mem_cgroup_put(memcg);
1376 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001377 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001378 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001379 /* add this page(page_cgroup) to the LRU we want. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001380
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001381}
1382
Daisuke Nishimura83aae4c2009-04-02 16:57:48 -07001383void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1384{
1385 __mem_cgroup_commit_charge_swapin(page, ptr,
1386 MEM_CGROUP_CHARGE_TYPE_MAPPED);
1387}
1388
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001389void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1390{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001391 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001392 return;
1393 if (!mem)
1394 return;
1395 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001396 if (do_swap_account)
1397 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001398 css_put(&mem->css);
1399}
1400
1401
Balbir Singh8697d332008-02-07 00:13:59 -08001402/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001403 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001404 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001405static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001406__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001407{
Hugh Dickins82895462008-03-04 14:29:08 -08001408 struct page_cgroup *pc;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001409 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001410 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001411
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001412 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001413 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07001414
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001415 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001416 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001417
Balbir Singh8697d332008-02-07 00:13:59 -08001418 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08001419 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08001420 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001421 pc = lookup_page_cgroup(page);
1422 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001423 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001424
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001425 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001426
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001427 mem = pc->mem_cgroup;
1428
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001429 if (!PageCgroupUsed(pc))
1430 goto unlock_out;
1431
1432 switch (ctype) {
1433 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1434 if (page_mapped(page))
1435 goto unlock_out;
1436 break;
1437 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1438 if (!PageAnon(page)) { /* Shared memory */
1439 if (page->mapping && !page_is_file_cache(page))
1440 goto unlock_out;
1441 } else if (page_mapped(page)) /* Anon */
1442 goto unlock_out;
1443 break;
1444 default:
1445 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001446 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001447
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001448 res_counter_uncharge(&mem->res, PAGE_SIZE);
1449 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1450 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001451 mem_cgroup_charge_statistics(mem, pc, false);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07001452
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001453 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001454 /*
1455 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1456 * freed from LRU. This is safe because uncharged page is expected not
1457 * to be reused (freed soon). Exception is SwapCache, it's handled by
1458 * special functions.
1459 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001460
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001461 mz = page_cgroup_zoneinfo(pc);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001462 unlock_page_cgroup(pc);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001463
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001464 /* at swapout, this memcg will be accessed to record to swap */
1465 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1466 css_put(&mem->css);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001467
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001468 return mem;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001469
1470unlock_out:
1471 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001472 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08001473}
1474
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001475void mem_cgroup_uncharge_page(struct page *page)
1476{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001477 /* early check. */
1478 if (page_mapped(page))
1479 return;
1480 if (page->mapping && !PageAnon(page))
1481 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001482 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1483}
1484
1485void mem_cgroup_uncharge_cache_page(struct page *page)
1486{
1487 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07001488 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001489 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1490}
1491
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001492/*
1493 * called from __delete_from_swap_cache() and drop "page" account.
1494 * memcg information is recorded to swap_cgroup of "ent"
1495 */
1496void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001497{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001498 struct mem_cgroup *memcg;
1499
1500 memcg = __mem_cgroup_uncharge_common(page,
1501 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1502 /* record memcg information */
1503 if (do_swap_account && memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001504 swap_cgroup_record(ent, css_id(&memcg->css));
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001505 mem_cgroup_get(memcg);
1506 }
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001507 if (memcg)
1508 css_put(&memcg->css);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001509}
1510
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001511#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1512/*
1513 * called from swap_entry_free(). remove record in swap_cgroup and
1514 * uncharge "memsw" account.
1515 */
1516void mem_cgroup_uncharge_swap(swp_entry_t ent)
1517{
1518 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001519 unsigned short id;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001520
1521 if (!do_swap_account)
1522 return;
1523
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001524 id = swap_cgroup_record(ent, 0);
1525 rcu_read_lock();
1526 memcg = mem_cgroup_lookup(id);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001527 if (memcg) {
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001528 /*
1529 * We uncharge this because swap is freed.
1530 * This memcg can be obsolete one. We avoid calling css_tryget
1531 */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001532 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1533 mem_cgroup_put(memcg);
1534 }
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07001535 rcu_read_unlock();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001536}
1537#endif
1538
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001539/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001540 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1541 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001542 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001543int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001544{
1545 struct page_cgroup *pc;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001546 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001547 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001548
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001549 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07001550 return 0;
1551
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001552 pc = lookup_page_cgroup(page);
1553 lock_page_cgroup(pc);
1554 if (PageCgroupUsed(pc)) {
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001555 mem = pc->mem_cgroup;
1556 css_get(&mem->css);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001557 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001558 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001559
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001560 if (mem) {
Daisuke Nishimura3bb4edf2009-01-07 18:08:28 -08001561 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001562 css_put(&mem->css);
1563 }
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001564 *ptr = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001565 return ret;
1566}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001567
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001568/* remove redundant charge if migration failed*/
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001569void mem_cgroup_end_migration(struct mem_cgroup *mem,
1570 struct page *oldpage, struct page *newpage)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001571{
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001572 struct page *target, *unused;
1573 struct page_cgroup *pc;
1574 enum charge_type ctype;
1575
1576 if (!mem)
1577 return;
1578
1579 /* at migration success, oldpage->mapping is NULL. */
1580 if (oldpage->mapping) {
1581 target = oldpage;
1582 unused = NULL;
1583 } else {
1584 target = newpage;
1585 unused = oldpage;
1586 }
1587
1588 if (PageAnon(target))
1589 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1590 else if (page_is_file_cache(target))
1591 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1592 else
1593 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1594
1595 /* unused page is not on radix-tree now. */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001596 if (unused)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001597 __mem_cgroup_uncharge_common(unused, ctype);
1598
1599 pc = lookup_page_cgroup(target);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001600 /*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001601 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1602 * So, double-counting is effectively avoided.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001603 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001604 __mem_cgroup_commit_charge(mem, pc, ctype);
1605
1606 /*
1607 * Both of oldpage and newpage are still under lock_page().
1608 * Then, we don't have to care about race in radix-tree.
1609 * But we have to be careful that this page is unmapped or not.
1610 *
1611 * There is a case for !page_mapped(). At the start of
1612 * migration, oldpage was mapped. But now, it's zapped.
1613 * But we know *target* page is not freed/reused under us.
1614 * mem_cgroup_uncharge_page() does all necessary checks.
1615 */
1616 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1617 mem_cgroup_uncharge_page(target);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001618}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001619
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001620/*
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001621 * A call to try to shrink memory usage under specified resource controller.
1622 * This is typically used for page reclaiming for shmem for reducing side
1623 * effect of page allocation from shmem, which is used by some mem_cgroup.
1624 */
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001625int mem_cgroup_shrink_usage(struct page *page,
1626 struct mm_struct *mm,
1627 gfp_t gfp_mask)
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001628{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001629 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001630 int progress = 0;
1631 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1632
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08001633 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001634 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001635 if (page)
1636 mem = try_get_mem_cgroup_from_swapcache(page);
1637 if (!mem && mm)
1638 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001639 if (unlikely(!mem))
Balbir Singh31a78f22008-09-28 23:09:31 +01001640 return 0;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001641
1642 do {
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001643 progress = mem_cgroup_hierarchical_reclaim(mem,
1644 gfp_mask, true, false);
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -08001645 progress += mem_cgroup_check_under_limit(mem);
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001646 } while (!progress && --retry);
1647
1648 css_put(&mem->css);
1649 if (!retry)
1650 return -ENOMEM;
1651 return 0;
1652}
1653
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001654static DEFINE_MUTEX(set_limit_mutex);
1655
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08001656static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001657 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001658{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001659 int retry_count;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001660 int progress;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001661 u64 memswlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001662 int ret = 0;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001663 int children = mem_cgroup_count_children(memcg);
1664 u64 curusage, oldusage;
1665
1666 /*
1667 * For keeping hierarchical_reclaim simple, how long we should retry
1668 * is depends on callers. We set our retry-count to be function
1669 * of # of children which we should visit in this loop.
1670 */
1671 retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
1672
1673 oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001674
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001675 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001676 if (signal_pending(current)) {
1677 ret = -EINTR;
1678 break;
1679 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001680 /*
1681 * Rather than hide all in some function, I do this in
1682 * open coded manner. You see what this really does.
1683 * We have to guarantee mem->res.limit < mem->memsw.limit.
1684 */
1685 mutex_lock(&set_limit_mutex);
1686 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1687 if (memswlimit < val) {
1688 ret = -EINVAL;
1689 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001690 break;
1691 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001692 ret = res_counter_set_limit(&memcg->res, val);
1693 mutex_unlock(&set_limit_mutex);
1694
1695 if (!ret)
1696 break;
1697
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001698 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001699 false, true);
1700 curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
1701 /* Usage is reduced ? */
1702 if (curusage >= oldusage)
1703 retry_count--;
1704 else
1705 oldusage = curusage;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001706 }
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001707
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001708 return ret;
1709}
1710
1711int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1712 unsigned long long val)
1713{
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001714 int retry_count;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001715 u64 memlimit, oldusage, curusage;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001716 int children = mem_cgroup_count_children(memcg);
1717 int ret = -EBUSY;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001718
1719 if (!do_swap_account)
1720 return -EINVAL;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001721 /* see mem_cgroup_resize_res_limit */
1722 retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
1723 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001724 while (retry_count) {
1725 if (signal_pending(current)) {
1726 ret = -EINTR;
1727 break;
1728 }
1729 /*
1730 * Rather than hide all in some function, I do this in
1731 * open coded manner. You see what this really does.
1732 * We have to guarantee mem->res.limit < mem->memsw.limit.
1733 */
1734 mutex_lock(&set_limit_mutex);
1735 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1736 if (memlimit > val) {
1737 ret = -EINVAL;
1738 mutex_unlock(&set_limit_mutex);
1739 break;
1740 }
1741 ret = res_counter_set_limit(&memcg->memsw, val);
1742 mutex_unlock(&set_limit_mutex);
1743
1744 if (!ret)
1745 break;
1746
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001747 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001748 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001749 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001750 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001751 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001752 else
1753 oldusage = curusage;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001754 }
1755 return ret;
1756}
1757
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001758/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001759 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001760 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1761 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001762static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001763 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001764{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001765 struct zone *zone;
1766 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001767 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001768 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001769 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001770 int ret = 0;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -08001771
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001772 zone = &NODE_DATA(node)->node_zones[zid];
1773 mz = mem_cgroup_zoneinfo(mem, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001774 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001775
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001776 loop = MEM_CGROUP_ZSTAT(mz, lru);
1777 /* give some margin against EBUSY etc...*/
1778 loop += 256;
1779 busy = NULL;
1780 while (loop--) {
1781 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001782 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001783 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001784 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001785 break;
1786 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001787 pc = list_entry(list->prev, struct page_cgroup, lru);
1788 if (busy == pc) {
1789 list_move(&pc->lru, list);
1790 busy = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001791 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001792 continue;
1793 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001794 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001795
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001796 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001797 if (ret == -ENOMEM)
1798 break;
1799
1800 if (ret == -EBUSY || ret == -EINVAL) {
1801 /* found lock contention or "pc" is obsolete. */
1802 busy = pc;
1803 cond_resched();
1804 } else
1805 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001806 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001807
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001808 if (!ret && !list_empty(list))
1809 return -EBUSY;
1810 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001811}
1812
1813/*
1814 * make mem_cgroup's charge to be 0 if there is no task.
1815 * This enables deleting this mem_cgroup.
1816 */
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001817static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001818{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001819 int ret;
1820 int node, zid, shrink;
1821 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001822 struct cgroup *cgrp = mem->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001823
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001824 css_get(&mem->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001825
1826 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001827 /* should free all ? */
1828 if (free_all)
1829 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001830move_account:
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001831 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001832 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001833 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001834 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001835 ret = -EINTR;
1836 if (signal_pending(current))
1837 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001838 /* This is for making all *used* pages to be on LRU. */
1839 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001840 ret = 0;
KAMEZAWA Hiroyuki299b4ea2009-01-29 14:25:17 -08001841 for_each_node_state(node, N_HIGH_MEMORY) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001842 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001843 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001844 for_each_lru(l) {
1845 ret = mem_cgroup_force_empty_list(mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001846 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001847 if (ret)
1848 break;
1849 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001850 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001851 if (ret)
1852 break;
1853 }
1854 /* it seems parent cgroup doesn't have enough mem */
1855 if (ret == -ENOMEM)
1856 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001857 cond_resched();
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001858 }
1859 ret = 0;
1860out:
1861 css_put(&mem->css);
1862 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001863
1864try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001865 /* returns EBUSY if there is a task or if we come here twice. */
1866 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001867 ret = -EBUSY;
1868 goto out;
1869 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001870 /* we call try-to-free pages for make this cgroup empty */
1871 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001872 /* try to free all pages in this cgroup */
1873 shrink = 1;
1874 while (nr_retries && mem->res.usage > 0) {
1875 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001876
1877 if (signal_pending(current)) {
1878 ret = -EINTR;
1879 goto out;
1880 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001881 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1882 false, get_swappiness(mem));
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001883 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001884 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001885 /* maybe some writeback is necessary */
1886 congestion_wait(WRITE, HZ/10);
1887 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001888
1889 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001890 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001891 /* try move_account...there may be some *locked* pages. */
1892 if (mem->res.usage)
1893 goto move_account;
1894 ret = 0;
1895 goto out;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001896}
1897
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001898int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1899{
1900 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1901}
1902
1903
Balbir Singh18f59ea2009-01-07 18:08:07 -08001904static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1905{
1906 return mem_cgroup_from_cont(cont)->use_hierarchy;
1907}
1908
1909static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1910 u64 val)
1911{
1912 int retval = 0;
1913 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1914 struct cgroup *parent = cont->parent;
1915 struct mem_cgroup *parent_mem = NULL;
1916
1917 if (parent)
1918 parent_mem = mem_cgroup_from_cont(parent);
1919
1920 cgroup_lock();
1921 /*
1922 * If parent's use_hiearchy is set, we can't make any modifications
1923 * in the child subtrees. If it is unset, then the change can
1924 * occur, provided the current cgroup has no children.
1925 *
1926 * For the root cgroup, parent_mem is NULL, we allow value to be
1927 * set if there are no children.
1928 */
1929 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1930 (val == 1 || val == 0)) {
1931 if (list_empty(&cont->children))
1932 mem->use_hierarchy = val;
1933 else
1934 retval = -EBUSY;
1935 } else
1936 retval = -EINVAL;
1937 cgroup_unlock();
1938
1939 return retval;
1940}
1941
Paul Menage2c3daa72008-04-29 00:59:58 -07001942static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001943{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001944 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1945 u64 val = 0;
1946 int type, name;
1947
1948 type = MEMFILE_TYPE(cft->private);
1949 name = MEMFILE_ATTR(cft->private);
1950 switch (type) {
1951 case _MEM:
1952 val = res_counter_read_u64(&mem->res, name);
1953 break;
1954 case _MEMSWAP:
1955 if (do_swap_account)
1956 val = res_counter_read_u64(&mem->memsw, name);
1957 break;
1958 default:
1959 BUG();
1960 break;
1961 }
1962 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001963}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001964/*
1965 * The user of this function is...
1966 * RES_LIMIT.
1967 */
Paul Menage856c13a2008-07-25 01:47:04 -07001968static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1969 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001970{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001971 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001972 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001973 unsigned long long val;
1974 int ret;
1975
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001976 type = MEMFILE_TYPE(cft->private);
1977 name = MEMFILE_ATTR(cft->private);
1978 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001979 case RES_LIMIT:
1980 /* This function does all necessary parse...reuse it */
1981 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001982 if (ret)
1983 break;
1984 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001985 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001986 else
1987 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001988 break;
1989 default:
1990 ret = -EINVAL; /* should be BUG() ? */
1991 break;
1992 }
1993 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001994}
1995
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08001996static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1997 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1998{
1999 struct cgroup *cgroup;
2000 unsigned long long min_limit, min_memsw_limit, tmp;
2001
2002 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2003 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2004 cgroup = memcg->css.cgroup;
2005 if (!memcg->use_hierarchy)
2006 goto out;
2007
2008 while (cgroup->parent) {
2009 cgroup = cgroup->parent;
2010 memcg = mem_cgroup_from_cont(cgroup);
2011 if (!memcg->use_hierarchy)
2012 break;
2013 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
2014 min_limit = min(min_limit, tmp);
2015 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2016 min_memsw_limit = min(min_memsw_limit, tmp);
2017 }
2018out:
2019 *mem_limit = min_limit;
2020 *memsw_limit = min_memsw_limit;
2021 return;
2022}
2023
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002024static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002025{
2026 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002027 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002028
2029 mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002030 type = MEMFILE_TYPE(event);
2031 name = MEMFILE_ATTR(event);
2032 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002033 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002034 if (type == _MEM)
2035 res_counter_reset_max(&mem->res);
2036 else
2037 res_counter_reset_max(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002038 break;
2039 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002040 if (type == _MEM)
2041 res_counter_reset_failcnt(&mem->res);
2042 else
2043 res_counter_reset_failcnt(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002044 break;
2045 }
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07002046 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002047}
2048
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002049
2050/* For read statistics */
2051enum {
2052 MCS_CACHE,
2053 MCS_RSS,
2054 MCS_PGPGIN,
2055 MCS_PGPGOUT,
2056 MCS_INACTIVE_ANON,
2057 MCS_ACTIVE_ANON,
2058 MCS_INACTIVE_FILE,
2059 MCS_ACTIVE_FILE,
2060 MCS_UNEVICTABLE,
2061 NR_MCS_STAT,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002062};
2063
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002064struct mcs_total_stat {
2065 s64 stat[NR_MCS_STAT];
2066};
2067
2068struct {
2069 char *local_name;
2070 char *total_name;
2071} memcg_stat_strings[NR_MCS_STAT] = {
2072 {"cache", "total_cache"},
2073 {"rss", "total_rss"},
2074 {"pgpgin", "total_pgpgin"},
2075 {"pgpgout", "total_pgpgout"},
2076 {"inactive_anon", "total_inactive_anon"},
2077 {"active_anon", "total_active_anon"},
2078 {"inactive_file", "total_inactive_file"},
2079 {"active_file", "total_active_file"},
2080 {"unevictable", "total_unevictable"}
2081};
2082
2083
2084static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
2085{
2086 struct mcs_total_stat *s = data;
2087 s64 val;
2088
2089 /* per cpu stat */
2090 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE);
2091 s->stat[MCS_CACHE] += val * PAGE_SIZE;
2092 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
2093 s->stat[MCS_RSS] += val * PAGE_SIZE;
2094 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT);
2095 s->stat[MCS_PGPGIN] += val;
2096 val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT);
2097 s->stat[MCS_PGPGOUT] += val;
2098
2099 /* per zone stat */
2100 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
2101 s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
2102 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
2103 s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
2104 val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
2105 s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
2106 val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
2107 s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
2108 val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
2109 s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
2110 return 0;
2111}
2112
2113static void
2114mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
2115{
2116 mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
2117}
2118
Paul Menagec64745c2008-04-29 01:00:02 -07002119static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
2120 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002121{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002122 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002123 struct mcs_total_stat mystat;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002124 int i;
2125
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002126 memset(&mystat, 0, sizeof(mystat));
2127 mem_cgroup_get_local_stat(mem_cont, &mystat);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002128
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002129 for (i = 0; i < NR_MCS_STAT; i++)
2130 cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002131
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002132 /* Hierarchical information */
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08002133 {
2134 unsigned long long limit, memsw_limit;
2135 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
2136 cb->fill(cb, "hierarchical_memory_limit", limit);
2137 if (do_swap_account)
2138 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
2139 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002140
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07002141 memset(&mystat, 0, sizeof(mystat));
2142 mem_cgroup_get_total_stat(mem_cont, &mystat);
2143 for (i = 0; i < NR_MCS_STAT; i++)
2144 cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
2145
2146
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002147#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08002148 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08002149
2150 {
2151 int nid, zid;
2152 struct mem_cgroup_per_zone *mz;
2153 unsigned long recent_rotated[2] = {0, 0};
2154 unsigned long recent_scanned[2] = {0, 0};
2155
2156 for_each_online_node(nid)
2157 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2158 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
2159
2160 recent_rotated[0] +=
2161 mz->reclaim_stat.recent_rotated[0];
2162 recent_rotated[1] +=
2163 mz->reclaim_stat.recent_rotated[1];
2164 recent_scanned[0] +=
2165 mz->reclaim_stat.recent_scanned[0];
2166 recent_scanned[1] +=
2167 mz->reclaim_stat.recent_scanned[1];
2168 }
2169 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
2170 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
2171 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
2172 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
2173 }
2174#endif
2175
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002176 return 0;
2177}
2178
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002179static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
2180{
2181 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2182
2183 return get_swappiness(memcg);
2184}
2185
2186static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
2187 u64 val)
2188{
2189 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
2190 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08002191
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002192 if (val > 100)
2193 return -EINVAL;
2194
2195 if (cgrp->parent == NULL)
2196 return -EINVAL;
2197
2198 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08002199
2200 cgroup_lock();
2201
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002202 /* If under hierarchy, only empty-root can set this value */
2203 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08002204 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2205 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002206 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08002207 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002208
2209 spin_lock(&memcg->reclaim_param_lock);
2210 memcg->swappiness = val;
2211 spin_unlock(&memcg->reclaim_param_lock);
2212
Li Zefan068b38c2009-01-15 13:51:26 -08002213 cgroup_unlock();
2214
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002215 return 0;
2216}
2217
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002218
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002219static struct cftype mem_cgroup_files[] = {
2220 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002221 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002222 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07002223 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002224 },
2225 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002226 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002227 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002228 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002229 .read_u64 = mem_cgroup_read,
2230 },
2231 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002232 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002233 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07002234 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07002235 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002236 },
2237 {
2238 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002239 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002240 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07002241 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002242 },
Balbir Singh8697d332008-02-07 00:13:59 -08002243 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002244 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07002245 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002246 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002247 {
2248 .name = "force_empty",
2249 .trigger = mem_cgroup_force_empty_write,
2250 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08002251 {
2252 .name = "use_hierarchy",
2253 .write_u64 = mem_cgroup_hierarchy_write,
2254 .read_u64 = mem_cgroup_hierarchy_read,
2255 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002256 {
2257 .name = "swappiness",
2258 .read_u64 = mem_cgroup_swappiness_read,
2259 .write_u64 = mem_cgroup_swappiness_write,
2260 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002261};
2262
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002263#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2264static struct cftype memsw_cgroup_files[] = {
2265 {
2266 .name = "memsw.usage_in_bytes",
2267 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2268 .read_u64 = mem_cgroup_read,
2269 },
2270 {
2271 .name = "memsw.max_usage_in_bytes",
2272 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2273 .trigger = mem_cgroup_reset,
2274 .read_u64 = mem_cgroup_read,
2275 },
2276 {
2277 .name = "memsw.limit_in_bytes",
2278 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2279 .write_string = mem_cgroup_write,
2280 .read_u64 = mem_cgroup_read,
2281 },
2282 {
2283 .name = "memsw.failcnt",
2284 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2285 .trigger = mem_cgroup_reset,
2286 .read_u64 = mem_cgroup_read,
2287 },
2288};
2289
2290static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2291{
2292 if (!do_swap_account)
2293 return 0;
2294 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2295 ARRAY_SIZE(memsw_cgroup_files));
2296};
2297#else
2298static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2299{
2300 return 0;
2301}
2302#endif
2303
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002304static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2305{
2306 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002307 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002308 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002309 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002310 /*
2311 * This routine is called against possible nodes.
2312 * But it's BUG to call kmalloc() against offline node.
2313 *
2314 * TODO: this routine can waste much memory for nodes which will
2315 * never be onlined. It's better to use memory hotplug callback
2316 * function.
2317 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002318 if (!node_state(node, N_NORMAL_MEMORY))
2319 tmp = -1;
2320 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002321 if (!pn)
2322 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002323
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002324 mem->info.nodeinfo[node] = pn;
2325 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002326
2327 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2328 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07002329 for_each_lru(l)
2330 INIT_LIST_HEAD(&mz->lists[l]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002331 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002332 return 0;
2333}
2334
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002335static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2336{
2337 kfree(mem->info.nodeinfo[node]);
2338}
2339
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002340static int mem_cgroup_size(void)
2341{
2342 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2343 return sizeof(struct mem_cgroup) + cpustat_size;
2344}
2345
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002346static struct mem_cgroup *mem_cgroup_alloc(void)
2347{
2348 struct mem_cgroup *mem;
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002349 int size = mem_cgroup_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002350
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002351 if (size < PAGE_SIZE)
2352 mem = kmalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002353 else
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002354 mem = vmalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002355
2356 if (mem)
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002357 memset(mem, 0, size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002358 return mem;
2359}
2360
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002361/*
2362 * At destroying mem_cgroup, references from swap_cgroup can remain.
2363 * (scanning all at force_empty is too costly...)
2364 *
2365 * Instead of clearing all references at force_empty, we remember
2366 * the number of reference from swap_cgroup and free mem_cgroup when
2367 * it goes down to 0.
2368 *
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002369 * Removal of cgroup itself succeeds regardless of refs from swap.
2370 */
2371
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002372static void __mem_cgroup_free(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002373{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002374 int node;
2375
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002376 free_css_id(&mem_cgroup_subsys, &mem->css);
2377
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002378 for_each_node_state(node, N_POSSIBLE)
2379 free_mem_cgroup_per_zone_info(mem, node);
2380
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002381 if (mem_cgroup_size() < PAGE_SIZE)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002382 kfree(mem);
2383 else
2384 vfree(mem);
2385}
2386
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002387static void mem_cgroup_get(struct mem_cgroup *mem)
2388{
2389 atomic_inc(&mem->refcnt);
2390}
2391
2392static void mem_cgroup_put(struct mem_cgroup *mem)
2393{
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002394 if (atomic_dec_and_test(&mem->refcnt)) {
2395 struct mem_cgroup *parent = parent_mem_cgroup(mem);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002396 __mem_cgroup_free(mem);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002397 if (parent)
2398 mem_cgroup_put(parent);
2399 }
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002400}
2401
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002402/*
2403 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
2404 */
2405static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
2406{
2407 if (!mem->res.parent)
2408 return NULL;
2409 return mem_cgroup_from_res_counter(mem->res.parent, res);
2410}
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002411
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002412#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2413static void __init enable_swap_cgroup(void)
2414{
Hirokazu Takahashif8d665422009-01-07 18:08:02 -08002415 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002416 do_swap_account = 1;
2417}
2418#else
2419static void __init enable_swap_cgroup(void)
2420{
2421}
2422#endif
2423
Li Zefan0eb253e2009-01-15 13:51:25 -08002424static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002425mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2426{
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002427 struct mem_cgroup *mem, *parent;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002428 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002429 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002430
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002431 mem = mem_cgroup_alloc();
2432 if (!mem)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002433 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08002434
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002435 for_each_node_state(node, N_POSSIBLE)
2436 if (alloc_mem_cgroup_per_zone_info(mem, node))
2437 goto free_out;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002438 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002439 if (cont->parent == NULL) {
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002440 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002441 parent = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002442 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002443 parent = mem_cgroup_from_cont(cont->parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002444 mem->use_hierarchy = parent->use_hierarchy;
2445 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002446
Balbir Singh18f59ea2009-01-07 18:08:07 -08002447 if (parent && parent->use_hierarchy) {
2448 res_counter_init(&mem->res, &parent->res);
2449 res_counter_init(&mem->memsw, &parent->memsw);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08002450 /*
2451 * We increment refcnt of the parent to ensure that we can
2452 * safely access it on res_counter_charge/uncharge.
2453 * This refcnt will be decremented when freeing this
2454 * mem_cgroup(see mem_cgroup_put).
2455 */
2456 mem_cgroup_get(parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002457 } else {
2458 res_counter_init(&mem->res, NULL);
2459 res_counter_init(&mem->memsw, NULL);
2460 }
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002461 mem->last_scanned_child = 0;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -08002462 spin_lock_init(&mem->reclaim_param_lock);
Balbir Singh6d61ef42009-01-07 18:08:06 -08002463
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002464 if (parent)
2465 mem->swappiness = get_swappiness(parent);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002466 atomic_set(&mem->refcnt, 1);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002467 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002468free_out:
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002469 __mem_cgroup_free(mem);
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002470 return ERR_PTR(error);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002471}
2472
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002473static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002474 struct cgroup *cont)
2475{
2476 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07002477
2478 return mem_cgroup_force_empty(mem, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002479}
2480
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002481static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2482 struct cgroup *cont)
2483{
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002484 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002485
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002486 mem_cgroup_put(mem);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002487}
2488
2489static int mem_cgroup_populate(struct cgroup_subsys *ss,
2490 struct cgroup *cont)
2491{
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002492 int ret;
2493
2494 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2495 ARRAY_SIZE(mem_cgroup_files));
2496
2497 if (!ret)
2498 ret = register_memsw_files(cont, ss);
2499 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002500}
2501
Balbir Singh67e465a2008-02-07 00:13:54 -08002502static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2503 struct cgroup *cont,
2504 struct cgroup *old_cont,
2505 struct task_struct *p)
2506{
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002507 mutex_lock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002508 /*
Nikanth Karthikesanf9717d22009-01-07 18:08:11 -08002509 * FIXME: It's better to move charges of this process from old
2510 * memcg to new memcg. But it's just on TODO-List now.
Balbir Singh67e465a2008-02-07 00:13:54 -08002511 */
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002512 mutex_unlock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002513}
2514
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002515struct cgroup_subsys mem_cgroup_subsys = {
2516 .name = "memory",
2517 .subsys_id = mem_cgroup_subsys_id,
2518 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002519 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002520 .destroy = mem_cgroup_destroy,
2521 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08002522 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002523 .early_init = 0,
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07002524 .use_id = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002525};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002526
2527#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2528
2529static int __init disable_swap_account(char *s)
2530{
2531 really_do_swap_account = 0;
2532 return 1;
2533}
2534__setup("noswapaccount", disable_swap_account);
2535#endif