blob: 4d0ea3ceba6d2e40431f64f7e8d8c0f3c0022165 [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>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080030#include <linux/mutex.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070031#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/swap.h>
33#include <linux/spinlock.h>
34#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080035#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070036#include <linux/vmalloc.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070037#include <linux/mm_inline.h>
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -070038#include <linux/page_cgroup.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080039#include "internal.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080040
Balbir Singh8697d332008-02-07 00:13:59 -080041#include <asm/uaccess.h>
42
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070043struct cgroup_subsys mem_cgroup_subsys __read_mostly;
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070044#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080045
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080046#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
47/* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */
48int do_swap_account __read_mostly;
49static int really_do_swap_account __initdata = 1; /* for remember boot option*/
50#else
51#define do_swap_account (0)
52#endif
53
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -080054static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080055
Balbir Singh8cdea7c2008-02-07 00:13:50 -080056/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080057 * Statistics for memory cgroup.
58 */
59enum mem_cgroup_stat_index {
60 /*
61 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
62 */
63 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
64 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
Balaji Rao55e462b2008-05-01 04:35:12 -070065 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
66 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080067
68 MEM_CGROUP_STAT_NSTATS,
69};
70
71struct mem_cgroup_stat_cpu {
72 s64 count[MEM_CGROUP_STAT_NSTATS];
73} ____cacheline_aligned_in_smp;
74
75struct mem_cgroup_stat {
Jan Blunckc8dad2b2009-01-07 18:07:53 -080076 struct mem_cgroup_stat_cpu cpustat[0];
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080077};
78
79/*
80 * For accounting under irq disable, no need for increment preempt count.
81 */
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070082static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat,
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080083 enum mem_cgroup_stat_index idx, int val)
84{
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -070085 stat->count[idx] += val;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080086}
87
88static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
89 enum mem_cgroup_stat_index idx)
90{
91 int cpu;
92 s64 ret = 0;
93 for_each_possible_cpu(cpu)
94 ret += stat->cpustat[cpu].count[idx];
95 return ret;
96}
97
98/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -080099 * per-zone information in memory controller.
100 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800101struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800102 /*
103 * spin_lock to protect the per cgroup LRU
104 */
Christoph Lameterb69408e2008-10-18 20:26:14 -0700105 struct list_head lists[NR_LRU_LISTS];
106 unsigned long count[NR_LRU_LISTS];
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800107
108 struct zone_reclaim_stat reclaim_stat;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800109};
110/* Macro for accessing counter */
111#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
112
113struct mem_cgroup_per_node {
114 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
115};
116
117struct mem_cgroup_lru_info {
118 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
119};
120
121/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800122 * The memory controller data structure. The memory controller controls both
123 * page cache and RSS per cgroup. We would eventually like to provide
124 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
125 * to help the administrator determine what knobs to tune.
126 *
127 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800128 * we hit the water mark. May be even add a low water mark, such that
129 * no reclaim occurs from a cgroup at it's low water mark, this is
130 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800131 */
132struct mem_cgroup {
133 struct cgroup_subsys_state css;
134 /*
135 * the counter to account for memory usage
136 */
137 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800138 /*
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800139 * the counter to account for mem+swap usage.
140 */
141 struct res_counter memsw;
142 /*
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800143 * Per cgroup active and inactive list, similar to the
144 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800145 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800146 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800147
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800148 /*
149 protect against reclaim related member.
150 */
151 spinlock_t reclaim_param_lock;
152
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800153 int prev_priority; /* for recording reclaim priority */
Balbir Singh6d61ef42009-01-07 18:08:06 -0800154
155 /*
156 * While reclaiming in a hiearchy, we cache the last child we
Paul Menage2cb378c2009-01-07 18:08:37 -0800157 * reclaimed from. Protected by hierarchy_mutex
Balbir Singh6d61ef42009-01-07 18:08:06 -0800158 */
159 struct mem_cgroup *last_scanned_child;
Balbir Singh18f59ea2009-01-07 18:08:07 -0800160 /*
161 * Should the accounting and control be hierarchical, per subtree?
162 */
163 bool use_hierarchy;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800164 unsigned long last_oom_jiffies;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800165 atomic_t refcnt;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800166
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800167 unsigned int swappiness;
168
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800169 /*
Jan Blunckc8dad2b2009-01-07 18:07:53 -0800170 * statistics. This must be placed at the end of memcg.
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800171 */
172 struct mem_cgroup_stat stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800173};
174
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800175enum charge_type {
176 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
177 MEM_CGROUP_CHARGE_TYPE_MAPPED,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700178 MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700179 MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800180 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700181 NR_CHARGE_TYPE,
182};
183
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700184/* only for here (for easy reading.) */
185#define PCGF_CACHE (1UL << PCG_CACHE)
186#define PCGF_USED (1UL << PCG_USED)
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700187#define PCGF_LOCK (1UL << PCG_LOCK)
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700188static const unsigned long
189pcg_default_flags[NR_CHARGE_TYPE] = {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800190 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
191 PCGF_USED | PCGF_LOCK, /* Anon */
192 PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700193 0, /* FORCE */
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800194};
195
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800196/* for encoding cft->private value on file */
197#define _MEM (0)
198#define _MEMSWAP (1)
199#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
200#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
201#define MEMFILE_ATTR(val) ((val) & 0xffff)
202
203static void mem_cgroup_get(struct mem_cgroup *mem);
204static void mem_cgroup_put(struct mem_cgroup *mem);
205
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700206static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
207 struct page_cgroup *pc,
208 bool charge)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800209{
210 int val = (charge)? 1 : -1;
211 struct mem_cgroup_stat *stat = &mem->stat;
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700212 struct mem_cgroup_stat_cpu *cpustat;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800213 int cpu = get_cpu();
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800214
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800215 cpustat = &stat->cpustat[cpu];
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700216 if (PageCgroupCache(pc))
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700217 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800218 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700219 __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val);
Balaji Rao55e462b2008-05-01 04:35:12 -0700220
221 if (charge)
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700222 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700223 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
224 else
KAMEZAWA Hiroyukiaddb9ef2008-10-18 20:28:10 -0700225 __mem_cgroup_stat_add_safe(cpustat,
Balaji Rao55e462b2008-05-01 04:35:12 -0700226 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800227 put_cpu();
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800228}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800229
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800230static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800231mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
232{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800233 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
234}
235
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800236static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800237page_cgroup_zoneinfo(struct page_cgroup *pc)
238{
239 struct mem_cgroup *mem = pc->mem_cgroup;
240 int nid = page_cgroup_nid(pc);
241 int zid = page_cgroup_zid(pc);
242
KOSAKI Motohiro54992762009-01-07 18:08:18 -0800243 if (!mem)
244 return NULL;
245
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800246 return mem_cgroup_zoneinfo(mem, nid, zid);
247}
248
249static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
Christoph Lameterb69408e2008-10-18 20:26:14 -0700250 enum lru_list idx)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800251{
252 int nid, zid;
253 struct mem_cgroup_per_zone *mz;
254 u64 total = 0;
255
256 for_each_online_node(nid)
257 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
258 mz = mem_cgroup_zoneinfo(mem, nid, zid);
259 total += MEM_CGROUP_ZSTAT(mz, idx);
260 }
261 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800262}
263
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800264static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800265{
266 return container_of(cgroup_subsys_state(cont,
267 mem_cgroup_subsys_id), struct mem_cgroup,
268 css);
269}
270
Balbir Singhcf475ad2008-04-29 01:00:16 -0700271struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800272{
Balbir Singh31a78f22008-09-28 23:09:31 +0100273 /*
274 * mm_update_next_owner() may clear mm->owner to NULL
275 * if it races with swapoff, page migration, etc.
276 * So this can be called with p == NULL.
277 */
278 if (unlikely(!p))
279 return NULL;
280
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800281 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
282 struct mem_cgroup, css);
283}
284
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800285static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
286{
287 struct mem_cgroup *mem = NULL;
288 /*
289 * Because we have no locks, mm->owner's may be being moved to other
290 * cgroup. We use css_tryget() here even if this looks
291 * pessimistic (rather than adding locks here).
292 */
293 rcu_read_lock();
294 do {
295 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
296 if (unlikely(!mem))
297 break;
298 } while (!css_tryget(&mem->css));
299 rcu_read_unlock();
300 return mem;
301}
302
303static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem)
304{
305 if (!mem)
306 return true;
307 return css_is_removed(&mem->css);
308}
309
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800310/*
311 * Following LRU functions are allowed to be used without PCG_LOCK.
312 * Operations are called by routine of global LRU independently from memcg.
313 * What we have to take care of here is validness of pc->mem_cgroup.
314 *
315 * Changes to pc->mem_cgroup happens when
316 * 1. charge
317 * 2. moving account
318 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
319 * It is added to LRU before charge.
320 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
321 * When moving account, the page is not on LRU. It's isolated.
322 */
323
324void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800325{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800326 struct page_cgroup *pc;
327 struct mem_cgroup *mem;
328 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700329
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800330 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800331 return;
332 pc = lookup_page_cgroup(page);
333 /* can happen while we handle swapcache. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800334 if (list_empty(&pc->lru) || !pc->mem_cgroup)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800335 return;
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800336 /*
337 * We don't check PCG_USED bit. It's cleared when the "page" is finally
338 * removed from global LRU.
339 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800340 mz = page_cgroup_zoneinfo(pc);
341 mem = pc->mem_cgroup;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700342 MEM_CGROUP_ZSTAT(mz, lru) -= 1;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800343 list_del_init(&pc->lru);
344 return;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800345}
346
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800347void mem_cgroup_del_lru(struct page *page)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800348{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800349 mem_cgroup_del_lru_list(page, page_lru(page));
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800350}
351
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800352void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
Balbir Singh66e17072008-02-07 00:13:56 -0800353{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800354 struct mem_cgroup_per_zone *mz;
355 struct page_cgroup *pc;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800356
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800357 if (mem_cgroup_disabled())
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700358 return;
Christoph Lameterb69408e2008-10-18 20:26:14 -0700359
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800360 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800361 /*
362 * Used bit is set without atomic ops but after smp_wmb().
363 * For making pc->mem_cgroup visible, insert smp_rmb() here.
364 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800365 smp_rmb();
366 /* unused page is not rotated. */
367 if (!PageCgroupUsed(pc))
368 return;
369 mz = page_cgroup_zoneinfo(pc);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700370 list_move(&pc->lru, &mz->lists[lru]);
Balbir Singh66e17072008-02-07 00:13:56 -0800371}
372
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800373void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
374{
375 struct page_cgroup *pc;
376 struct mem_cgroup_per_zone *mz;
377
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800378 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800379 return;
380 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800381 /*
382 * Used bit is set without atomic ops but after smp_wmb().
383 * For making pc->mem_cgroup visible, insert smp_rmb() here.
384 */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800385 smp_rmb();
386 if (!PageCgroupUsed(pc))
387 return;
388
389 mz = page_cgroup_zoneinfo(pc);
390 MEM_CGROUP_ZSTAT(mz, lru) += 1;
391 list_add(&pc->lru, &mz->lists[lru]);
392}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800393
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800394/*
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800395 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
396 * lru because the page may.be reused after it's fully uncharged (because of
397 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
398 * it again. This function is only used to charge SwapCache. It's done under
399 * lock_page and expected that zone->lru_lock is never held.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800400 */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800401static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800402{
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800403 unsigned long flags;
404 struct zone *zone = page_zone(page);
405 struct page_cgroup *pc = lookup_page_cgroup(page);
406
407 spin_lock_irqsave(&zone->lru_lock, flags);
408 /*
409 * Forget old LRU when this page_cgroup is *not* used. This Used bit
410 * is guarded by lock_page() because the page is SwapCache.
411 */
412 if (!PageCgroupUsed(pc))
413 mem_cgroup_del_lru_list(page, page_lru(page));
414 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800415}
416
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -0800417static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
418{
419 unsigned long flags;
420 struct zone *zone = page_zone(page);
421 struct page_cgroup *pc = lookup_page_cgroup(page);
422
423 spin_lock_irqsave(&zone->lru_lock, flags);
424 /* link when the page is linked to LRU but page_cgroup isn't */
425 if (PageLRU(page) && list_empty(&pc->lru))
426 mem_cgroup_add_lru_list(page, page_lru(page));
427 spin_unlock_irqrestore(&zone->lru_lock, flags);
428}
429
430
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800431void mem_cgroup_move_lists(struct page *page,
432 enum lru_list from, enum lru_list to)
433{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -0800434 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800435 return;
436 mem_cgroup_del_lru_list(page, from);
437 mem_cgroup_add_lru_list(page, to);
438}
439
David Rientjes4c4a2212008-02-07 00:14:06 -0800440int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
441{
442 int ret;
443
444 task_lock(task);
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800445 ret = task->mm && mm_match_cgroup(task->mm, mem);
David Rientjes4c4a2212008-02-07 00:14:06 -0800446 task_unlock(task);
447 return ret;
448}
449
Balbir Singh66e17072008-02-07 00:13:56 -0800450/*
KAMEZAWA Hiroyuki58ae83d2008-02-07 00:14:32 -0800451 * Calculate mapped_ratio under memory controller. This will be used in
452 * vmscan.c for deteremining we have to reclaim mapped pages.
453 */
454int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
455{
456 long total, rss;
457
458 /*
459 * usage is recorded in bytes. But, here, we assume the number of
460 * physical pages can be represented by "long" on any arch.
461 */
462 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
463 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
464 return (int)((rss * 100L) / total);
465}
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800466
KAMEZAWA Hiroyuki5932f362008-02-07 00:14:33 -0800467/*
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800468 * prev_priority control...this will be used in memory reclaim path.
469 */
470int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
471{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800472 int prev_priority;
473
474 spin_lock(&mem->reclaim_param_lock);
475 prev_priority = mem->prev_priority;
476 spin_unlock(&mem->reclaim_param_lock);
477
478 return prev_priority;
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800479}
480
481void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
482{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800483 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800484 if (priority < mem->prev_priority)
485 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800486 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800487}
488
489void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
490{
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800491 spin_lock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800492 mem->prev_priority = priority;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -0800493 spin_unlock(&mem->reclaim_param_lock);
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800494}
495
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800496static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800497{
498 unsigned long active;
499 unsigned long inactive;
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800500 unsigned long gb;
501 unsigned long inactive_ratio;
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800502
503 inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
504 active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
505
KOSAKI Motohiroc772be92009-01-07 18:08:25 -0800506 gb = (inactive + active) >> (30 - PAGE_SHIFT);
507 if (gb)
508 inactive_ratio = int_sqrt(10 * gb);
509 else
510 inactive_ratio = 1;
511
512 if (present_pages) {
513 present_pages[0] = inactive;
514 present_pages[1] = active;
515 }
516
517 return inactive_ratio;
518}
519
520int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
521{
522 unsigned long active;
523 unsigned long inactive;
524 unsigned long present_pages[2];
525 unsigned long inactive_ratio;
526
527 inactive_ratio = calc_inactive_ratio(memcg, present_pages);
528
529 inactive = present_pages[0];
530 active = present_pages[1];
531
532 if (inactive * inactive_ratio < active)
KOSAKI Motohiro14797e22009-01-07 18:08:18 -0800533 return 1;
534
535 return 0;
536}
537
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800538unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
539 struct zone *zone,
540 enum lru_list lru)
541{
542 int nid = zone->zone_pgdat->node_id;
543 int zid = zone_idx(zone);
544 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
545
546 return MEM_CGROUP_ZSTAT(mz, lru);
547}
548
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800549struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
550 struct zone *zone)
551{
552 int nid = zone->zone_pgdat->node_id;
553 int zid = zone_idx(zone);
554 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
555
556 return &mz->reclaim_stat;
557}
558
559struct zone_reclaim_stat *
560mem_cgroup_get_reclaim_stat_from_page(struct page *page)
561{
562 struct page_cgroup *pc;
563 struct mem_cgroup_per_zone *mz;
564
565 if (mem_cgroup_disabled())
566 return NULL;
567
568 pc = lookup_page_cgroup(page);
Daisuke Nishimurabd112db2009-01-15 13:51:11 -0800569 /*
570 * Used bit is set without atomic ops but after smp_wmb().
571 * For making pc->mem_cgroup visible, insert smp_rmb() here.
572 */
573 smp_rmb();
574 if (!PageCgroupUsed(pc))
575 return NULL;
576
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800577 mz = page_cgroup_zoneinfo(pc);
578 if (!mz)
579 return NULL;
580
581 return &mz->reclaim_stat;
582}
583
Balbir Singh66e17072008-02-07 00:13:56 -0800584unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
585 struct list_head *dst,
586 unsigned long *scanned, int order,
587 int mode, struct zone *z,
588 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700589 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800590{
591 unsigned long nr_taken = 0;
592 struct page *page;
593 unsigned long scan;
594 LIST_HEAD(pc_list);
595 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800596 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800597 int nid = z->zone_pgdat->node_id;
598 int zid = zone_idx(z);
599 struct mem_cgroup_per_zone *mz;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700600 int lru = LRU_FILE * !!file + !!active;
Balbir Singh66e17072008-02-07 00:13:56 -0800601
Balbir Singhcf475ad2008-04-29 01:00:16 -0700602 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800603 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -0700604 src = &mz->lists[lru];
Balbir Singh66e17072008-02-07 00:13:56 -0800605
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800606 scan = 0;
607 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800608 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800609 break;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800610
611 page = pc->page;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700612 if (unlikely(!PageCgroupUsed(pc)))
613 continue;
Hugh Dickins436c65412008-02-07 00:14:12 -0800614 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800615 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800616
Hugh Dickins436c65412008-02-07 00:14:12 -0800617 scan++;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700618 if (__isolate_lru_page(page, mode, file) == 0) {
Balbir Singh66e17072008-02-07 00:13:56 -0800619 list_move(&page->lru, dst);
620 nr_taken++;
621 }
622 }
623
Balbir Singh66e17072008-02-07 00:13:56 -0800624 *scanned = scan;
625 return nr_taken;
626}
627
Balbir Singh6d61ef42009-01-07 18:08:06 -0800628#define mem_cgroup_from_res_counter(counter, member) \
629 container_of(counter, struct mem_cgroup, member)
630
631/*
632 * This routine finds the DFS walk successor. This routine should be
Paul Menage2cb378c2009-01-07 18:08:37 -0800633 * called with hierarchy_mutex held
Balbir Singh6d61ef42009-01-07 18:08:06 -0800634 */
635static struct mem_cgroup *
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800636__mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem)
Balbir Singh6d61ef42009-01-07 18:08:06 -0800637{
638 struct cgroup *cgroup, *curr_cgroup, *root_cgroup;
639
640 curr_cgroup = curr->css.cgroup;
641 root_cgroup = root_mem->css.cgroup;
642
643 if (!list_empty(&curr_cgroup->children)) {
644 /*
645 * Walk down to children
646 */
Balbir Singh6d61ef42009-01-07 18:08:06 -0800647 cgroup = list_entry(curr_cgroup->children.next,
648 struct cgroup, sibling);
649 curr = mem_cgroup_from_cont(cgroup);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800650 goto done;
651 }
652
653visit_parent:
654 if (curr_cgroup == root_cgroup) {
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800655 /* caller handles NULL case */
656 curr = NULL;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800657 goto done;
658 }
659
660 /*
661 * Goto next sibling
662 */
663 if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) {
Balbir Singh6d61ef42009-01-07 18:08:06 -0800664 cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup,
665 sibling);
666 curr = mem_cgroup_from_cont(cgroup);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800667 goto done;
668 }
669
670 /*
671 * Go up to next parent and next parent's sibling if need be
672 */
673 curr_cgroup = curr_cgroup->parent;
674 goto visit_parent;
675
676done:
Balbir Singh6d61ef42009-01-07 18:08:06 -0800677 return curr;
678}
679
680/*
681 * Visit the first child (need not be the first child as per the ordering
682 * of the cgroup list, since we track last_scanned_child) of @mem and use
683 * that to reclaim free pages from.
684 */
685static struct mem_cgroup *
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800686mem_cgroup_get_next_node(struct mem_cgroup *root_mem)
Balbir Singh6d61ef42009-01-07 18:08:06 -0800687{
688 struct cgroup *cgroup;
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800689 struct mem_cgroup *orig, *next;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800690 bool obsolete;
691
Balbir Singh6d61ef42009-01-07 18:08:06 -0800692 /*
693 * Scan all children under the mem_cgroup mem
694 */
Paul Menage2cb378c2009-01-07 18:08:37 -0800695 mutex_lock(&mem_cgroup_subsys.hierarchy_mutex);
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800696
697 orig = root_mem->last_scanned_child;
698 obsolete = mem_cgroup_is_obsolete(orig);
699
Balbir Singh6d61ef42009-01-07 18:08:06 -0800700 if (list_empty(&root_mem->css.cgroup->children)) {
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800701 /*
702 * root_mem might have children before and last_scanned_child
703 * may point to one of them. We put it later.
704 */
705 if (orig)
706 VM_BUG_ON(!obsolete);
707 next = NULL;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800708 goto done;
709 }
710
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800711 if (!orig || obsolete) {
Balbir Singh6d61ef42009-01-07 18:08:06 -0800712 cgroup = list_first_entry(&root_mem->css.cgroup->children,
713 struct cgroup, sibling);
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800714 next = mem_cgroup_from_cont(cgroup);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800715 } else
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800716 next = __mem_cgroup_get_next_node(orig, root_mem);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800717
718done:
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800719 if (next)
720 mem_cgroup_get(next);
721 root_mem->last_scanned_child = next;
722 if (orig)
723 mem_cgroup_put(orig);
Paul Menage2cb378c2009-01-07 18:08:37 -0800724 mutex_unlock(&mem_cgroup_subsys.hierarchy_mutex);
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800725 return (next) ? next : root_mem;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800726}
727
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800728static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
729{
730 if (do_swap_account) {
731 if (res_counter_check_under_limit(&mem->res) &&
732 res_counter_check_under_limit(&mem->memsw))
733 return true;
734 } else
735 if (res_counter_check_under_limit(&mem->res))
736 return true;
737 return false;
738}
739
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800740static unsigned int get_swappiness(struct mem_cgroup *memcg)
741{
742 struct cgroup *cgrp = memcg->css.cgroup;
743 unsigned int swappiness;
744
745 /* root ? */
746 if (cgrp->parent == NULL)
747 return vm_swappiness;
748
749 spin_lock(&memcg->reclaim_param_lock);
750 swappiness = memcg->swappiness;
751 spin_unlock(&memcg->reclaim_param_lock);
752
753 return swappiness;
754}
755
Balbir Singh6d61ef42009-01-07 18:08:06 -0800756/*
757 * Dance down the hierarchy if needed to reclaim memory. We remember the
758 * last child we reclaimed from, so that we don't end up penalizing
759 * one child extensively based on its position in the children list.
760 *
761 * root_mem is the original ancestor that we've been reclaim from.
762 */
763static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
764 gfp_t gfp_mask, bool noswap)
765{
766 struct mem_cgroup *next_mem;
767 int ret = 0;
768
769 /*
770 * Reclaim unconditionally and don't check for return value.
771 * We need to reclaim in the current group and down the tree.
772 * One might think about checking for children before reclaiming,
773 * but there might be left over accounting, even after children
774 * have left.
775 */
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800776 ret += try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800777 get_swappiness(root_mem));
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800778 if (mem_cgroup_check_under_limit(root_mem))
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800779 return 1; /* indicate reclaim has succeeded */
Daisuke Nishimura670ec2f2009-01-07 18:08:13 -0800780 if (!root_mem->use_hierarchy)
781 return ret;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800782
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800783 next_mem = mem_cgroup_get_next_node(root_mem);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800784
785 while (next_mem != root_mem) {
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800786 if (mem_cgroup_is_obsolete(next_mem)) {
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800787 next_mem = mem_cgroup_get_next_node(root_mem);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800788 continue;
789 }
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800790 ret += try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -0800791 get_swappiness(next_mem));
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800792 if (mem_cgroup_check_under_limit(root_mem))
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800793 return 1; /* indicate reclaim has succeeded */
Daisuke Nishimurac268e992009-01-15 13:51:13 -0800794 next_mem = mem_cgroup_get_next_node(root_mem);
Balbir Singh6d61ef42009-01-07 18:08:06 -0800795 }
796 return ret;
797}
798
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800799bool mem_cgroup_oom_called(struct task_struct *task)
800{
801 bool ret = false;
802 struct mem_cgroup *mem;
803 struct mm_struct *mm;
804
805 rcu_read_lock();
806 mm = task->mm;
807 if (!mm)
808 mm = &init_mm;
809 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
810 if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10))
811 ret = true;
812 rcu_read_unlock();
813 return ret;
814}
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800815/*
816 * Unlike exported interface, "oom" parameter is added. if oom==true,
817 * oom-killer can be invoked.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800818 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800819static int __mem_cgroup_try_charge(struct mm_struct *mm,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800820 gfp_t gfp_mask, struct mem_cgroup **memcg,
821 bool oom)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800822{
Balbir Singh6d61ef42009-01-07 18:08:06 -0800823 struct mem_cgroup *mem, *mem_over_limit;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800824 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800825 struct res_counter *fail_res;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800826
827 if (unlikely(test_thread_flag(TIF_MEMDIE))) {
828 /* Don't account this! */
829 *memcg = NULL;
830 return 0;
831 }
832
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800833 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800834 * We always charge the cgroup the mm_struct belongs to.
835 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800836 * thread group leader migrates. It's possible that mm is not
837 * set, if so charge the init_mm (happens for pagecache usage).
838 */
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800839 mem = *memcg;
840 if (likely(!mem)) {
841 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800842 *memcg = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700843 } else {
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800844 css_get(&mem->css);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700845 }
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800846 if (unlikely(!mem))
847 return 0;
848
849 VM_BUG_ON(mem_cgroup_is_obsolete(mem));
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800850
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800851 while (1) {
852 int ret;
853 bool noswap = false;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800854
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800855 ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800856 if (likely(!ret)) {
857 if (!do_swap_account)
858 break;
Balbir Singh28dbc4b2009-01-07 18:08:05 -0800859 ret = res_counter_charge(&mem->memsw, PAGE_SIZE,
860 &fail_res);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800861 if (likely(!ret))
862 break;
863 /* mem+swap counter fails */
864 res_counter_uncharge(&mem->res, PAGE_SIZE);
865 noswap = true;
Balbir Singh6d61ef42009-01-07 18:08:06 -0800866 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
867 memsw);
868 } else
869 /* mem counter fails */
870 mem_over_limit = mem_cgroup_from_res_counter(fail_res,
871 res);
872
Hugh Dickins3be91272008-02-07 00:14:19 -0800873 if (!(gfp_mask & __GFP_WAIT))
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800874 goto nomem;
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800875
Balbir Singh6d61ef42009-01-07 18:08:06 -0800876 ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask,
877 noswap);
Daisuke Nishimura4d1c6272009-01-15 13:51:14 -0800878 if (ret)
879 continue;
Balbir Singh66e17072008-02-07 00:13:56 -0800880
881 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800882 * try_to_free_mem_cgroup_pages() might not give us a full
883 * picture of reclaim. Some pages are reclaimed and might be
884 * moved to swap cache or just unmapped from the cgroup.
885 * Check the limit again to see if the reclaim reduced the
886 * current usage of the cgroup before giving up
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800887 *
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800888 */
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -0800889 if (mem_cgroup_check_under_limit(mem_over_limit))
890 continue;
Hugh Dickins3be91272008-02-07 00:14:19 -0800891
892 if (!nr_retries--) {
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800893 if (oom) {
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -0800894 mutex_lock(&memcg_tasklist);
KAMEZAWA Hiroyuki88700752009-01-07 18:08:09 -0800895 mem_cgroup_out_of_memory(mem_over_limit, gfp_mask);
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -0800896 mutex_unlock(&memcg_tasklist);
KAMEZAWA Hiroyuki88700752009-01-07 18:08:09 -0800897 mem_over_limit->last_oom_jiffies = jiffies;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800898 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800899 goto nomem;
Balbir Singh66e17072008-02-07 00:13:56 -0800900 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800901 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800902 return 0;
903nomem:
904 css_put(&mem->css);
905 return -ENOMEM;
906}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800907
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -0800908static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
909{
910 struct mem_cgroup *mem;
911 swp_entry_t ent;
912
913 if (!PageSwapCache(page))
914 return NULL;
915
916 ent.val = page_private(page);
917 mem = lookup_swap_cgroup(ent);
918 if (!mem)
919 return NULL;
920 if (!css_tryget(&mem->css))
921 return NULL;
922 return mem;
923}
924
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800925/*
Daisuke Nishimuraa5e924f2009-01-07 18:08:28 -0800926 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800927 * USED state. If already USED, uncharge and return.
928 */
929
930static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
931 struct page_cgroup *pc,
932 enum charge_type ctype)
933{
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800934 /* try_charge() can return NULL to *memcg, taking care of it. */
935 if (!mem)
936 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700937
938 lock_page_cgroup(pc);
939 if (unlikely(PageCgroupUsed(pc))) {
940 unlock_page_cgroup(pc);
941 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -0800942 if (do_swap_account)
943 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700944 css_put(&mem->css);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -0800945 return;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700946 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800947 pc->mem_cgroup = mem;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800948 smp_wmb();
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700949 pc->flags = pcg_default_flags[ctype];
Hugh Dickins3be91272008-02-07 00:14:19 -0800950
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800951 mem_cgroup_charge_statistics(mem, pc, true);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700952
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -0700953 unlock_page_cgroup(pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800954}
955
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800956/**
957 * mem_cgroup_move_account - move account of the page
958 * @pc: page_cgroup of the page.
959 * @from: mem_cgroup which the page is moved from.
960 * @to: mem_cgroup which the page is moved to. @from != @to.
961 *
962 * The caller must confirm following.
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800963 * - page is not on LRU (isolate_page() is useful.)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800964 *
965 * returns 0 at success,
966 * returns -EBUSY when lock is busy or "pc" is unstable.
967 *
968 * This function does "uncharge" from old cgroup but doesn't do "charge" to
969 * new cgroup. It should be done by a caller.
970 */
971
972static int mem_cgroup_move_account(struct page_cgroup *pc,
973 struct mem_cgroup *from, struct mem_cgroup *to)
974{
975 struct mem_cgroup_per_zone *from_mz, *to_mz;
976 int nid, zid;
977 int ret = -EBUSY;
978
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800979 VM_BUG_ON(from == to);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800980 VM_BUG_ON(PageLRU(pc->page));
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800981
982 nid = page_cgroup_nid(pc);
983 zid = page_cgroup_zid(pc);
984 from_mz = mem_cgroup_zoneinfo(from, nid, zid);
985 to_mz = mem_cgroup_zoneinfo(to, nid, zid);
986
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -0800987 if (!trylock_page_cgroup(pc))
988 return ret;
989
990 if (!PageCgroupUsed(pc))
991 goto out;
992
993 if (pc->mem_cgroup != from)
994 goto out;
995
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800996 res_counter_uncharge(&from->res, PAGE_SIZE);
997 mem_cgroup_charge_statistics(from, pc, false);
998 if (do_swap_account)
999 res_counter_uncharge(&from->memsw, PAGE_SIZE);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001000 css_put(&from->css);
1001
1002 css_get(&to->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001003 pc->mem_cgroup = to;
1004 mem_cgroup_charge_statistics(to, pc, true);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001005 ret = 0;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001006out:
1007 unlock_page_cgroup(pc);
1008 return ret;
1009}
1010
1011/*
1012 * move charges to its parent.
1013 */
1014
1015static int mem_cgroup_move_parent(struct page_cgroup *pc,
1016 struct mem_cgroup *child,
1017 gfp_t gfp_mask)
1018{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001019 struct page *page = pc->page;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001020 struct cgroup *cg = child->css.cgroup;
1021 struct cgroup *pcg = cg->parent;
1022 struct mem_cgroup *parent;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001023 int ret;
1024
1025 /* Is ROOT ? */
1026 if (!pcg)
1027 return -EINVAL;
1028
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001029
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001030 parent = mem_cgroup_from_cont(pcg);
1031
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001032
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001033 ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001034 if (ret || !parent)
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001035 return ret;
1036
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001037 if (!get_page_unless_zero(page)) {
1038 ret = -EBUSY;
1039 goto uncharge;
1040 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001041
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001042 ret = isolate_lru_page(page);
1043
1044 if (ret)
1045 goto cancel;
1046
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001047 ret = mem_cgroup_move_account(pc, child, parent);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001048
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001049 putback_lru_page(page);
1050 if (!ret) {
1051 put_page(page);
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001052 /* drop extra refcnt by try_charge() */
1053 css_put(&parent->css);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001054 return 0;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001055 }
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001056
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001057cancel:
Daisuke Nishimura40d58132009-01-15 13:51:12 -08001058 put_page(page);
1059uncharge:
1060 /* drop extra refcnt by try_charge() */
1061 css_put(&parent->css);
1062 /* uncharge if move fails */
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001063 res_counter_uncharge(&parent->res, PAGE_SIZE);
1064 if (do_swap_account)
1065 res_counter_uncharge(&parent->memsw, PAGE_SIZE);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001066 return ret;
1067}
1068
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001069/*
1070 * Charge the memory controller for page usage.
1071 * Return
1072 * 0 if the charge was successful
1073 * < 0 if the cgroup is over its limit
1074 */
1075static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1076 gfp_t gfp_mask, enum charge_type ctype,
1077 struct mem_cgroup *memcg)
1078{
1079 struct mem_cgroup *mem;
1080 struct page_cgroup *pc;
1081 int ret;
1082
1083 pc = lookup_page_cgroup(page);
1084 /* can happen at boot */
1085 if (unlikely(!pc))
1086 return 0;
1087 prefetchw(pc);
1088
1089 mem = memcg;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001090 ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001091 if (ret || !mem)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001092 return ret;
1093
1094 __mem_cgroup_commit_charge(mem, pc, ctype);
1095 return 0;
1096}
1097
1098int mem_cgroup_newpage_charge(struct page *page,
1099 struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001100{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001101 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001102 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001103 if (PageCompound(page))
1104 return 0;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001105 /*
1106 * If already mapped, we don't have to account.
1107 * If page cache, page->mapping has address_space.
1108 * But page->mapping may have out-of-use anon_vma pointer,
1109 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1110 * is NULL.
1111 */
1112 if (page_mapped(page) || (page->mapping && !PageAnon(page)))
1113 return 0;
1114 if (unlikely(!mm))
1115 mm = &init_mm;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001116 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001117 MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -08001118}
1119
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001120int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
1121 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -08001122{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001123 struct mem_cgroup *mem = NULL;
1124 int ret;
1125
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001126 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001127 return 0;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001128 if (PageCompound(page))
1129 return 0;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001130 /*
1131 * Corner case handling. This is called from add_to_page_cache()
1132 * in usual. But some FS (shmem) precharges this page before calling it
1133 * and call add_to_page_cache() with GFP_NOWAIT.
1134 *
1135 * For GFP_NOWAIT case, the page may be pre-charged before calling
1136 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
1137 * charge twice. (It works but has to pay a bit larger cost.)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001138 * And when the page is SwapCache, it should take swap information
1139 * into account. This is under lock_page() now.
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001140 */
1141 if (!(gfp_mask & __GFP_WAIT)) {
1142 struct page_cgroup *pc;
1143
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001144
1145 pc = lookup_page_cgroup(page);
1146 if (!pc)
1147 return 0;
1148 lock_page_cgroup(pc);
1149 if (PageCgroupUsed(pc)) {
1150 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001151 return 0;
1152 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001153 unlock_page_cgroup(pc);
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001154 }
1155
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001156 if (do_swap_account && PageSwapCache(page)) {
1157 mem = try_get_mem_cgroup_from_swapcache(page);
1158 if (mem)
1159 mm = NULL;
1160 else
1161 mem = NULL;
1162 /* SwapCache may be still linked to LRU now. */
1163 mem_cgroup_lru_del_before_commit_swapcache(page);
1164 }
1165
1166 if (unlikely(!mm && !mem))
Balbir Singh8697d332008-02-07 00:13:59 -08001167 mm = &init_mm;
KAMEZAWA Hiroyukiaccf1632008-07-25 01:47:17 -07001168
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -07001169 if (page_is_file_cache(page))
1170 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001171 MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001172
1173 ret = mem_cgroup_charge_common(page, mm, gfp_mask,
1174 MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
1175 if (mem)
1176 css_put(&mem->css);
1177 if (PageSwapCache(page))
1178 mem_cgroup_lru_add_after_commit_swapcache(page);
1179
1180 if (do_swap_account && !ret && PageSwapCache(page)) {
1181 swp_entry_t ent = {.val = page_private(page)};
1182 /* avoid double counting */
1183 mem = swap_cgroup_record(ent, NULL);
1184 if (mem) {
1185 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1186 mem_cgroup_put(mem);
1187 }
1188 }
1189 return ret;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001190}
1191
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001192/*
1193 * While swap-in, try_charge -> commit or cancel, the page is locked.
1194 * And when try_charge() successfully returns, one refcnt to memcg without
1195 * struct page_cgroup is aquired. This refcnt will be cumsumed by
1196 * "commit()" or removed by "cancel()"
1197 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001198int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
1199 struct page *page,
1200 gfp_t mask, struct mem_cgroup **ptr)
1201{
1202 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001203 int ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001204
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001205 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001206 return 0;
1207
1208 if (!do_swap_account)
1209 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001210 /*
1211 * A racing thread's fault, or swapoff, may have already updated
1212 * the pte, and even removed page from swap cache: return success
1213 * to go on to do_swap_page()'s pte_same() test, which should fail.
1214 */
1215 if (!PageSwapCache(page))
1216 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001217 mem = try_get_mem_cgroup_from_swapcache(page);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001218 if (!mem)
1219 goto charge_cur_mm;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001220 *ptr = mem;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001221 ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
1222 /* drop extra refcnt from tryget */
1223 css_put(&mem->css);
1224 return ret;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001225charge_cur_mm:
1226 if (unlikely(!mm))
1227 mm = &init_mm;
1228 return __mem_cgroup_try_charge(mm, mask, ptr, true);
1229}
1230
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001231void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
1232{
1233 struct page_cgroup *pc;
1234
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001235 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001236 return;
1237 if (!ptr)
1238 return;
1239 pc = lookup_page_cgroup(page);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001240 mem_cgroup_lru_del_before_commit_swapcache(page);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001241 __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001242 mem_cgroup_lru_add_after_commit_swapcache(page);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001243 /*
1244 * Now swap is on-memory. This means this page may be
1245 * counted both as mem and swap....double count.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001246 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
1247 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
1248 * may call delete_from_swap_cache() before reach here.
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001249 */
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08001250 if (do_swap_account && PageSwapCache(page)) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001251 swp_entry_t ent = {.val = page_private(page)};
1252 struct mem_cgroup *memcg;
1253 memcg = swap_cgroup_record(ent, NULL);
1254 if (memcg) {
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001255 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1256 mem_cgroup_put(memcg);
1257 }
1258
1259 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001260 /* add this page(page_cgroup) to the LRU we want. */
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001261
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001262}
1263
1264void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
1265{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001266 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001267 return;
1268 if (!mem)
1269 return;
1270 res_counter_uncharge(&mem->res, PAGE_SIZE);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001271 if (do_swap_account)
1272 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001273 css_put(&mem->css);
1274}
1275
1276
Balbir Singh8697d332008-02-07 00:13:59 -08001277/*
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001278 * uncharge if !page_mapped(page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001279 */
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001280static struct mem_cgroup *
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001281__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001282{
Hugh Dickins82895462008-03-04 14:29:08 -08001283 struct page_cgroup *pc;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001284 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08001285 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001286
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001287 if (mem_cgroup_disabled())
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001288 return NULL;
Balbir Singh40779602008-04-04 14:29:59 -07001289
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001290 if (PageSwapCache(page))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001291 return NULL;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001292
Balbir Singh8697d332008-02-07 00:13:59 -08001293 /*
Balbir Singh3c541e12008-02-07 00:14:41 -08001294 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -08001295 */
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001296 pc = lookup_page_cgroup(page);
1297 if (unlikely(!pc || !PageCgroupUsed(pc)))
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001298 return NULL;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001299
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001300 lock_page_cgroup(pc);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001301
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001302 mem = pc->mem_cgroup;
1303
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001304 if (!PageCgroupUsed(pc))
1305 goto unlock_out;
1306
1307 switch (ctype) {
1308 case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1309 if (page_mapped(page))
1310 goto unlock_out;
1311 break;
1312 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
1313 if (!PageAnon(page)) { /* Shared memory */
1314 if (page->mapping && !page_is_file_cache(page))
1315 goto unlock_out;
1316 } else if (page_mapped(page)) /* Anon */
1317 goto unlock_out;
1318 break;
1319 default:
1320 break;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001321 }
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001322
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001323 res_counter_uncharge(&mem->res, PAGE_SIZE);
1324 if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
1325 res_counter_uncharge(&mem->memsw, PAGE_SIZE);
1326
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001327 mem_cgroup_charge_statistics(mem, pc, false);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001328 ClearPageCgroupUsed(pc);
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001329 /*
1330 * pc->mem_cgroup is not cleared here. It will be accessed when it's
1331 * freed from LRU. This is safe because uncharged page is expected not
1332 * to be reused (freed soon). Exception is SwapCache, it's handled by
1333 * special functions.
1334 */
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001335
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001336 mz = page_cgroup_zoneinfo(pc);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001337 unlock_page_cgroup(pc);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001338
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001339 /* at swapout, this memcg will be accessed to record to swap */
1340 if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
1341 css_put(&mem->css);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001342
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001343 return mem;
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001344
1345unlock_out:
1346 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001347 return NULL;
Balbir Singh3c541e12008-02-07 00:14:41 -08001348}
1349
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001350void mem_cgroup_uncharge_page(struct page *page)
1351{
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001352 /* early check. */
1353 if (page_mapped(page))
1354 return;
1355 if (page->mapping && !PageAnon(page))
1356 return;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001357 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
1358}
1359
1360void mem_cgroup_uncharge_cache_page(struct page *page)
1361{
1362 VM_BUG_ON(page_mapped(page));
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -07001363 VM_BUG_ON(page->mapping);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001364 __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
1365}
1366
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001367/*
1368 * called from __delete_from_swap_cache() and drop "page" account.
1369 * memcg information is recorded to swap_cgroup of "ent"
1370 */
1371void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001372{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001373 struct mem_cgroup *memcg;
1374
1375 memcg = __mem_cgroup_uncharge_common(page,
1376 MEM_CGROUP_CHARGE_TYPE_SWAPOUT);
1377 /* record memcg information */
1378 if (do_swap_account && memcg) {
1379 swap_cgroup_record(ent, memcg);
1380 mem_cgroup_get(memcg);
1381 }
KAMEZAWA Hiroyukia7fe9422009-01-07 18:08:13 -08001382 if (memcg)
1383 css_put(&memcg->css);
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001384}
1385
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001386#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
1387/*
1388 * called from swap_entry_free(). remove record in swap_cgroup and
1389 * uncharge "memsw" account.
1390 */
1391void mem_cgroup_uncharge_swap(swp_entry_t ent)
1392{
1393 struct mem_cgroup *memcg;
1394
1395 if (!do_swap_account)
1396 return;
1397
1398 memcg = swap_cgroup_record(ent, NULL);
1399 if (memcg) {
1400 res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
1401 mem_cgroup_put(memcg);
1402 }
1403}
1404#endif
1405
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001406/*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001407 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
1408 * page belongs to.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001409 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001410int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr)
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001411{
1412 struct page_cgroup *pc;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001413 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001414 int ret = 0;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001415
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001416 if (mem_cgroup_disabled())
Balbir Singh40779602008-04-04 14:29:59 -07001417 return 0;
1418
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001419 pc = lookup_page_cgroup(page);
1420 lock_page_cgroup(pc);
1421 if (PageCgroupUsed(pc)) {
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001422 mem = pc->mem_cgroup;
1423 css_get(&mem->css);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -08001424 }
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001425 unlock_page_cgroup(pc);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001426
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001427 if (mem) {
Daisuke Nishimura3bb4edf2009-01-07 18:08:28 -08001428 ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001429 css_put(&mem->css);
1430 }
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001431 *ptr = mem;
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001432 return ret;
1433}
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -08001434
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001435/* remove redundant charge if migration failed*/
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001436void mem_cgroup_end_migration(struct mem_cgroup *mem,
1437 struct page *oldpage, struct page *newpage)
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -07001438{
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001439 struct page *target, *unused;
1440 struct page_cgroup *pc;
1441 enum charge_type ctype;
1442
1443 if (!mem)
1444 return;
1445
1446 /* at migration success, oldpage->mapping is NULL. */
1447 if (oldpage->mapping) {
1448 target = oldpage;
1449 unused = NULL;
1450 } else {
1451 target = newpage;
1452 unused = oldpage;
1453 }
1454
1455 if (PageAnon(target))
1456 ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
1457 else if (page_is_file_cache(target))
1458 ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
1459 else
1460 ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
1461
1462 /* unused page is not on radix-tree now. */
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001463 if (unused)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001464 __mem_cgroup_uncharge_common(unused, ctype);
1465
1466 pc = lookup_page_cgroup(target);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001467 /*
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001468 * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup.
1469 * So, double-counting is effectively avoided.
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -07001470 */
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -08001471 __mem_cgroup_commit_charge(mem, pc, ctype);
1472
1473 /*
1474 * Both of oldpage and newpage are still under lock_page().
1475 * Then, we don't have to care about race in radix-tree.
1476 * But we have to be careful that this page is unmapped or not.
1477 *
1478 * There is a case for !page_mapped(). At the start of
1479 * migration, oldpage was mapped. But now, it's zapped.
1480 * But we know *target* page is not freed/reused under us.
1481 * mem_cgroup_uncharge_page() does all necessary checks.
1482 */
1483 if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
1484 mem_cgroup_uncharge_page(target);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -08001485}
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001486
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001487/*
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001488 * A call to try to shrink memory usage under specified resource controller.
1489 * This is typically used for page reclaiming for shmem for reducing side
1490 * effect of page allocation from shmem, which is used by some mem_cgroup.
1491 */
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001492int mem_cgroup_shrink_usage(struct page *page,
1493 struct mm_struct *mm,
1494 gfp_t gfp_mask)
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001495{
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001496 struct mem_cgroup *mem = NULL;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001497 int progress = 0;
1498 int retry = MEM_CGROUP_RECLAIM_RETRIES;
1499
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08001500 if (mem_cgroup_disabled())
Li Zefancede86a2008-07-25 01:47:18 -07001501 return 0;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08001502 if (page)
1503 mem = try_get_mem_cgroup_from_swapcache(page);
1504 if (!mem && mm)
1505 mem = try_get_mem_cgroup_from_mm(mm);
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001506 if (unlikely(!mem))
Balbir Singh31a78f22008-09-28 23:09:31 +01001507 return 0;
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001508
1509 do {
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001510 progress = mem_cgroup_hierarchical_reclaim(mem, gfp_mask, true);
Daisuke Nishimurab85a96c2009-01-07 18:08:12 -08001511 progress += mem_cgroup_check_under_limit(mem);
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001512 } while (!progress && --retry);
1513
1514 css_put(&mem->css);
1515 if (!retry)
1516 return -ENOMEM;
1517 return 0;
1518}
1519
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001520static DEFINE_MUTEX(set_limit_mutex);
1521
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08001522static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001523 unsigned long long val)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001524{
1525
1526 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1527 int progress;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001528 u64 memswlimit;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001529 int ret = 0;
1530
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001531 while (retry_count) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001532 if (signal_pending(current)) {
1533 ret = -EINTR;
1534 break;
1535 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001536 /*
1537 * Rather than hide all in some function, I do this in
1538 * open coded manner. You see what this really does.
1539 * We have to guarantee mem->res.limit < mem->memsw.limit.
1540 */
1541 mutex_lock(&set_limit_mutex);
1542 memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1543 if (memswlimit < val) {
1544 ret = -EINVAL;
1545 mutex_unlock(&set_limit_mutex);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001546 break;
1547 }
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001548 ret = res_counter_set_limit(&memcg->res, val);
1549 mutex_unlock(&set_limit_mutex);
1550
1551 if (!ret)
1552 break;
1553
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001554 progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL,
1555 false);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001556 if (!progress) retry_count--;
1557 }
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001558
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001559 return ret;
1560}
1561
1562int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
1563 unsigned long long val)
1564{
1565 int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
1566 u64 memlimit, oldusage, curusage;
1567 int ret;
1568
1569 if (!do_swap_account)
1570 return -EINVAL;
1571
1572 while (retry_count) {
1573 if (signal_pending(current)) {
1574 ret = -EINTR;
1575 break;
1576 }
1577 /*
1578 * Rather than hide all in some function, I do this in
1579 * open coded manner. You see what this really does.
1580 * We have to guarantee mem->res.limit < mem->memsw.limit.
1581 */
1582 mutex_lock(&set_limit_mutex);
1583 memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1584 if (memlimit > val) {
1585 ret = -EINVAL;
1586 mutex_unlock(&set_limit_mutex);
1587 break;
1588 }
1589 ret = res_counter_set_limit(&memcg->memsw, val);
1590 mutex_unlock(&set_limit_mutex);
1591
1592 if (!ret)
1593 break;
1594
1595 oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
Daisuke Nishimura42e9abb2009-01-07 18:08:30 -08001596 mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001597 curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
1598 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001599 retry_count--;
1600 }
1601 return ret;
1602}
1603
KAMEZAWA Hiroyukic9b0ed52008-07-25 01:47:15 -07001604/*
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001605 * This routine traverse page_cgroup in given list and drop them all.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001606 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
1607 */
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001608static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001609 int node, int zid, enum lru_list lru)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001610{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001611 struct zone *zone;
1612 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001613 struct page_cgroup *pc, *busy;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001614 unsigned long flags, loop;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08001615 struct list_head *list;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001616 int ret = 0;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -08001617
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001618 zone = &NODE_DATA(node)->node_zones[zid];
1619 mz = mem_cgroup_zoneinfo(mem, node, zid);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001620 list = &mz->lists[lru];
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001621
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001622 loop = MEM_CGROUP_ZSTAT(mz, lru);
1623 /* give some margin against EBUSY etc...*/
1624 loop += 256;
1625 busy = NULL;
1626 while (loop--) {
1627 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001628 spin_lock_irqsave(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001629 if (list_empty(list)) {
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001630 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001631 break;
1632 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001633 pc = list_entry(list->prev, struct page_cgroup, lru);
1634 if (busy == pc) {
1635 list_move(&pc->lru, list);
1636 busy = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001637 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001638 continue;
1639 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001640 spin_unlock_irqrestore(&zone->lru_lock, flags);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001641
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08001642 ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001643 if (ret == -ENOMEM)
1644 break;
1645
1646 if (ret == -EBUSY || ret == -EINVAL) {
1647 /* found lock contention or "pc" is obsolete. */
1648 busy = pc;
1649 cond_resched();
1650 } else
1651 busy = NULL;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001652 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001653
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001654 if (!ret && !list_empty(list))
1655 return -EBUSY;
1656 return ret;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001657}
1658
1659/*
1660 * make mem_cgroup's charge to be 0 if there is no task.
1661 * This enables deleting this mem_cgroup.
1662 */
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001663static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001664{
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001665 int ret;
1666 int node, zid, shrink;
1667 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001668 struct cgroup *cgrp = mem->css.cgroup;
Hugh Dickins8869b8f2008-03-04 14:29:09 -08001669
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001670 css_get(&mem->css);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001671
1672 shrink = 0;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001673 /* should free all ? */
1674 if (free_all)
1675 goto try_to_free;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001676move_account:
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001677 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001678 ret = -EBUSY;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001679 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001680 goto out;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001681 ret = -EINTR;
1682 if (signal_pending(current))
1683 goto out;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001684 /* This is for making all *used* pages to be on LRU. */
1685 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001686 ret = 0;
1687 for_each_node_state(node, N_POSSIBLE) {
1688 for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001689 enum lru_list l;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001690 for_each_lru(l) {
1691 ret = mem_cgroup_force_empty_list(mem,
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001692 node, zid, l);
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001693 if (ret)
1694 break;
1695 }
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001696 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001697 if (ret)
1698 break;
1699 }
1700 /* it seems parent cgroup doesn't have enough mem */
1701 if (ret == -ENOMEM)
1702 goto try_to_free;
KAMEZAWA Hiroyuki52d4b9a2008-10-18 20:28:16 -07001703 cond_resched();
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001704 }
1705 ret = 0;
1706out:
1707 css_put(&mem->css);
1708 return ret;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001709
1710try_to_free:
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001711 /* returns EBUSY if there is a task or if we come here twice. */
1712 if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001713 ret = -EBUSY;
1714 goto out;
1715 }
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001716 /* we call try-to-free pages for make this cgroup empty */
1717 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001718 /* try to free all pages in this cgroup */
1719 shrink = 1;
1720 while (nr_retries && mem->res.usage > 0) {
1721 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001722
1723 if (signal_pending(current)) {
1724 ret = -EINTR;
1725 goto out;
1726 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001727 progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
1728 false, get_swappiness(mem));
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001729 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001730 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001731 /* maybe some writeback is necessary */
1732 congestion_wait(WRITE, HZ/10);
1733 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001734
1735 }
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001736 lru_add_drain();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001737 /* try move_account...there may be some *locked* pages. */
1738 if (mem->res.usage)
1739 goto move_account;
1740 ret = 0;
1741 goto out;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08001742}
1743
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08001744int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
1745{
1746 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
1747}
1748
1749
Balbir Singh18f59ea2009-01-07 18:08:07 -08001750static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
1751{
1752 return mem_cgroup_from_cont(cont)->use_hierarchy;
1753}
1754
1755static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
1756 u64 val)
1757{
1758 int retval = 0;
1759 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1760 struct cgroup *parent = cont->parent;
1761 struct mem_cgroup *parent_mem = NULL;
1762
1763 if (parent)
1764 parent_mem = mem_cgroup_from_cont(parent);
1765
1766 cgroup_lock();
1767 /*
1768 * If parent's use_hiearchy is set, we can't make any modifications
1769 * in the child subtrees. If it is unset, then the change can
1770 * occur, provided the current cgroup has no children.
1771 *
1772 * For the root cgroup, parent_mem is NULL, we allow value to be
1773 * set if there are no children.
1774 */
1775 if ((!parent_mem || !parent_mem->use_hierarchy) &&
1776 (val == 1 || val == 0)) {
1777 if (list_empty(&cont->children))
1778 mem->use_hierarchy = val;
1779 else
1780 retval = -EBUSY;
1781 } else
1782 retval = -EINVAL;
1783 cgroup_unlock();
1784
1785 return retval;
1786}
1787
Paul Menage2c3daa72008-04-29 00:59:58 -07001788static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001789{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001790 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1791 u64 val = 0;
1792 int type, name;
1793
1794 type = MEMFILE_TYPE(cft->private);
1795 name = MEMFILE_ATTR(cft->private);
1796 switch (type) {
1797 case _MEM:
1798 val = res_counter_read_u64(&mem->res, name);
1799 break;
1800 case _MEMSWAP:
1801 if (do_swap_account)
1802 val = res_counter_read_u64(&mem->memsw, name);
1803 break;
1804 default:
1805 BUG();
1806 break;
1807 }
1808 return val;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001809}
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001810/*
1811 * The user of this function is...
1812 * RES_LIMIT.
1813 */
Paul Menage856c13a2008-07-25 01:47:04 -07001814static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
1815 const char *buffer)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001816{
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001817 struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001818 int type, name;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001819 unsigned long long val;
1820 int ret;
1821
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001822 type = MEMFILE_TYPE(cft->private);
1823 name = MEMFILE_ATTR(cft->private);
1824 switch (name) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001825 case RES_LIMIT:
1826 /* This function does all necessary parse...reuse it */
1827 ret = res_counter_memparse_write_strategy(buffer, &val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001828 if (ret)
1829 break;
1830 if (type == _MEM)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001831 ret = mem_cgroup_resize_limit(memcg, val);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001832 else
1833 ret = mem_cgroup_resize_memsw_limit(memcg, val);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07001834 break;
1835 default:
1836 ret = -EINVAL; /* should be BUG() ? */
1837 break;
1838 }
1839 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001840}
1841
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08001842static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
1843 unsigned long long *mem_limit, unsigned long long *memsw_limit)
1844{
1845 struct cgroup *cgroup;
1846 unsigned long long min_limit, min_memsw_limit, tmp;
1847
1848 min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1849 min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1850 cgroup = memcg->css.cgroup;
1851 if (!memcg->use_hierarchy)
1852 goto out;
1853
1854 while (cgroup->parent) {
1855 cgroup = cgroup->parent;
1856 memcg = mem_cgroup_from_cont(cgroup);
1857 if (!memcg->use_hierarchy)
1858 break;
1859 tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
1860 min_limit = min(min_limit, tmp);
1861 tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1862 min_memsw_limit = min(min_memsw_limit, tmp);
1863 }
1864out:
1865 *mem_limit = min_limit;
1866 *memsw_limit = min_memsw_limit;
1867 return;
1868}
1869
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07001870static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07001871{
1872 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001873 int type, name;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07001874
1875 mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001876 type = MEMFILE_TYPE(event);
1877 name = MEMFILE_ATTR(event);
1878 switch (name) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07001879 case RES_MAX_USAGE:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001880 if (type == _MEM)
1881 res_counter_reset_max(&mem->res);
1882 else
1883 res_counter_reset_max(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07001884 break;
1885 case RES_FAILCNT:
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08001886 if (type == _MEM)
1887 res_counter_reset_failcnt(&mem->res);
1888 else
1889 res_counter_reset_failcnt(&mem->memsw);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07001890 break;
1891 }
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -07001892 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07001893}
1894
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001895static const struct mem_cgroup_stat_desc {
1896 const char *msg;
1897 u64 unit;
1898} mem_cgroup_stat_desc[] = {
1899 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
1900 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
Balaji Rao55e462b2008-05-01 04:35:12 -07001901 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
1902 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001903};
1904
Paul Menagec64745c2008-04-29 01:00:02 -07001905static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1906 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001907{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001908 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1909 struct mem_cgroup_stat *stat = &mem_cont->stat;
1910 int i;
1911
1912 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1913 s64 val;
1914
1915 val = mem_cgroup_read_stat(stat, i);
1916 val *= mem_cgroup_stat_desc[i].unit;
Paul Menagec64745c2008-04-29 01:00:02 -07001917 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001918 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001919 /* showing # of active pages */
1920 {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001921 unsigned long active_anon, inactive_anon;
1922 unsigned long active_file, inactive_file;
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001923 unsigned long unevictable;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001924
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001925 inactive_anon = mem_cgroup_get_all_zonestat(mem_cont,
1926 LRU_INACTIVE_ANON);
1927 active_anon = mem_cgroup_get_all_zonestat(mem_cont,
1928 LRU_ACTIVE_ANON);
1929 inactive_file = mem_cgroup_get_all_zonestat(mem_cont,
1930 LRU_INACTIVE_FILE);
1931 active_file = mem_cgroup_get_all_zonestat(mem_cont,
1932 LRU_ACTIVE_FILE);
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001933 unevictable = mem_cgroup_get_all_zonestat(mem_cont,
1934 LRU_UNEVICTABLE);
1935
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001936 cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE);
1937 cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE);
1938 cb->fill(cb, "active_file", (active_file) * PAGE_SIZE);
1939 cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE);
Lee Schermerhorn7b854122008-10-18 20:26:40 -07001940 cb->fill(cb, "unevictable", unevictable * PAGE_SIZE);
1941
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001942 }
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08001943 {
1944 unsigned long long limit, memsw_limit;
1945 memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
1946 cb->fill(cb, "hierarchical_memory_limit", limit);
1947 if (do_swap_account)
1948 cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
1949 }
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08001950
1951#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001952 cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08001953
1954 {
1955 int nid, zid;
1956 struct mem_cgroup_per_zone *mz;
1957 unsigned long recent_rotated[2] = {0, 0};
1958 unsigned long recent_scanned[2] = {0, 0};
1959
1960 for_each_online_node(nid)
1961 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1962 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1963
1964 recent_rotated[0] +=
1965 mz->reclaim_stat.recent_rotated[0];
1966 recent_rotated[1] +=
1967 mz->reclaim_stat.recent_rotated[1];
1968 recent_scanned[0] +=
1969 mz->reclaim_stat.recent_scanned[0];
1970 recent_scanned[1] +=
1971 mz->reclaim_stat.recent_scanned[1];
1972 }
1973 cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
1974 cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
1975 cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
1976 cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
1977 }
1978#endif
1979
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08001980 return 0;
1981}
1982
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001983static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
1984{
1985 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1986
1987 return get_swappiness(memcg);
1988}
1989
1990static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
1991 u64 val)
1992{
1993 struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
1994 struct mem_cgroup *parent;
Li Zefan068b38c2009-01-15 13:51:26 -08001995
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001996 if (val > 100)
1997 return -EINVAL;
1998
1999 if (cgrp->parent == NULL)
2000 return -EINVAL;
2001
2002 parent = mem_cgroup_from_cont(cgrp->parent);
Li Zefan068b38c2009-01-15 13:51:26 -08002003
2004 cgroup_lock();
2005
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002006 /* If under hierarchy, only empty-root can set this value */
2007 if ((parent->use_hierarchy) ||
Li Zefan068b38c2009-01-15 13:51:26 -08002008 (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
2009 cgroup_unlock();
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002010 return -EINVAL;
Li Zefan068b38c2009-01-15 13:51:26 -08002011 }
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002012
2013 spin_lock(&memcg->reclaim_param_lock);
2014 memcg->swappiness = val;
2015 spin_unlock(&memcg->reclaim_param_lock);
2016
Li Zefan068b38c2009-01-15 13:51:26 -08002017 cgroup_unlock();
2018
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002019 return 0;
2020}
2021
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002022
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002023static struct cftype mem_cgroup_files[] = {
2024 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002025 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002026 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Paul Menage2c3daa72008-04-29 00:59:58 -07002027 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002028 },
2029 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002030 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002031 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002032 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07002033 .read_u64 = mem_cgroup_read,
2034 },
2035 {
Balbir Singh0eea1032008-02-07 00:13:57 -08002036 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002037 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Paul Menage856c13a2008-07-25 01:47:04 -07002038 .write_string = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -07002039 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002040 },
2041 {
2042 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002043 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07002044 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -07002045 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002046 },
Balbir Singh8697d332008-02-07 00:13:59 -08002047 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002048 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -07002049 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08002050 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002051 {
2052 .name = "force_empty",
2053 .trigger = mem_cgroup_force_empty_write,
2054 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08002055 {
2056 .name = "use_hierarchy",
2057 .write_u64 = mem_cgroup_hierarchy_write,
2058 .read_u64 = mem_cgroup_hierarchy_read,
2059 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002060 {
2061 .name = "swappiness",
2062 .read_u64 = mem_cgroup_swappiness_read,
2063 .write_u64 = mem_cgroup_swappiness_write,
2064 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002065};
2066
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002067#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2068static struct cftype memsw_cgroup_files[] = {
2069 {
2070 .name = "memsw.usage_in_bytes",
2071 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
2072 .read_u64 = mem_cgroup_read,
2073 },
2074 {
2075 .name = "memsw.max_usage_in_bytes",
2076 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
2077 .trigger = mem_cgroup_reset,
2078 .read_u64 = mem_cgroup_read,
2079 },
2080 {
2081 .name = "memsw.limit_in_bytes",
2082 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
2083 .write_string = mem_cgroup_write,
2084 .read_u64 = mem_cgroup_read,
2085 },
2086 {
2087 .name = "memsw.failcnt",
2088 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
2089 .trigger = mem_cgroup_reset,
2090 .read_u64 = mem_cgroup_read,
2091 },
2092};
2093
2094static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2095{
2096 if (!do_swap_account)
2097 return 0;
2098 return cgroup_add_files(cont, ss, memsw_cgroup_files,
2099 ARRAY_SIZE(memsw_cgroup_files));
2100};
2101#else
2102static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
2103{
2104 return 0;
2105}
2106#endif
2107
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002108static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2109{
2110 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002111 struct mem_cgroup_per_zone *mz;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002112 enum lru_list l;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002113 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002114 /*
2115 * This routine is called against possible nodes.
2116 * But it's BUG to call kmalloc() against offline node.
2117 *
2118 * TODO: this routine can waste much memory for nodes which will
2119 * never be onlined. It's better to use memory hotplug callback
2120 * function.
2121 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07002122 if (!node_state(node, N_NORMAL_MEMORY))
2123 tmp = -1;
2124 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002125 if (!pn)
2126 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002127
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002128 mem->info.nodeinfo[node] = pn;
2129 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002130
2131 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
2132 mz = &pn->zoneinfo[zone];
Christoph Lameterb69408e2008-10-18 20:26:14 -07002133 for_each_lru(l)
2134 INIT_LIST_HEAD(&mz->lists[l]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002135 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002136 return 0;
2137}
2138
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08002139static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
2140{
2141 kfree(mem->info.nodeinfo[node]);
2142}
2143
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002144static int mem_cgroup_size(void)
2145{
2146 int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu);
2147 return sizeof(struct mem_cgroup) + cpustat_size;
2148}
2149
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002150static struct mem_cgroup *mem_cgroup_alloc(void)
2151{
2152 struct mem_cgroup *mem;
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002153 int size = mem_cgroup_size();
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002154
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002155 if (size < PAGE_SIZE)
2156 mem = kmalloc(size, GFP_KERNEL);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002157 else
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002158 mem = vmalloc(size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002159
2160 if (mem)
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002161 memset(mem, 0, size);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002162 return mem;
2163}
2164
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002165/*
2166 * At destroying mem_cgroup, references from swap_cgroup can remain.
2167 * (scanning all at force_empty is too costly...)
2168 *
2169 * Instead of clearing all references at force_empty, we remember
2170 * the number of reference from swap_cgroup and free mem_cgroup when
2171 * it goes down to 0.
2172 *
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002173 * Removal of cgroup itself succeeds regardless of refs from swap.
2174 */
2175
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002176static void __mem_cgroup_free(struct mem_cgroup *mem)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002177{
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002178 int node;
2179
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002180 for_each_node_state(node, N_POSSIBLE)
2181 free_mem_cgroup_per_zone_info(mem, node);
2182
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002183 if (mem_cgroup_size() < PAGE_SIZE)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002184 kfree(mem);
2185 else
2186 vfree(mem);
2187}
2188
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002189static void mem_cgroup_get(struct mem_cgroup *mem)
2190{
2191 atomic_inc(&mem->refcnt);
2192}
2193
2194static void mem_cgroup_put(struct mem_cgroup *mem)
2195{
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002196 if (atomic_dec_and_test(&mem->refcnt))
2197 __mem_cgroup_free(mem);
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002198}
2199
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07002200
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002201#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2202static void __init enable_swap_cgroup(void)
2203{
Hirokazu Takahashif8d66542009-01-07 18:08:02 -08002204 if (!mem_cgroup_disabled() && really_do_swap_account)
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002205 do_swap_account = 1;
2206}
2207#else
2208static void __init enable_swap_cgroup(void)
2209{
2210}
2211#endif
2212
Li Zefan0eb253e2009-01-15 13:51:25 -08002213static struct cgroup_subsys_state * __ref
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002214mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
2215{
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002216 struct mem_cgroup *mem, *parent;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002217 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002218
Jan Blunckc8dad2b2009-01-07 18:07:53 -08002219 mem = mem_cgroup_alloc();
2220 if (!mem)
2221 return ERR_PTR(-ENOMEM);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08002222
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002223 for_each_node_state(node, N_POSSIBLE)
2224 if (alloc_mem_cgroup_per_zone_info(mem, node))
2225 goto free_out;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002226 /* root ? */
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002227 if (cont->parent == NULL) {
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002228 enable_swap_cgroup();
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002229 parent = NULL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002230 } else {
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002231 parent = mem_cgroup_from_cont(cont->parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002232 mem->use_hierarchy = parent->use_hierarchy;
2233 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08002234
Balbir Singh18f59ea2009-01-07 18:08:07 -08002235 if (parent && parent->use_hierarchy) {
2236 res_counter_init(&mem->res, &parent->res);
2237 res_counter_init(&mem->memsw, &parent->memsw);
2238 } else {
2239 res_counter_init(&mem->res, NULL);
2240 res_counter_init(&mem->memsw, NULL);
2241 }
Balbir Singh6d61ef42009-01-07 18:08:06 -08002242 mem->last_scanned_child = NULL;
KOSAKI Motohiro2733c062009-01-07 18:08:23 -08002243 spin_lock_init(&mem->reclaim_param_lock);
Balbir Singh6d61ef42009-01-07 18:08:06 -08002244
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08002245 if (parent)
2246 mem->swappiness = get_swappiness(parent);
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002247 atomic_set(&mem->refcnt, 1);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002248 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002249free_out:
KAMEZAWA Hiroyukia7ba0ee2009-01-07 18:08:32 -08002250 __mem_cgroup_free(mem);
Li Zefan2dda81c2008-02-23 15:24:14 -08002251 return ERR_PTR(-ENOMEM);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002252}
2253
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002254static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
2255 struct cgroup *cont)
2256{
2257 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002258 mem_cgroup_force_empty(mem, false);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002259}
2260
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002261static void mem_cgroup_destroy(struct cgroup_subsys *ss,
2262 struct cgroup *cont)
2263{
Daisuke Nishimurac268e992009-01-15 13:51:13 -08002264 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
2265 struct mem_cgroup *last_scanned_child = mem->last_scanned_child;
2266
2267 if (last_scanned_child) {
2268 VM_BUG_ON(!mem_cgroup_is_obsolete(last_scanned_child));
2269 mem_cgroup_put(last_scanned_child);
2270 }
2271 mem_cgroup_put(mem);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002272}
2273
2274static int mem_cgroup_populate(struct cgroup_subsys *ss,
2275 struct cgroup *cont)
2276{
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002277 int ret;
2278
2279 ret = cgroup_add_files(cont, ss, mem_cgroup_files,
2280 ARRAY_SIZE(mem_cgroup_files));
2281
2282 if (!ret)
2283 ret = register_memsw_files(cont, ss);
2284 return ret;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002285}
2286
Balbir Singh67e465a2008-02-07 00:13:54 -08002287static void mem_cgroup_move_task(struct cgroup_subsys *ss,
2288 struct cgroup *cont,
2289 struct cgroup *old_cont,
2290 struct task_struct *p)
2291{
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002292 mutex_lock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002293 /*
Nikanth Karthikesanf9717d22009-01-07 18:08:11 -08002294 * FIXME: It's better to move charges of this process from old
2295 * memcg to new memcg. But it's just on TODO-List now.
Balbir Singh67e465a2008-02-07 00:13:54 -08002296 */
Daisuke Nishimura7f4d4542009-01-07 18:08:29 -08002297 mutex_unlock(&memcg_tasklist);
Balbir Singh67e465a2008-02-07 00:13:54 -08002298}
2299
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002300struct cgroup_subsys mem_cgroup_subsys = {
2301 .name = "memory",
2302 .subsys_id = mem_cgroup_subsys_id,
2303 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08002304 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002305 .destroy = mem_cgroup_destroy,
2306 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08002307 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08002308 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002309};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08002310
2311#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2312
2313static int __init disable_swap_account(char *s)
2314{
2315 really_do_swap_account = 0;
2316 return 1;
2317}
2318__setup("noswapaccount", disable_swap_account);
2319#endif