blob: e46451e1d9b793563b08d8ef015927b484b0be1d [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 Hiroyukid52aa412008-02-07 00:14:24 -080024#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080025#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080026#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080027#include <linux/bit_spinlock.h>
28#include <linux/rcupdate.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070029#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080030#include <linux/swap.h>
31#include <linux/spinlock.h>
32#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080033#include <linux/seq_file.h>
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -070034#include <linux/vmalloc.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035
Balbir Singh8697d332008-02-07 00:13:59 -080036#include <asm/uaccess.h>
37
Balbir Singh8cdea7c2008-02-07 00:13:50 -080038struct cgroup_subsys mem_cgroup_subsys;
Balbir Singh66e17072008-02-07 00:13:56 -080039static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
Balbir Singhb6ac57d2008-04-29 01:00:19 -070040static struct kmem_cache *page_cgroup_cache;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080041
42/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080043 * Statistics for memory cgroup.
44 */
45enum mem_cgroup_stat_index {
46 /*
47 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
48 */
49 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
50 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
Balaji Rao55e462b2008-05-01 04:35:12 -070051 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
52 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080053
54 MEM_CGROUP_STAT_NSTATS,
55};
56
57struct mem_cgroup_stat_cpu {
58 s64 count[MEM_CGROUP_STAT_NSTATS];
59} ____cacheline_aligned_in_smp;
60
61struct mem_cgroup_stat {
62 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
63};
64
65/*
66 * For accounting under irq disable, no need for increment preempt count.
67 */
68static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
69 enum mem_cgroup_stat_index idx, int val)
70{
71 int cpu = smp_processor_id();
72 stat->cpustat[cpu].count[idx] += val;
73}
74
75static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
76 enum mem_cgroup_stat_index idx)
77{
78 int cpu;
79 s64 ret = 0;
80 for_each_possible_cpu(cpu)
81 ret += stat->cpustat[cpu].count[idx];
82 return ret;
83}
84
85/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -080086 * per-zone information in memory controller.
87 */
88
89enum mem_cgroup_zstat_index {
90 MEM_CGROUP_ZSTAT_ACTIVE,
91 MEM_CGROUP_ZSTAT_INACTIVE,
92
93 NR_MEM_CGROUP_ZSTAT,
94};
95
96struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -080097 /*
98 * spin_lock to protect the per cgroup LRU
99 */
100 spinlock_t lru_lock;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800101 struct list_head active_list;
102 struct list_head inactive_list;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800103 unsigned long count[NR_MEM_CGROUP_ZSTAT];
104};
105/* Macro for accessing counter */
106#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
107
108struct mem_cgroup_per_node {
109 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
110};
111
112struct mem_cgroup_lru_info {
113 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
114};
115
116/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800117 * The memory controller data structure. The memory controller controls both
118 * page cache and RSS per cgroup. We would eventually like to provide
119 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
120 * to help the administrator determine what knobs to tune.
121 *
122 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800123 * we hit the water mark. May be even add a low water mark, such that
124 * no reclaim occurs from a cgroup at it's low water mark, this is
125 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800126 */
127struct mem_cgroup {
128 struct cgroup_subsys_state css;
129 /*
130 * the counter to account for memory usage
131 */
132 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800133 /*
134 * Per cgroup active and inactive list, similar to the
135 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800136 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800137 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800138
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800139 int prev_priority; /* for recording reclaim priority */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800140 /*
141 * statistics.
142 */
143 struct mem_cgroup_stat stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800144};
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800145static struct mem_cgroup init_mem_cgroup;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800146
147/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800148 * We use the lower bit of the page->page_cgroup pointer as a bit spin
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800149 * lock. We need to ensure that page->page_cgroup is at least two
150 * byte aligned (based on comments from Nick Piggin). But since
151 * bit_spin_lock doesn't actually set that lock bit in a non-debug
152 * uniprocessor kernel, we should avoid setting it here too.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800153 */
154#define PAGE_CGROUP_LOCK_BIT 0x0
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800155#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
156#define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
157#else
158#define PAGE_CGROUP_LOCK 0x0
159#endif
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800160
161/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800162 * A page_cgroup page is associated with every page descriptor. The
163 * page_cgroup helps us identify information about the cgroup
164 */
165struct page_cgroup {
166 struct list_head lru; /* per cgroup LRU list */
167 struct page *page;
168 struct mem_cgroup *mem_cgroup;
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800169 int ref_cnt; /* cached, mapped, migrating */
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800170 int flags;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800171};
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800172#define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800173#define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800174
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800175static int page_cgroup_nid(struct page_cgroup *pc)
KAMEZAWA Hiroyukic0149532008-02-07 00:14:30 -0800176{
177 return page_to_nid(pc->page);
178}
179
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800180static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
KAMEZAWA Hiroyukic0149532008-02-07 00:14:30 -0800181{
182 return page_zonenum(pc->page);
183}
184
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800185enum charge_type {
186 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
187 MEM_CGROUP_CHARGE_TYPE_MAPPED,
188};
189
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800190/*
191 * Always modified under lru lock. Then, not necessary to preempt_disable()
192 */
193static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
194 bool charge)
195{
196 int val = (charge)? 1 : -1;
197 struct mem_cgroup_stat *stat = &mem->stat;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800198
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800199 VM_BUG_ON(!irqs_disabled());
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800200 if (flags & PAGE_CGROUP_FLAG_CACHE)
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800201 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800202 else
203 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
Balaji Rao55e462b2008-05-01 04:35:12 -0700204
205 if (charge)
206 __mem_cgroup_stat_add_safe(stat,
207 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
208 else
209 __mem_cgroup_stat_add_safe(stat,
210 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800211}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800212
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800213static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800214mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
215{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800216 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
217}
218
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800219static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800220page_cgroup_zoneinfo(struct page_cgroup *pc)
221{
222 struct mem_cgroup *mem = pc->mem_cgroup;
223 int nid = page_cgroup_nid(pc);
224 int zid = page_cgroup_zid(pc);
225
226 return mem_cgroup_zoneinfo(mem, nid, zid);
227}
228
229static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 enum mem_cgroup_zstat_index idx)
231{
232 int nid, zid;
233 struct mem_cgroup_per_zone *mz;
234 u64 total = 0;
235
236 for_each_online_node(nid)
237 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
238 mz = mem_cgroup_zoneinfo(mem, nid, zid);
239 total += MEM_CGROUP_ZSTAT(mz, idx);
240 }
241 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800242}
243
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800244static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800245{
246 return container_of(cgroup_subsys_state(cont,
247 mem_cgroup_subsys_id), struct mem_cgroup,
248 css);
249}
250
Balbir Singhcf475ad2008-04-29 01:00:16 -0700251struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800252{
253 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
254 struct mem_cgroup, css);
255}
256
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800257static inline int page_cgroup_locked(struct page *page)
258{
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800259 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800260}
261
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800262static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800263{
Hugh Dickins9442ec9d2008-03-04 14:29:07 -0800264 VM_BUG_ON(!page_cgroup_locked(page));
265 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800266}
267
268struct page_cgroup *page_get_page_cgroup(struct page *page)
269{
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800270 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800271}
272
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800273static void lock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800274{
275 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800276}
277
Hugh Dickins2680eed2008-03-04 14:29:13 -0800278static int try_lock_page_cgroup(struct page *page)
279{
280 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
281}
282
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800283static void unlock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800284{
285 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
286}
287
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700288static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
289 struct page_cgroup *pc)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800290{
291 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800292
293 if (from)
294 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
295 else
296 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
297
298 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
299 list_del_init(&pc->lru);
300}
301
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700302static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
303 struct page_cgroup *pc)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800304{
305 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800306
307 if (!to) {
308 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800309 list_add(&pc->lru, &mz->inactive_list);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800310 } else {
311 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800312 list_add(&pc->lru, &mz->active_list);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800313 }
314 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
315}
316
Balbir Singh8697d332008-02-07 00:13:59 -0800317static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
Balbir Singh66e17072008-02-07 00:13:56 -0800318{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800319 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
320 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
321
322 if (from)
323 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
324 else
325 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
326
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800327 if (active) {
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800328 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800329 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800330 list_move(&pc->lru, &mz->active_list);
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800331 } else {
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800332 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800333 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800334 list_move(&pc->lru, &mz->inactive_list);
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800335 }
Balbir Singh66e17072008-02-07 00:13:56 -0800336}
337
David Rientjes4c4a2212008-02-07 00:14:06 -0800338int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
339{
340 int ret;
341
342 task_lock(task);
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800343 ret = task->mm && mm_match_cgroup(task->mm, mem);
David Rientjes4c4a2212008-02-07 00:14:06 -0800344 task_unlock(task);
345 return ret;
346}
347
Balbir Singh66e17072008-02-07 00:13:56 -0800348/*
349 * This routine assumes that the appropriate zone's lru lock is already held
350 */
Hugh Dickins427d5412008-03-04 14:29:03 -0800351void mem_cgroup_move_lists(struct page *page, bool active)
Balbir Singh66e17072008-02-07 00:13:56 -0800352{
Hugh Dickins427d5412008-03-04 14:29:03 -0800353 struct page_cgroup *pc;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800354 struct mem_cgroup_per_zone *mz;
355 unsigned long flags;
356
Hugh Dickins2680eed2008-03-04 14:29:13 -0800357 /*
358 * We cannot lock_page_cgroup while holding zone's lru_lock,
359 * because other holders of lock_page_cgroup can be interrupted
360 * with an attempt to rotate_reclaimable_page. But we cannot
361 * safely get to page_cgroup without it, so just try_lock it:
362 * mem_cgroup_isolate_pages allows for page left on wrong list.
363 */
364 if (!try_lock_page_cgroup(page))
Balbir Singh66e17072008-02-07 00:13:56 -0800365 return;
366
Hugh Dickins2680eed2008-03-04 14:29:13 -0800367 pc = page_get_page_cgroup(page);
368 if (pc) {
Hugh Dickins2680eed2008-03-04 14:29:13 -0800369 mz = page_cgroup_zoneinfo(pc);
Hugh Dickins2680eed2008-03-04 14:29:13 -0800370 spin_lock_irqsave(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800371 __mem_cgroup_move_lists(pc, active);
Hugh Dickins2680eed2008-03-04 14:29:13 -0800372 spin_unlock_irqrestore(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800373 }
374 unlock_page_cgroup(page);
Balbir Singh66e17072008-02-07 00:13:56 -0800375}
376
KAMEZAWA Hiroyuki58ae83d2008-02-07 00:14:32 -0800377/*
378 * Calculate mapped_ratio under memory controller. This will be used in
379 * vmscan.c for deteremining we have to reclaim mapped pages.
380 */
381int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
382{
383 long total, rss;
384
385 /*
386 * usage is recorded in bytes. But, here, we assume the number of
387 * physical pages can be represented by "long" on any arch.
388 */
389 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
390 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
391 return (int)((rss * 100L) / total);
392}
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800393
KAMEZAWA Hiroyuki5932f362008-02-07 00:14:33 -0800394/*
395 * This function is called from vmscan.c. In page reclaiming loop. balance
396 * between active and inactive list is calculated. For memory controller
397 * page reclaiming, we should use using mem_cgroup's imbalance rather than
398 * zone's global lru imbalance.
399 */
400long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
401{
402 unsigned long active, inactive;
403 /* active and inactive are the number of pages. 'long' is ok.*/
404 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
405 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
406 return (long) (active / (inactive + 1));
407}
KAMEZAWA Hiroyuki58ae83d2008-02-07 00:14:32 -0800408
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800409/*
410 * prev_priority control...this will be used in memory reclaim path.
411 */
412int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
413{
414 return mem->prev_priority;
415}
416
417void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
418{
419 if (priority < mem->prev_priority)
420 mem->prev_priority = priority;
421}
422
423void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
424{
425 mem->prev_priority = priority;
426}
427
KAMEZAWA Hiroyukicc381082008-02-07 00:14:35 -0800428/*
429 * Calculate # of pages to be scanned in this priority/zone.
430 * See also vmscan.c
431 *
432 * priority starts from "DEF_PRIORITY" and decremented in each loop.
433 * (see include/linux/mmzone.h)
434 */
435
436long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
437 struct zone *zone, int priority)
438{
439 long nr_active;
440 int nid = zone->zone_pgdat->node_id;
441 int zid = zone_idx(zone);
442 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
443
444 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
445 return (nr_active >> priority);
446}
447
448long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
449 struct zone *zone, int priority)
450{
451 long nr_inactive;
452 int nid = zone->zone_pgdat->node_id;
453 int zid = zone_idx(zone);
454 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
455
456 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
KAMEZAWA Hiroyukicc381082008-02-07 00:14:35 -0800457 return (nr_inactive >> priority);
458}
459
Balbir Singh66e17072008-02-07 00:13:56 -0800460unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
461 struct list_head *dst,
462 unsigned long *scanned, int order,
463 int mode, struct zone *z,
464 struct mem_cgroup *mem_cont,
465 int active)
466{
467 unsigned long nr_taken = 0;
468 struct page *page;
469 unsigned long scan;
470 LIST_HEAD(pc_list);
471 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800472 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800473 int nid = z->zone_pgdat->node_id;
474 int zid = zone_idx(z);
475 struct mem_cgroup_per_zone *mz;
Balbir Singh66e17072008-02-07 00:13:56 -0800476
Balbir Singhcf475ad2008-04-29 01:00:16 -0700477 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800478 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Balbir Singh66e17072008-02-07 00:13:56 -0800479 if (active)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800480 src = &mz->active_list;
Balbir Singh66e17072008-02-07 00:13:56 -0800481 else
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800482 src = &mz->inactive_list;
483
Balbir Singh66e17072008-02-07 00:13:56 -0800484
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800485 spin_lock(&mz->lru_lock);
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800486 scan = 0;
487 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800488 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800489 break;
Balbir Singh66e17072008-02-07 00:13:56 -0800490 page = pc->page;
Balbir Singh66e17072008-02-07 00:13:56 -0800491
Hugh Dickins436c65412008-02-07 00:14:12 -0800492 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800493 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800494
Balbir Singh66e17072008-02-07 00:13:56 -0800495 if (PageActive(page) && !active) {
496 __mem_cgroup_move_lists(pc, true);
Balbir Singh66e17072008-02-07 00:13:56 -0800497 continue;
498 }
499 if (!PageActive(page) && active) {
500 __mem_cgroup_move_lists(pc, false);
Balbir Singh66e17072008-02-07 00:13:56 -0800501 continue;
502 }
503
Hugh Dickins436c65412008-02-07 00:14:12 -0800504 scan++;
505 list_move(&pc->lru, &pc_list);
Balbir Singh66e17072008-02-07 00:13:56 -0800506
507 if (__isolate_lru_page(page, mode) == 0) {
508 list_move(&page->lru, dst);
509 nr_taken++;
510 }
511 }
512
513 list_splice(&pc_list, src);
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800514 spin_unlock(&mz->lru_lock);
Balbir Singh66e17072008-02-07 00:13:56 -0800515
516 *scanned = scan;
517 return nr_taken;
518}
519
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800520/*
521 * Charge the memory controller for page usage.
522 * Return
523 * 0 if the charge was successful
524 * < 0 if the cgroup is over its limit
525 */
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800526static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
527 gfp_t gfp_mask, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800528{
529 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800530 struct page_cgroup *pc;
Balbir Singh66e17072008-02-07 00:13:56 -0800531 unsigned long flags;
532 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800533 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800534
Balbir Singh40779602008-04-04 14:29:59 -0700535 if (mem_cgroup_subsys.disabled)
536 return 0;
537
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800538 /*
539 * Should page_cgroup's go to their own slab?
540 * One could optimize the performance of the charging routine
541 * by saving a bit in the page_flags and using it as a lock
542 * to see if the cgroup page already has a page_cgroup associated
543 * with it
544 */
Balbir Singh66e17072008-02-07 00:13:56 -0800545retry:
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800546 lock_page_cgroup(page);
547 pc = page_get_page_cgroup(page);
548 /*
549 * The page_cgroup exists and
550 * the page has already been accounted.
551 */
552 if (pc) {
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800553 VM_BUG_ON(pc->page != page);
554 VM_BUG_ON(pc->ref_cnt <= 0);
555
556 pc->ref_cnt++;
557 unlock_page_cgroup(page);
558 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800559 }
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800560 unlock_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800561
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700562 pc = kmem_cache_zalloc(page_cgroup_cache, gfp_mask);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800563 if (pc == NULL)
564 goto err;
565
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800566 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800567 * We always charge the cgroup the mm_struct belongs to.
568 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800569 * thread group leader migrates. It's possible that mm is not
570 * set, if so charge the init_mm (happens for pagecache usage).
571 */
572 if (!mm)
573 mm = &init_mm;
574
Hugh Dickins3be91272008-02-07 00:14:19 -0800575 rcu_read_lock();
Balbir Singhcf475ad2008-04-29 01:00:16 -0700576 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800577 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800578 * For every charge from the cgroup, increment reference count
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800579 */
580 css_get(&mem->css);
581 rcu_read_unlock();
582
Balbir Singh0eea1032008-02-07 00:13:57 -0800583 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
Hugh Dickins3be91272008-02-07 00:14:19 -0800584 if (!(gfp_mask & __GFP_WAIT))
585 goto out;
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800586
587 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
Balbir Singh66e17072008-02-07 00:13:56 -0800588 continue;
589
590 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800591 * try_to_free_mem_cgroup_pages() might not give us a full
592 * picture of reclaim. Some pages are reclaimed and might be
593 * moved to swap cache or just unmapped from the cgroup.
594 * Check the limit again to see if the reclaim reduced the
595 * current usage of the cgroup before giving up
596 */
Balbir Singh66e17072008-02-07 00:13:56 -0800597 if (res_counter_check_under_limit(&mem->res))
598 continue;
Hugh Dickins3be91272008-02-07 00:14:19 -0800599
600 if (!nr_retries--) {
601 mem_cgroup_out_of_memory(mem, gfp_mask);
602 goto out;
Balbir Singh66e17072008-02-07 00:13:56 -0800603 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800604 }
605
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800606 pc->ref_cnt = 1;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800607 pc->mem_cgroup = mem;
608 pc->page = page;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800609 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800610 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
Balbir Singh4a56d022008-04-29 01:00:23 -0700611 pc->flags = PAGE_CGROUP_FLAG_CACHE;
Hugh Dickins3be91272008-02-07 00:14:19 -0800612
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800613 lock_page_cgroup(page);
614 if (page_get_page_cgroup(page)) {
615 unlock_page_cgroup(page);
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800616 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800617 * Another charge has been added to this page already.
618 * We take lock_page_cgroup(page) again and read
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800619 * page->cgroup, increment refcnt.... just retry is OK.
620 */
621 res_counter_uncharge(&mem->res, PAGE_SIZE);
622 css_put(&mem->css);
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700623 kmem_cache_free(page_cgroup_cache, pc);
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800624 goto retry;
625 }
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800626 page_assign_page_cgroup(page, pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800627
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800628 mz = page_cgroup_zoneinfo(pc);
629 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700630 __mem_cgroup_add_list(mz, pc);
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800631 spin_unlock_irqrestore(&mz->lru_lock, flags);
Balbir Singh66e17072008-02-07 00:13:56 -0800632
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800633 unlock_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800634done:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800635 return 0;
Hugh Dickins3be91272008-02-07 00:14:19 -0800636out:
637 css_put(&mem->css);
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700638 kmem_cache_free(page_cgroup_cache, pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800639err:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800640 return -ENOMEM;
641}
642
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800643int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800644{
645 return mem_cgroup_charge_common(page, mm, gfp_mask,
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800646 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800647}
648
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800649int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
650 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -0800651{
Balbir Singh8697d332008-02-07 00:13:59 -0800652 if (!mm)
653 mm = &init_mm;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800654 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800655 MEM_CGROUP_CHARGE_TYPE_CACHE);
Balbir Singh8697d332008-02-07 00:13:59 -0800656}
657
658/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800659 * Uncharging is always a welcome operation, we never complain, simply
Hugh Dickins82895462008-03-04 14:29:08 -0800660 * uncharge.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800661 */
Hugh Dickins82895462008-03-04 14:29:08 -0800662void mem_cgroup_uncharge_page(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800663{
Hugh Dickins82895462008-03-04 14:29:08 -0800664 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800665 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800666 struct mem_cgroup_per_zone *mz;
Balbir Singh66e17072008-02-07 00:13:56 -0800667 unsigned long flags;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800668
Balbir Singh40779602008-04-04 14:29:59 -0700669 if (mem_cgroup_subsys.disabled)
670 return;
671
Balbir Singh8697d332008-02-07 00:13:59 -0800672 /*
Balbir Singh3c541e12008-02-07 00:14:41 -0800673 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -0800674 */
Hugh Dickins82895462008-03-04 14:29:08 -0800675 lock_page_cgroup(page);
676 pc = page_get_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800677 if (!pc)
Hugh Dickins82895462008-03-04 14:29:08 -0800678 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800679
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800680 VM_BUG_ON(pc->page != page);
681 VM_BUG_ON(pc->ref_cnt <= 0);
682
683 if (--(pc->ref_cnt) == 0) {
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800684 mz = page_cgroup_zoneinfo(pc);
685 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700686 __mem_cgroup_remove_list(mz, pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800687 spin_unlock_irqrestore(&mz->lru_lock, flags);
688
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800689 page_assign_page_cgroup(page, NULL);
690 unlock_page_cgroup(page);
691
Hugh Dickins6d48ff82008-03-04 14:29:12 -0800692 mem = pc->mem_cgroup;
693 res_counter_uncharge(&mem->res, PAGE_SIZE);
694 css_put(&mem->css);
695
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700696 kmem_cache_free(page_cgroup_cache, pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800697 return;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800698 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800699
Hugh Dickins82895462008-03-04 14:29:08 -0800700unlock:
Balbir Singh3c541e12008-02-07 00:14:41 -0800701 unlock_page_cgroup(page);
702}
703
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800704/*
705 * Returns non-zero if a page (under migration) has valid page_cgroup member.
706 * Refcnt of page_cgroup is incremented.
707 */
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800708int mem_cgroup_prepare_migration(struct page *page)
709{
710 struct page_cgroup *pc;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800711
Balbir Singh40779602008-04-04 14:29:59 -0700712 if (mem_cgroup_subsys.disabled)
713 return 0;
714
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800715 lock_page_cgroup(page);
716 pc = page_get_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800717 if (pc)
718 pc->ref_cnt++;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800719 unlock_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800720 return pc != NULL;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800721}
722
723void mem_cgroup_end_migration(struct page *page)
724{
Hugh Dickins82895462008-03-04 14:29:08 -0800725 mem_cgroup_uncharge_page(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800726}
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800727
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800728/*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800729 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800730 * And no race with uncharge() routines because page_cgroup for *page*
731 * has extra one reference by mem_cgroup_prepare_migration.
732 */
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800733void mem_cgroup_page_migration(struct page *page, struct page *newpage)
734{
735 struct page_cgroup *pc;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800736 struct mem_cgroup_per_zone *mz;
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800737 unsigned long flags;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800738
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800739 lock_page_cgroup(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800740 pc = page_get_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800741 if (!pc) {
742 unlock_page_cgroup(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800743 return;
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800744 }
745
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800746 mz = page_cgroup_zoneinfo(pc);
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800747 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700748 __mem_cgroup_remove_list(mz, pc);
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800749 spin_unlock_irqrestore(&mz->lru_lock, flags);
750
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800751 page_assign_page_cgroup(page, NULL);
752 unlock_page_cgroup(page);
753
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800754 pc->page = newpage;
755 lock_page_cgroup(newpage);
756 page_assign_page_cgroup(newpage, pc);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800757
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800758 mz = page_cgroup_zoneinfo(pc);
759 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki3eae90c2008-04-29 01:00:22 -0700760 __mem_cgroup_add_list(mz, pc);
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800761 spin_unlock_irqrestore(&mz->lru_lock, flags);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800762
763 unlock_page_cgroup(newpage);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800764}
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800765
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800766/*
767 * This routine traverse page_cgroup in given list and drop them all.
768 * This routine ignores page_cgroup->ref_cnt.
769 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
770 */
771#define FORCE_UNCHARGE_BATCH (128)
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800772static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800773 struct mem_cgroup_per_zone *mz,
774 int active)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800775{
776 struct page_cgroup *pc;
777 struct page *page;
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800778 int count = FORCE_UNCHARGE_BATCH;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800779 unsigned long flags;
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800780 struct list_head *list;
781
782 if (active)
783 list = &mz->active_list;
784 else
785 list = &mz->inactive_list;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800786
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800787 spin_lock_irqsave(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800788 while (!list_empty(list)) {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800789 pc = list_entry(list->prev, struct page_cgroup, lru);
790 page = pc->page;
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800791 get_page(page);
792 spin_unlock_irqrestore(&mz->lru_lock, flags);
793 mem_cgroup_uncharge_page(page);
794 put_page(page);
795 if (--count <= 0) {
796 count = FORCE_UNCHARGE_BATCH;
797 cond_resched();
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800798 }
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800799 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800800 }
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800801 spin_unlock_irqrestore(&mz->lru_lock, flags);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800802}
803
804/*
805 * make mem_cgroup's charge to be 0 if there is no task.
806 * This enables deleting this mem_cgroup.
807 */
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800808static int mem_cgroup_force_empty(struct mem_cgroup *mem)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800809{
810 int ret = -EBUSY;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800811 int node, zid;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800812
Balbir Singh40779602008-04-04 14:29:59 -0700813 if (mem_cgroup_subsys.disabled)
814 return 0;
815
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800816 css_get(&mem->css);
817 /*
818 * page reclaim code (kswapd etc..) will move pages between
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800819 * active_list <-> inactive_list while we don't take a lock.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800820 * So, we have to do loop here until all lists are empty.
821 */
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800822 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800823 if (atomic_read(&mem->css.cgroup->count) > 0)
824 goto out;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800825 for_each_node_state(node, N_POSSIBLE)
826 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
827 struct mem_cgroup_per_zone *mz;
828 mz = mem_cgroup_zoneinfo(mem, node, zid);
829 /* drop all page_cgroup in active_list */
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800830 mem_cgroup_force_empty_list(mem, mz, 1);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800831 /* drop all page_cgroup in inactive_list */
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800832 mem_cgroup_force_empty_list(mem, mz, 0);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800833 }
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800834 }
835 ret = 0;
836out:
837 css_put(&mem->css);
838 return ret;
839}
840
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800841static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
Balbir Singh0eea1032008-02-07 00:13:57 -0800842{
843 *tmp = memparse(buf, &buf);
844 if (*buf != '\0')
845 return -EINVAL;
846
847 /*
848 * Round up the value to the closest page size
849 */
850 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
851 return 0;
852}
853
Paul Menage2c3daa72008-04-29 00:59:58 -0700854static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800855{
Paul Menage2c3daa72008-04-29 00:59:58 -0700856 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
857 cft->private);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800858}
859
860static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
861 struct file *file, const char __user *userbuf,
862 size_t nbytes, loff_t *ppos)
863{
864 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
Balbir Singh0eea1032008-02-07 00:13:57 -0800865 cft->private, userbuf, nbytes, ppos,
866 mem_cgroup_write_strategy);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800867}
868
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -0700869static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700870{
871 struct mem_cgroup *mem;
872
873 mem = mem_cgroup_from_cont(cont);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -0700874 switch (event) {
875 case RES_MAX_USAGE:
876 res_counter_reset_max(&mem->res);
877 break;
878 case RES_FAILCNT:
879 res_counter_reset_failcnt(&mem->res);
880 break;
881 }
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700882 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700883}
884
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700885static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800886{
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700887 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800888}
889
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800890static const struct mem_cgroup_stat_desc {
891 const char *msg;
892 u64 unit;
893} mem_cgroup_stat_desc[] = {
894 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
895 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
Balaji Rao55e462b2008-05-01 04:35:12 -0700896 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
897 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800898};
899
Paul Menagec64745c2008-04-29 01:00:02 -0700900static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
901 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800902{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800903 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
904 struct mem_cgroup_stat *stat = &mem_cont->stat;
905 int i;
906
907 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
908 s64 val;
909
910 val = mem_cgroup_read_stat(stat, i);
911 val *= mem_cgroup_stat_desc[i].unit;
Paul Menagec64745c2008-04-29 01:00:02 -0700912 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800913 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800914 /* showing # of active pages */
915 {
916 unsigned long active, inactive;
917
918 inactive = mem_cgroup_get_all_zonestat(mem_cont,
919 MEM_CGROUP_ZSTAT_INACTIVE);
920 active = mem_cgroup_get_all_zonestat(mem_cont,
921 MEM_CGROUP_ZSTAT_ACTIVE);
Paul Menagec64745c2008-04-29 01:00:02 -0700922 cb->fill(cb, "active", (active) * PAGE_SIZE);
923 cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800924 }
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800925 return 0;
926}
927
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800928static struct cftype mem_cgroup_files[] = {
929 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800930 .name = "usage_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800931 .private = RES_USAGE,
Paul Menage2c3daa72008-04-29 00:59:58 -0700932 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800933 },
934 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700935 .name = "max_usage_in_bytes",
936 .private = RES_MAX_USAGE,
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -0700937 .trigger = mem_cgroup_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700938 .read_u64 = mem_cgroup_read,
939 },
940 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800941 .name = "limit_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800942 .private = RES_LIMIT,
943 .write = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -0700944 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800945 },
946 {
947 .name = "failcnt",
948 .private = RES_FAILCNT,
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -0700949 .trigger = mem_cgroup_reset,
Paul Menage2c3daa72008-04-29 00:59:58 -0700950 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800951 },
Balbir Singh8697d332008-02-07 00:13:59 -0800952 {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800953 .name = "force_empty",
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700954 .trigger = mem_force_empty_write,
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800955 },
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800956 {
957 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -0700958 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800959 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800960};
961
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800962static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
963{
964 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800965 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -0700966 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800967 /*
968 * This routine is called against possible nodes.
969 * But it's BUG to call kmalloc() against offline node.
970 *
971 * TODO: this routine can waste much memory for nodes which will
972 * never be onlined. It's better to use memory hotplug callback
973 * function.
974 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -0700975 if (!node_state(node, N_NORMAL_MEMORY))
976 tmp = -1;
977 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800978 if (!pn)
979 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800980
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800981 mem->info.nodeinfo[node] = pn;
982 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800983
984 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
985 mz = &pn->zoneinfo[zone];
986 INIT_LIST_HEAD(&mz->active_list);
987 INIT_LIST_HEAD(&mz->inactive_list);
KAMEZAWA Hiroyuki072c56c2008-02-07 00:14:39 -0800988 spin_lock_init(&mz->lru_lock);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800989 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800990 return 0;
991}
992
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800993static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
994{
995 kfree(mem->info.nodeinfo[node]);
996}
997
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -0700998static struct mem_cgroup *mem_cgroup_alloc(void)
999{
1000 struct mem_cgroup *mem;
1001
1002 if (sizeof(*mem) < PAGE_SIZE)
1003 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1004 else
1005 mem = vmalloc(sizeof(*mem));
1006
1007 if (mem)
1008 memset(mem, 0, sizeof(*mem));
1009 return mem;
1010}
1011
1012static void mem_cgroup_free(struct mem_cgroup *mem)
1013{
1014 if (sizeof(*mem) < PAGE_SIZE)
1015 kfree(mem);
1016 else
1017 vfree(mem);
1018}
1019
1020
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001021static struct cgroup_subsys_state *
1022mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1023{
1024 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001025 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001026
Balbir Singhb6ac57d2008-04-29 01:00:19 -07001027 if (unlikely((cont->parent) == NULL)) {
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001028 mem = &init_mem_cgroup;
Balbir Singhb6ac57d2008-04-29 01:00:19 -07001029 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
1030 } else {
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07001031 mem = mem_cgroup_alloc();
1032 if (!mem)
1033 return ERR_PTR(-ENOMEM);
Balbir Singhb6ac57d2008-04-29 01:00:19 -07001034 }
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001035
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001036 res_counter_init(&mem->res);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001037
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001038 for_each_node_state(node, N_POSSIBLE)
1039 if (alloc_mem_cgroup_per_zone_info(mem, node))
1040 goto free_out;
1041
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001042 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001043free_out:
1044 for_each_node_state(node, N_POSSIBLE)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001045 free_mem_cgroup_per_zone_info(mem, node);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001046 if (cont->parent != NULL)
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07001047 mem_cgroup_free(mem);
Li Zefan2dda81c2008-02-23 15:24:14 -08001048 return ERR_PTR(-ENOMEM);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001049}
1050
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08001051static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1052 struct cgroup *cont)
1053{
1054 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1055 mem_cgroup_force_empty(mem);
1056}
1057
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001058static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1059 struct cgroup *cont)
1060{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001061 int node;
1062 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1063
1064 for_each_node_state(node, N_POSSIBLE)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001065 free_mem_cgroup_per_zone_info(mem, node);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001066
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07001067 mem_cgroup_free(mem_cgroup_from_cont(cont));
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001068}
1069
1070static int mem_cgroup_populate(struct cgroup_subsys *ss,
1071 struct cgroup *cont)
1072{
Balbir Singh40779602008-04-04 14:29:59 -07001073 if (mem_cgroup_subsys.disabled)
1074 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001075 return cgroup_add_files(cont, ss, mem_cgroup_files,
1076 ARRAY_SIZE(mem_cgroup_files));
1077}
1078
Balbir Singh67e465a2008-02-07 00:13:54 -08001079static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1080 struct cgroup *cont,
1081 struct cgroup *old_cont,
1082 struct task_struct *p)
1083{
1084 struct mm_struct *mm;
1085 struct mem_cgroup *mem, *old_mem;
1086
Balbir Singh40779602008-04-04 14:29:59 -07001087 if (mem_cgroup_subsys.disabled)
1088 return;
1089
Balbir Singh67e465a2008-02-07 00:13:54 -08001090 mm = get_task_mm(p);
1091 if (mm == NULL)
1092 return;
1093
1094 mem = mem_cgroup_from_cont(cont);
1095 old_mem = mem_cgroup_from_cont(old_cont);
1096
1097 if (mem == old_mem)
1098 goto out;
1099
1100 /*
1101 * Only thread group leaders are allowed to migrate, the mm_struct is
1102 * in effect owned by the leader
1103 */
Pavel Emelyanov52ea27e2008-03-19 17:00:45 -07001104 if (!thread_group_leader(p))
Balbir Singh67e465a2008-02-07 00:13:54 -08001105 goto out;
1106
Balbir Singh67e465a2008-02-07 00:13:54 -08001107out:
1108 mmput(mm);
Balbir Singh67e465a2008-02-07 00:13:54 -08001109}
1110
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001111struct cgroup_subsys mem_cgroup_subsys = {
1112 .name = "memory",
1113 .subsys_id = mem_cgroup_subsys_id,
1114 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08001115 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001116 .destroy = mem_cgroup_destroy,
1117 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08001118 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001119 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001120};