blob: dc3472f9f68c54f884ec4b62b39ed0f9a75591ce [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>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080034
Balbir Singh8697d332008-02-07 00:13:59 -080035#include <asm/uaccess.h>
36
Balbir Singh8cdea7c2008-02-07 00:13:50 -080037struct cgroup_subsys mem_cgroup_subsys;
Balbir Singh66e17072008-02-07 00:13:56 -080038static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
Balbir Singhb6ac57d2008-04-29 01:00:19 -070039static struct kmem_cache *page_cgroup_cache;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080040
41/*
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080042 * Statistics for memory cgroup.
43 */
44enum mem_cgroup_stat_index {
45 /*
46 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
47 */
48 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
49 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
50
51 MEM_CGROUP_STAT_NSTATS,
52};
53
54struct mem_cgroup_stat_cpu {
55 s64 count[MEM_CGROUP_STAT_NSTATS];
56} ____cacheline_aligned_in_smp;
57
58struct mem_cgroup_stat {
59 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
60};
61
62/*
63 * For accounting under irq disable, no need for increment preempt count.
64 */
65static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
66 enum mem_cgroup_stat_index idx, int val)
67{
68 int cpu = smp_processor_id();
69 stat->cpustat[cpu].count[idx] += val;
70}
71
72static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
73 enum mem_cgroup_stat_index idx)
74{
75 int cpu;
76 s64 ret = 0;
77 for_each_possible_cpu(cpu)
78 ret += stat->cpustat[cpu].count[idx];
79 return ret;
80}
81
82/*
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -080083 * per-zone information in memory controller.
84 */
85
86enum mem_cgroup_zstat_index {
87 MEM_CGROUP_ZSTAT_ACTIVE,
88 MEM_CGROUP_ZSTAT_INACTIVE,
89
90 NR_MEM_CGROUP_ZSTAT,
91};
92
93struct mem_cgroup_per_zone {
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -080094 /*
95 * spin_lock to protect the per cgroup LRU
96 */
97 spinlock_t lru_lock;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -080098 struct list_head active_list;
99 struct list_head inactive_list;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800100 unsigned long count[NR_MEM_CGROUP_ZSTAT];
101};
102/* Macro for accessing counter */
103#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
104
105struct mem_cgroup_per_node {
106 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
107};
108
109struct mem_cgroup_lru_info {
110 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
111};
112
113/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800114 * The memory controller data structure. The memory controller controls both
115 * page cache and RSS per cgroup. We would eventually like to provide
116 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
117 * to help the administrator determine what knobs to tune.
118 *
119 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800120 * we hit the water mark. May be even add a low water mark, such that
121 * no reclaim occurs from a cgroup at it's low water mark, this is
122 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800123 */
124struct mem_cgroup {
125 struct cgroup_subsys_state css;
126 /*
127 * the counter to account for memory usage
128 */
129 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800130 /*
131 * Per cgroup active and inactive list, similar to the
132 * per zone LRU lists.
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800133 */
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800134 struct mem_cgroup_lru_info info;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800135
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800136 int prev_priority; /* for recording reclaim priority */
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800137 /*
138 * statistics.
139 */
140 struct mem_cgroup_stat stat;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800141};
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800142static struct mem_cgroup init_mem_cgroup;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800143
144/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800145 * We use the lower bit of the page->page_cgroup pointer as a bit spin
Hugh Dickins9442ec92008-03-04 14:29:07 -0800146 * lock. We need to ensure that page->page_cgroup is at least two
147 * byte aligned (based on comments from Nick Piggin). But since
148 * bit_spin_lock doesn't actually set that lock bit in a non-debug
149 * uniprocessor kernel, we should avoid setting it here too.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800150 */
151#define PAGE_CGROUP_LOCK_BIT 0x0
Hugh Dickins9442ec92008-03-04 14:29:07 -0800152#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
153#define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
154#else
155#define PAGE_CGROUP_LOCK 0x0
156#endif
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800157
158/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800159 * A page_cgroup page is associated with every page descriptor. The
160 * page_cgroup helps us identify information about the cgroup
161 */
162struct page_cgroup {
163 struct list_head lru; /* per cgroup LRU list */
164 struct page *page;
165 struct mem_cgroup *mem_cgroup;
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800166 int ref_cnt; /* cached, mapped, migrating */
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800167 int flags;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800168};
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800169#define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800170#define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800171
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800172static int page_cgroup_nid(struct page_cgroup *pc)
KAMEZAWA Hiroyukic01495302008-02-07 00:14:30 -0800173{
174 return page_to_nid(pc->page);
175}
176
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800177static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
KAMEZAWA Hiroyukic01495302008-02-07 00:14:30 -0800178{
179 return page_zonenum(pc->page);
180}
181
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800182enum charge_type {
183 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
184 MEM_CGROUP_CHARGE_TYPE_MAPPED,
185};
186
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800187/*
188 * Always modified under lru lock. Then, not necessary to preempt_disable()
189 */
190static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
191 bool charge)
192{
193 int val = (charge)? 1 : -1;
194 struct mem_cgroup_stat *stat = &mem->stat;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800195
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800196 VM_BUG_ON(!irqs_disabled());
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800197 if (flags & PAGE_CGROUP_FLAG_CACHE)
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800198 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800199 else
200 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800201}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800202
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800203static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800204mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
205{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800206 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
207}
208
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800209static struct mem_cgroup_per_zone *
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800210page_cgroup_zoneinfo(struct page_cgroup *pc)
211{
212 struct mem_cgroup *mem = pc->mem_cgroup;
213 int nid = page_cgroup_nid(pc);
214 int zid = page_cgroup_zid(pc);
215
216 return mem_cgroup_zoneinfo(mem, nid, zid);
217}
218
219static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
220 enum mem_cgroup_zstat_index idx)
221{
222 int nid, zid;
223 struct mem_cgroup_per_zone *mz;
224 u64 total = 0;
225
226 for_each_online_node(nid)
227 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
228 mz = mem_cgroup_zoneinfo(mem, nid, zid);
229 total += MEM_CGROUP_ZSTAT(mz, idx);
230 }
231 return total;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800232}
233
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800234static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800235{
236 return container_of(cgroup_subsys_state(cont,
237 mem_cgroup_subsys_id), struct mem_cgroup,
238 css);
239}
240
Balbir Singhcf475ad2008-04-29 01:00:16 -0700241struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800242{
243 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
244 struct mem_cgroup, css);
245}
246
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800247static inline int page_cgroup_locked(struct page *page)
248{
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800249 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800250}
251
Hugh Dickins9442ec92008-03-04 14:29:07 -0800252static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800253{
Hugh Dickins9442ec92008-03-04 14:29:07 -0800254 VM_BUG_ON(!page_cgroup_locked(page));
255 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800256}
257
258struct page_cgroup *page_get_page_cgroup(struct page *page)
259{
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800260 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800261}
262
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800263static void lock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800264{
265 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800266}
267
Hugh Dickins2680eed2008-03-04 14:29:13 -0800268static int try_lock_page_cgroup(struct page *page)
269{
270 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
271}
272
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800273static void unlock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800274{
275 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
276}
277
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800278static void __mem_cgroup_remove_list(struct page_cgroup *pc)
279{
280 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
281 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
282
283 if (from)
284 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
285 else
286 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
287
288 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
289 list_del_init(&pc->lru);
290}
291
292static void __mem_cgroup_add_list(struct page_cgroup *pc)
293{
294 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
295 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
296
297 if (!to) {
298 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800299 list_add(&pc->lru, &mz->inactive_list);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800300 } else {
301 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800302 list_add(&pc->lru, &mz->active_list);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800303 }
304 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
305}
306
Balbir Singh8697d332008-02-07 00:13:59 -0800307static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
Balbir Singh66e17072008-02-07 00:13:56 -0800308{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800309 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
310 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
311
312 if (from)
313 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
314 else
315 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
316
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800317 if (active) {
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800318 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800319 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800320 list_move(&pc->lru, &mz->active_list);
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800321 } else {
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800322 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800323 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800324 list_move(&pc->lru, &mz->inactive_list);
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800325 }
Balbir Singh66e17072008-02-07 00:13:56 -0800326}
327
David Rientjes4c4a2212008-02-07 00:14:06 -0800328int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
329{
330 int ret;
331
332 task_lock(task);
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800333 ret = task->mm && mm_match_cgroup(task->mm, mem);
David Rientjes4c4a2212008-02-07 00:14:06 -0800334 task_unlock(task);
335 return ret;
336}
337
Balbir Singh66e17072008-02-07 00:13:56 -0800338/*
339 * This routine assumes that the appropriate zone's lru lock is already held
340 */
Hugh Dickins427d5412008-03-04 14:29:03 -0800341void mem_cgroup_move_lists(struct page *page, bool active)
Balbir Singh66e17072008-02-07 00:13:56 -0800342{
Hugh Dickins427d5412008-03-04 14:29:03 -0800343 struct page_cgroup *pc;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800344 struct mem_cgroup_per_zone *mz;
345 unsigned long flags;
346
Hugh Dickins2680eed2008-03-04 14:29:13 -0800347 /*
348 * We cannot lock_page_cgroup while holding zone's lru_lock,
349 * because other holders of lock_page_cgroup can be interrupted
350 * with an attempt to rotate_reclaimable_page. But we cannot
351 * safely get to page_cgroup without it, so just try_lock it:
352 * mem_cgroup_isolate_pages allows for page left on wrong list.
353 */
354 if (!try_lock_page_cgroup(page))
Balbir Singh66e17072008-02-07 00:13:56 -0800355 return;
356
Hugh Dickins2680eed2008-03-04 14:29:13 -0800357 pc = page_get_page_cgroup(page);
358 if (pc) {
Hugh Dickins2680eed2008-03-04 14:29:13 -0800359 mz = page_cgroup_zoneinfo(pc);
Hugh Dickins2680eed2008-03-04 14:29:13 -0800360 spin_lock_irqsave(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800361 __mem_cgroup_move_lists(pc, active);
Hugh Dickins2680eed2008-03-04 14:29:13 -0800362 spin_unlock_irqrestore(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800363 }
364 unlock_page_cgroup(page);
Balbir Singh66e17072008-02-07 00:13:56 -0800365}
366
KAMEZAWA Hiroyuki58ae83d2008-02-07 00:14:32 -0800367/*
368 * Calculate mapped_ratio under memory controller. This will be used in
369 * vmscan.c for deteremining we have to reclaim mapped pages.
370 */
371int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
372{
373 long total, rss;
374
375 /*
376 * usage is recorded in bytes. But, here, we assume the number of
377 * physical pages can be represented by "long" on any arch.
378 */
379 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
380 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
381 return (int)((rss * 100L) / total);
382}
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800383
KAMEZAWA Hiroyuki5932f362008-02-07 00:14:33 -0800384/*
385 * This function is called from vmscan.c. In page reclaiming loop. balance
386 * between active and inactive list is calculated. For memory controller
387 * page reclaiming, we should use using mem_cgroup's imbalance rather than
388 * zone's global lru imbalance.
389 */
390long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
391{
392 unsigned long active, inactive;
393 /* active and inactive are the number of pages. 'long' is ok.*/
394 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
395 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
396 return (long) (active / (inactive + 1));
397}
KAMEZAWA Hiroyuki58ae83d2008-02-07 00:14:32 -0800398
KAMEZAWA Hiroyuki6c48a1d2008-02-07 00:14:34 -0800399/*
400 * prev_priority control...this will be used in memory reclaim path.
401 */
402int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
403{
404 return mem->prev_priority;
405}
406
407void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
408{
409 if (priority < mem->prev_priority)
410 mem->prev_priority = priority;
411}
412
413void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
414{
415 mem->prev_priority = priority;
416}
417
KAMEZAWA Hiroyukicc381082008-02-07 00:14:35 -0800418/*
419 * Calculate # of pages to be scanned in this priority/zone.
420 * See also vmscan.c
421 *
422 * priority starts from "DEF_PRIORITY" and decremented in each loop.
423 * (see include/linux/mmzone.h)
424 */
425
426long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
427 struct zone *zone, int priority)
428{
429 long nr_active;
430 int nid = zone->zone_pgdat->node_id;
431 int zid = zone_idx(zone);
432 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
433
434 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
435 return (nr_active >> priority);
436}
437
438long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
439 struct zone *zone, int priority)
440{
441 long nr_inactive;
442 int nid = zone->zone_pgdat->node_id;
443 int zid = zone_idx(zone);
444 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
445
446 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
KAMEZAWA Hiroyukicc381082008-02-07 00:14:35 -0800447 return (nr_inactive >> priority);
448}
449
Balbir Singh66e17072008-02-07 00:13:56 -0800450unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
451 struct list_head *dst,
452 unsigned long *scanned, int order,
453 int mode, struct zone *z,
454 struct mem_cgroup *mem_cont,
455 int active)
456{
457 unsigned long nr_taken = 0;
458 struct page *page;
459 unsigned long scan;
460 LIST_HEAD(pc_list);
461 struct list_head *src;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800462 struct page_cgroup *pc, *tmp;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800463 int nid = z->zone_pgdat->node_id;
464 int zid = zone_idx(z);
465 struct mem_cgroup_per_zone *mz;
Balbir Singh66e17072008-02-07 00:13:56 -0800466
Balbir Singhcf475ad2008-04-29 01:00:16 -0700467 BUG_ON(!mem_cont);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800468 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
Balbir Singh66e17072008-02-07 00:13:56 -0800469 if (active)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800470 src = &mz->active_list;
Balbir Singh66e17072008-02-07 00:13:56 -0800471 else
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800472 src = &mz->inactive_list;
473
Balbir Singh66e17072008-02-07 00:13:56 -0800474
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800475 spin_lock(&mz->lru_lock);
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800476 scan = 0;
477 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
Hugh Dickins436c65412008-02-07 00:14:12 -0800478 if (scan >= nr_to_scan)
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800479 break;
Balbir Singh66e17072008-02-07 00:13:56 -0800480 page = pc->page;
Balbir Singh66e17072008-02-07 00:13:56 -0800481
Hugh Dickins436c65412008-02-07 00:14:12 -0800482 if (unlikely(!PageLRU(page)))
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800483 continue;
KAMEZAWA Hiroyukiff7283f2008-02-07 00:14:11 -0800484
Balbir Singh66e17072008-02-07 00:13:56 -0800485 if (PageActive(page) && !active) {
486 __mem_cgroup_move_lists(pc, true);
Balbir Singh66e17072008-02-07 00:13:56 -0800487 continue;
488 }
489 if (!PageActive(page) && active) {
490 __mem_cgroup_move_lists(pc, false);
Balbir Singh66e17072008-02-07 00:13:56 -0800491 continue;
492 }
493
Hugh Dickins436c65412008-02-07 00:14:12 -0800494 scan++;
495 list_move(&pc->lru, &pc_list);
Balbir Singh66e17072008-02-07 00:13:56 -0800496
497 if (__isolate_lru_page(page, mode) == 0) {
498 list_move(&page->lru, dst);
499 nr_taken++;
500 }
501 }
502
503 list_splice(&pc_list, src);
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800504 spin_unlock(&mz->lru_lock);
Balbir Singh66e17072008-02-07 00:13:56 -0800505
506 *scanned = scan;
507 return nr_taken;
508}
509
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800510/*
511 * Charge the memory controller for page usage.
512 * Return
513 * 0 if the charge was successful
514 * < 0 if the cgroup is over its limit
515 */
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800516static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
517 gfp_t gfp_mask, enum charge_type ctype)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800518{
519 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800520 struct page_cgroup *pc;
Balbir Singh66e17072008-02-07 00:13:56 -0800521 unsigned long flags;
522 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800523 struct mem_cgroup_per_zone *mz;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800524
Balbir Singh40779602008-04-04 14:29:59 -0700525 if (mem_cgroup_subsys.disabled)
526 return 0;
527
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800528 /*
529 * Should page_cgroup's go to their own slab?
530 * One could optimize the performance of the charging routine
531 * by saving a bit in the page_flags and using it as a lock
532 * to see if the cgroup page already has a page_cgroup associated
533 * with it
534 */
Balbir Singh66e17072008-02-07 00:13:56 -0800535retry:
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800536 lock_page_cgroup(page);
537 pc = page_get_page_cgroup(page);
538 /*
539 * The page_cgroup exists and
540 * the page has already been accounted.
541 */
542 if (pc) {
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800543 VM_BUG_ON(pc->page != page);
544 VM_BUG_ON(pc->ref_cnt <= 0);
545
546 pc->ref_cnt++;
547 unlock_page_cgroup(page);
548 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800549 }
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800550 unlock_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800551
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700552 pc = kmem_cache_zalloc(page_cgroup_cache, gfp_mask);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800553 if (pc == NULL)
554 goto err;
555
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800556 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800557 * We always charge the cgroup the mm_struct belongs to.
558 * The mm_struct's mem_cgroup changes on task migration if the
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800559 * thread group leader migrates. It's possible that mm is not
560 * set, if so charge the init_mm (happens for pagecache usage).
561 */
562 if (!mm)
563 mm = &init_mm;
564
Hugh Dickins3be91272008-02-07 00:14:19 -0800565 rcu_read_lock();
Balbir Singhcf475ad2008-04-29 01:00:16 -0700566 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800567 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800568 * For every charge from the cgroup, increment reference count
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800569 */
570 css_get(&mem->css);
571 rcu_read_unlock();
572
Balbir Singh0eea1032008-02-07 00:13:57 -0800573 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
Hugh Dickins3be91272008-02-07 00:14:19 -0800574 if (!(gfp_mask & __GFP_WAIT))
575 goto out;
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800576
577 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
Balbir Singh66e17072008-02-07 00:13:56 -0800578 continue;
579
580 /*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800581 * try_to_free_mem_cgroup_pages() might not give us a full
582 * picture of reclaim. Some pages are reclaimed and might be
583 * moved to swap cache or just unmapped from the cgroup.
584 * Check the limit again to see if the reclaim reduced the
585 * current usage of the cgroup before giving up
586 */
Balbir Singh66e17072008-02-07 00:13:56 -0800587 if (res_counter_check_under_limit(&mem->res))
588 continue;
Hugh Dickins3be91272008-02-07 00:14:19 -0800589
590 if (!nr_retries--) {
591 mem_cgroup_out_of_memory(mem, gfp_mask);
592 goto out;
Balbir Singh66e17072008-02-07 00:13:56 -0800593 }
Hugh Dickins3be91272008-02-07 00:14:19 -0800594 congestion_wait(WRITE, HZ/10);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800595 }
596
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800597 pc->ref_cnt = 1;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800598 pc->mem_cgroup = mem;
599 pc->page = page;
KAMEZAWA Hiroyuki3564c7c2008-02-07 00:14:23 -0800600 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800601 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
602 pc->flags |= PAGE_CGROUP_FLAG_CACHE;
Hugh Dickins3be91272008-02-07 00:14:19 -0800603
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800604 lock_page_cgroup(page);
605 if (page_get_page_cgroup(page)) {
606 unlock_page_cgroup(page);
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800607 /*
Hugh Dickins3be91272008-02-07 00:14:19 -0800608 * Another charge has been added to this page already.
609 * We take lock_page_cgroup(page) again and read
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800610 * page->cgroup, increment refcnt.... just retry is OK.
611 */
612 res_counter_uncharge(&mem->res, PAGE_SIZE);
613 css_put(&mem->css);
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700614 kmem_cache_free(page_cgroup_cache, pc);
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800615 goto retry;
616 }
Hugh Dickins7e924aa2008-03-04 14:29:08 -0800617 page_assign_page_cgroup(page, pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800618
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800619 mz = page_cgroup_zoneinfo(pc);
620 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800621 __mem_cgroup_add_list(pc);
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800622 spin_unlock_irqrestore(&mz->lru_lock, flags);
Balbir Singh66e17072008-02-07 00:13:56 -0800623
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800624 unlock_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800625done:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800626 return 0;
Hugh Dickins3be91272008-02-07 00:14:19 -0800627out:
628 css_put(&mem->css);
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700629 kmem_cache_free(page_cgroup_cache, pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800630err:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800631 return -ENOMEM;
632}
633
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800634int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800635{
636 return mem_cgroup_charge_common(page, mm, gfp_mask,
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800637 MEM_CGROUP_CHARGE_TYPE_MAPPED);
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800638}
639
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800640int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
641 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -0800642{
Balbir Singh8697d332008-02-07 00:13:59 -0800643 if (!mm)
644 mm = &init_mm;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800645 return mem_cgroup_charge_common(page, mm, gfp_mask,
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800646 MEM_CGROUP_CHARGE_TYPE_CACHE);
Balbir Singh8697d332008-02-07 00:13:59 -0800647}
648
649/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800650 * Uncharging is always a welcome operation, we never complain, simply
Hugh Dickins82895462008-03-04 14:29:08 -0800651 * uncharge.
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800652 */
Hugh Dickins82895462008-03-04 14:29:08 -0800653void mem_cgroup_uncharge_page(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800654{
Hugh Dickins82895462008-03-04 14:29:08 -0800655 struct page_cgroup *pc;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800656 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800657 struct mem_cgroup_per_zone *mz;
Balbir Singh66e17072008-02-07 00:13:56 -0800658 unsigned long flags;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800659
Balbir Singh40779602008-04-04 14:29:59 -0700660 if (mem_cgroup_subsys.disabled)
661 return;
662
Balbir Singh8697d332008-02-07 00:13:59 -0800663 /*
Balbir Singh3c541e12008-02-07 00:14:41 -0800664 * Check if our page_cgroup is valid
Balbir Singh8697d332008-02-07 00:13:59 -0800665 */
Hugh Dickins82895462008-03-04 14:29:08 -0800666 lock_page_cgroup(page);
667 pc = page_get_page_cgroup(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800668 if (!pc)
Hugh Dickins82895462008-03-04 14:29:08 -0800669 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800670
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800671 VM_BUG_ON(pc->page != page);
672 VM_BUG_ON(pc->ref_cnt <= 0);
673
674 if (--(pc->ref_cnt) == 0) {
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800675 mz = page_cgroup_zoneinfo(pc);
676 spin_lock_irqsave(&mz->lru_lock, flags);
677 __mem_cgroup_remove_list(pc);
678 spin_unlock_irqrestore(&mz->lru_lock, flags);
679
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800680 page_assign_page_cgroup(page, NULL);
681 unlock_page_cgroup(page);
682
Hugh Dickins6d48ff82008-03-04 14:29:12 -0800683 mem = pc->mem_cgroup;
684 res_counter_uncharge(&mem->res, PAGE_SIZE);
685 css_put(&mem->css);
686
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700687 kmem_cache_free(page_cgroup_cache, pc);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800688 return;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800689 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800690
Hugh Dickins82895462008-03-04 14:29:08 -0800691unlock:
Balbir Singh3c541e12008-02-07 00:14:41 -0800692 unlock_page_cgroup(page);
693}
694
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800695/*
696 * Returns non-zero if a page (under migration) has valid page_cgroup member.
697 * Refcnt of page_cgroup is incremented.
698 */
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800699int mem_cgroup_prepare_migration(struct page *page)
700{
701 struct page_cgroup *pc;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800702
Balbir Singh40779602008-04-04 14:29:59 -0700703 if (mem_cgroup_subsys.disabled)
704 return 0;
705
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800706 lock_page_cgroup(page);
707 pc = page_get_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800708 if (pc)
709 pc->ref_cnt++;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800710 unlock_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800711 return pc != NULL;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800712}
713
714void mem_cgroup_end_migration(struct page *page)
715{
Hugh Dickins82895462008-03-04 14:29:08 -0800716 mem_cgroup_uncharge_page(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800717}
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800718
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800719/*
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800720 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800721 * And no race with uncharge() routines because page_cgroup for *page*
722 * has extra one reference by mem_cgroup_prepare_migration.
723 */
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800724void mem_cgroup_page_migration(struct page *page, struct page *newpage)
725{
726 struct page_cgroup *pc;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800727 struct mem_cgroup_per_zone *mz;
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800728 unsigned long flags;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800729
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800730 lock_page_cgroup(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800731 pc = page_get_page_cgroup(page);
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800732 if (!pc) {
733 unlock_page_cgroup(page);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800734 return;
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800735 }
736
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800737 mz = page_cgroup_zoneinfo(pc);
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800738 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800739 __mem_cgroup_remove_list(pc);
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800740 spin_unlock_irqrestore(&mz->lru_lock, flags);
741
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800742 page_assign_page_cgroup(page, NULL);
743 unlock_page_cgroup(page);
744
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800745 pc->page = newpage;
746 lock_page_cgroup(newpage);
747 page_assign_page_cgroup(newpage, pc);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800748
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800749 mz = page_cgroup_zoneinfo(pc);
750 spin_lock_irqsave(&mz->lru_lock, flags);
751 __mem_cgroup_add_list(pc);
752 spin_unlock_irqrestore(&mz->lru_lock, flags);
Hugh Dickinsfb59e9f2008-03-04 14:29:16 -0800753
754 unlock_page_cgroup(newpage);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800755}
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800756
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800757/*
758 * This routine traverse page_cgroup in given list and drop them all.
759 * This routine ignores page_cgroup->ref_cnt.
760 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
761 */
762#define FORCE_UNCHARGE_BATCH (128)
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800763static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800764 struct mem_cgroup_per_zone *mz,
765 int active)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800766{
767 struct page_cgroup *pc;
768 struct page *page;
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800769 int count = FORCE_UNCHARGE_BATCH;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800770 unsigned long flags;
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800771 struct list_head *list;
772
773 if (active)
774 list = &mz->active_list;
775 else
776 list = &mz->inactive_list;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800777
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800778 spin_lock_irqsave(&mz->lru_lock, flags);
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800779 while (!list_empty(list)) {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800780 pc = list_entry(list->prev, struct page_cgroup, lru);
781 page = pc->page;
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800782 get_page(page);
783 spin_unlock_irqrestore(&mz->lru_lock, flags);
784 mem_cgroup_uncharge_page(page);
785 put_page(page);
786 if (--count <= 0) {
787 count = FORCE_UNCHARGE_BATCH;
788 cond_resched();
Hugh Dickinsb9c565d2008-03-04 14:29:11 -0800789 }
Hirokazu Takahashi9b3c0a02008-03-04 14:29:15 -0800790 spin_lock_irqsave(&mz->lru_lock, flags);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800791 }
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800792 spin_unlock_irqrestore(&mz->lru_lock, flags);
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800793}
794
795/*
796 * make mem_cgroup's charge to be 0 if there is no task.
797 * This enables deleting this mem_cgroup.
798 */
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800799static int mem_cgroup_force_empty(struct mem_cgroup *mem)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800800{
801 int ret = -EBUSY;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800802 int node, zid;
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800803
Balbir Singh40779602008-04-04 14:29:59 -0700804 if (mem_cgroup_subsys.disabled)
805 return 0;
806
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800807 css_get(&mem->css);
808 /*
809 * page reclaim code (kswapd etc..) will move pages between
Hugh Dickins8869b8f2008-03-04 14:29:09 -0800810 * active_list <-> inactive_list while we don't take a lock.
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800811 * So, we have to do loop here until all lists are empty.
812 */
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800813 while (mem->res.usage > 0) {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800814 if (atomic_read(&mem->css.cgroup->count) > 0)
815 goto out;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800816 for_each_node_state(node, N_POSSIBLE)
817 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
818 struct mem_cgroup_per_zone *mz;
819 mz = mem_cgroup_zoneinfo(mem, node, zid);
820 /* drop all page_cgroup in active_list */
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800821 mem_cgroup_force_empty_list(mem, mz, 1);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800822 /* drop all page_cgroup in inactive_list */
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800823 mem_cgroup_force_empty_list(mem, mz, 0);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800824 }
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800825 }
826 ret = 0;
827out:
828 css_put(&mem->css);
829 return ret;
830}
831
Hugh Dickinsd5b69e32008-03-04 14:29:10 -0800832static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
Balbir Singh0eea1032008-02-07 00:13:57 -0800833{
834 *tmp = memparse(buf, &buf);
835 if (*buf != '\0')
836 return -EINVAL;
837
838 /*
839 * Round up the value to the closest page size
840 */
841 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
842 return 0;
843}
844
Paul Menage2c3daa72008-04-29 00:59:58 -0700845static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800846{
Paul Menage2c3daa72008-04-29 00:59:58 -0700847 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
848 cft->private);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800849}
850
851static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
852 struct file *file, const char __user *userbuf,
853 size_t nbytes, loff_t *ppos)
854{
855 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
Balbir Singh0eea1032008-02-07 00:13:57 -0800856 cft->private, userbuf, nbytes, ppos,
857 mem_cgroup_write_strategy);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800858}
859
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700860static int mem_cgroup_max_reset(struct cgroup *cont, unsigned int event)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700861{
862 struct mem_cgroup *mem;
863
864 mem = mem_cgroup_from_cont(cont);
865 res_counter_reset_max(&mem->res);
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700866 return 0;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700867}
868
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700869static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800870{
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700871 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800872}
873
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800874static const struct mem_cgroup_stat_desc {
875 const char *msg;
876 u64 unit;
877} mem_cgroup_stat_desc[] = {
878 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
879 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
880};
881
Paul Menagec64745c2008-04-29 01:00:02 -0700882static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
883 struct cgroup_map_cb *cb)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800884{
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800885 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
886 struct mem_cgroup_stat *stat = &mem_cont->stat;
887 int i;
888
889 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
890 s64 val;
891
892 val = mem_cgroup_read_stat(stat, i);
893 val *= mem_cgroup_stat_desc[i].unit;
Paul Menagec64745c2008-04-29 01:00:02 -0700894 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800895 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800896 /* showing # of active pages */
897 {
898 unsigned long active, inactive;
899
900 inactive = mem_cgroup_get_all_zonestat(mem_cont,
901 MEM_CGROUP_ZSTAT_INACTIVE);
902 active = mem_cgroup_get_all_zonestat(mem_cont,
903 MEM_CGROUP_ZSTAT_ACTIVE);
Paul Menagec64745c2008-04-29 01:00:02 -0700904 cb->fill(cb, "active", (active) * PAGE_SIZE);
905 cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800906 }
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800907 return 0;
908}
909
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800910static struct cftype mem_cgroup_files[] = {
911 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800912 .name = "usage_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800913 .private = RES_USAGE,
Paul Menage2c3daa72008-04-29 00:59:58 -0700914 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800915 },
916 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700917 .name = "max_usage_in_bytes",
918 .private = RES_MAX_USAGE,
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700919 .trigger = mem_cgroup_max_reset,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -0700920 .read_u64 = mem_cgroup_read,
921 },
922 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800923 .name = "limit_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800924 .private = RES_LIMIT,
925 .write = mem_cgroup_write,
Paul Menage2c3daa72008-04-29 00:59:58 -0700926 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800927 },
928 {
929 .name = "failcnt",
930 .private = RES_FAILCNT,
Paul Menage2c3daa72008-04-29 00:59:58 -0700931 .read_u64 = mem_cgroup_read,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800932 },
Balbir Singh8697d332008-02-07 00:13:59 -0800933 {
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800934 .name = "force_empty",
Pavel Emelyanov85cc59d2008-04-29 01:00:20 -0700935 .trigger = mem_force_empty_write,
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -0800936 },
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800937 {
938 .name = "stat",
Paul Menagec64745c2008-04-29 01:00:02 -0700939 .read_map = mem_control_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -0800940 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800941};
942
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800943static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
944{
945 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800946 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -0700947 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800948 /*
949 * This routine is called against possible nodes.
950 * But it's BUG to call kmalloc() against offline node.
951 *
952 * TODO: this routine can waste much memory for nodes which will
953 * never be onlined. It's better to use memory hotplug callback
954 * function.
955 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -0700956 if (!node_state(node, N_NORMAL_MEMORY))
957 tmp = -1;
958 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800959 if (!pn)
960 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800961
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800962 mem->info.nodeinfo[node] = pn;
963 memset(pn, 0, sizeof(*pn));
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800964
965 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
966 mz = &pn->zoneinfo[zone];
967 INIT_LIST_HEAD(&mz->active_list);
968 INIT_LIST_HEAD(&mz->inactive_list);
KAMEZAWA Hiroyuki072c56c12008-02-07 00:14:39 -0800969 spin_lock_init(&mz->lru_lock);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800970 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800971 return 0;
972}
973
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800974static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
975{
976 kfree(mem->info.nodeinfo[node]);
977}
978
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800979static struct cgroup_subsys_state *
980mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
981{
982 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800983 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800984
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700985 if (unlikely((cont->parent) == NULL)) {
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800986 mem = &init_mem_cgroup;
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700987 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
988 } else {
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800989 mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
Balbir Singhb6ac57d2008-04-29 01:00:19 -0700990 }
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800991
992 if (mem == NULL)
Li Zefan2dda81c2008-02-23 15:24:14 -0800993 return ERR_PTR(-ENOMEM);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800994
995 res_counter_init(&mem->res);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -0800996
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800997 memset(&mem->info, 0, sizeof(mem->info));
998
999 for_each_node_state(node, N_POSSIBLE)
1000 if (alloc_mem_cgroup_per_zone_info(mem, node))
1001 goto free_out;
1002
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001003 return &mem->css;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001004free_out:
1005 for_each_node_state(node, N_POSSIBLE)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001006 free_mem_cgroup_per_zone_info(mem, node);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001007 if (cont->parent != NULL)
1008 kfree(mem);
Li Zefan2dda81c2008-02-23 15:24:14 -08001009 return ERR_PTR(-ENOMEM);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001010}
1011
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08001012static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1013 struct cgroup *cont)
1014{
1015 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1016 mem_cgroup_force_empty(mem);
1017}
1018
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001019static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1020 struct cgroup *cont)
1021{
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001022 int node;
1023 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1024
1025 for_each_node_state(node, N_POSSIBLE)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08001026 free_mem_cgroup_per_zone_info(mem, node);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001027
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001028 kfree(mem_cgroup_from_cont(cont));
1029}
1030
1031static int mem_cgroup_populate(struct cgroup_subsys *ss,
1032 struct cgroup *cont)
1033{
Balbir Singh40779602008-04-04 14:29:59 -07001034 if (mem_cgroup_subsys.disabled)
1035 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001036 return cgroup_add_files(cont, ss, mem_cgroup_files,
1037 ARRAY_SIZE(mem_cgroup_files));
1038}
1039
Balbir Singh67e465a2008-02-07 00:13:54 -08001040static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1041 struct cgroup *cont,
1042 struct cgroup *old_cont,
1043 struct task_struct *p)
1044{
1045 struct mm_struct *mm;
1046 struct mem_cgroup *mem, *old_mem;
1047
Balbir Singh40779602008-04-04 14:29:59 -07001048 if (mem_cgroup_subsys.disabled)
1049 return;
1050
Balbir Singh67e465a2008-02-07 00:13:54 -08001051 mm = get_task_mm(p);
1052 if (mm == NULL)
1053 return;
1054
1055 mem = mem_cgroup_from_cont(cont);
1056 old_mem = mem_cgroup_from_cont(old_cont);
1057
1058 if (mem == old_mem)
1059 goto out;
1060
1061 /*
1062 * Only thread group leaders are allowed to migrate, the mm_struct is
1063 * in effect owned by the leader
1064 */
Pavel Emelyanov52ea27e2008-03-19 17:00:45 -07001065 if (!thread_group_leader(p))
Balbir Singh67e465a2008-02-07 00:13:54 -08001066 goto out;
1067
Balbir Singh67e465a2008-02-07 00:13:54 -08001068out:
1069 mmput(mm);
Balbir Singh67e465a2008-02-07 00:13:54 -08001070}
1071
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001072struct cgroup_subsys mem_cgroup_subsys = {
1073 .name = "memory",
1074 .subsys_id = mem_cgroup_subsys_id,
1075 .create = mem_cgroup_create,
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08001076 .pre_destroy = mem_cgroup_pre_destroy,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001077 .destroy = mem_cgroup_destroy,
1078 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -08001079 .attach = mem_cgroup_move_task,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001080 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001081};