Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
8cdea7c0 BS |
9 | * 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> | |
78fb7466 | 23 | #include <linux/mm.h> |
d52aa412 | 24 | #include <linux/smp.h> |
8a9f3ccd | 25 | #include <linux/page-flags.h> |
66e1707b | 26 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
27 | #include <linux/bit_spinlock.h> |
28 | #include <linux/rcupdate.h> | |
b6ac57d5 | 29 | #include <linux/slab.h> |
66e1707b BS |
30 | #include <linux/swap.h> |
31 | #include <linux/spinlock.h> | |
32 | #include <linux/fs.h> | |
d2ceb9b7 | 33 | #include <linux/seq_file.h> |
33327948 | 34 | #include <linux/vmalloc.h> |
8cdea7c0 | 35 | |
8697d331 BS |
36 | #include <asm/uaccess.h> |
37 | ||
a181b0e8 KH |
38 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
39 | static struct kmem_cache *page_cgroup_cache __read_mostly; | |
40 | #define MEM_CGROUP_RECLAIM_RETRIES 5 | |
8cdea7c0 | 41 | |
d52aa412 KH |
42 | /* |
43 | * Statistics for memory cgroup. | |
44 | */ | |
45 | enum 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 */ | |
55e462b0 BR |
51 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
52 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
53 | |
54 | MEM_CGROUP_STAT_NSTATS, | |
55 | }; | |
56 | ||
57 | struct mem_cgroup_stat_cpu { | |
58 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
59 | } ____cacheline_aligned_in_smp; | |
60 | ||
61 | struct 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 | */ | |
68 | static 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 | ||
75 | static 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 | ||
6d12e2d8 KH |
85 | /* |
86 | * per-zone information in memory controller. | |
87 | */ | |
88 | ||
89 | enum mem_cgroup_zstat_index { | |
90 | MEM_CGROUP_ZSTAT_ACTIVE, | |
91 | MEM_CGROUP_ZSTAT_INACTIVE, | |
92 | ||
93 | NR_MEM_CGROUP_ZSTAT, | |
94 | }; | |
95 | ||
96 | struct mem_cgroup_per_zone { | |
072c56c1 KH |
97 | /* |
98 | * spin_lock to protect the per cgroup LRU | |
99 | */ | |
100 | spinlock_t lru_lock; | |
1ecaab2b KH |
101 | struct list_head active_list; |
102 | struct list_head inactive_list; | |
6d12e2d8 KH |
103 | unsigned long count[NR_MEM_CGROUP_ZSTAT]; |
104 | }; | |
105 | /* Macro for accessing counter */ | |
106 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
107 | ||
108 | struct mem_cgroup_per_node { | |
109 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
110 | }; | |
111 | ||
112 | struct mem_cgroup_lru_info { | |
113 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
114 | }; | |
115 | ||
8cdea7c0 BS |
116 | /* |
117 | * 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 | |
8a9f3ccd BS |
123 | * 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. | |
8cdea7c0 BS |
126 | */ |
127 | struct mem_cgroup { | |
128 | struct cgroup_subsys_state css; | |
129 | /* | |
130 | * the counter to account for memory usage | |
131 | */ | |
132 | struct res_counter res; | |
78fb7466 PE |
133 | /* |
134 | * Per cgroup active and inactive list, similar to the | |
135 | * per zone LRU lists. | |
78fb7466 | 136 | */ |
6d12e2d8 | 137 | struct mem_cgroup_lru_info info; |
072c56c1 | 138 | |
6c48a1d0 | 139 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
140 | /* |
141 | * statistics. | |
142 | */ | |
143 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 144 | }; |
8869b8f6 | 145 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 146 | |
8a9f3ccd BS |
147 | /* |
148 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
149 | * 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. | |
8a9f3ccd BS |
153 | */ |
154 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
155 | #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 | |
8a9f3ccd | 160 | |
8cdea7c0 BS |
161 | /* |
162 | * A page_cgroup page is associated with every page descriptor. The | |
163 | * page_cgroup helps us identify information about the cgroup | |
164 | */ | |
165 | struct page_cgroup { | |
166 | struct list_head lru; /* per cgroup LRU list */ | |
167 | struct page *page; | |
168 | struct mem_cgroup *mem_cgroup; | |
b9c565d5 | 169 | int ref_cnt; /* cached, mapped, migrating */ |
8869b8f6 | 170 | int flags; |
8cdea7c0 | 171 | }; |
217bc319 | 172 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 173 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
8cdea7c0 | 174 | |
d5b69e38 | 175 | static int page_cgroup_nid(struct page_cgroup *pc) |
c0149530 KH |
176 | { |
177 | return page_to_nid(pc->page); | |
178 | } | |
179 | ||
d5b69e38 | 180 | static enum zone_type page_cgroup_zid(struct page_cgroup *pc) |
c0149530 KH |
181 | { |
182 | return page_zonenum(pc->page); | |
183 | } | |
184 | ||
217bc319 KH |
185 | enum charge_type { |
186 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
187 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
188 | }; | |
189 | ||
d52aa412 KH |
190 | /* |
191 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
192 | */ | |
193 | static 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; | |
d52aa412 | 198 | |
8869b8f6 | 199 | VM_BUG_ON(!irqs_disabled()); |
d52aa412 | 200 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
8869b8f6 | 201 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 KH |
202 | else |
203 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
55e462b0 BR |
204 | |
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); | |
6d12e2d8 KH |
211 | } |
212 | ||
d5b69e38 | 213 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
214 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
215 | { | |
6d12e2d8 KH |
216 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
217 | } | |
218 | ||
d5b69e38 | 219 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
220 | page_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); | |
d52aa412 | 225 | |
6d12e2d8 KH |
226 | return mem_cgroup_zoneinfo(mem, nid, zid); |
227 | } | |
228 | ||
229 | static 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; | |
d52aa412 KH |
242 | } |
243 | ||
d5b69e38 | 244 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
245 | { |
246 | return container_of(cgroup_subsys_state(cont, | |
247 | mem_cgroup_subsys_id), struct mem_cgroup, | |
248 | css); | |
249 | } | |
250 | ||
cf475ad2 | 251 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 PE |
252 | { |
253 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
254 | struct mem_cgroup, css); | |
255 | } | |
256 | ||
8a9f3ccd BS |
257 | static inline int page_cgroup_locked(struct page *page) |
258 | { | |
8869b8f6 | 259 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
260 | } |
261 | ||
9442ec9d | 262 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 263 | { |
9442ec9d HD |
264 | VM_BUG_ON(!page_cgroup_locked(page)); |
265 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
266 | } |
267 | ||
268 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
269 | { | |
8869b8f6 | 270 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
271 | } |
272 | ||
d5b69e38 | 273 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
274 | { |
275 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
276 | } |
277 | ||
2680eed7 HD |
278 | static int try_lock_page_cgroup(struct page *page) |
279 | { | |
280 | return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
281 | } | |
282 | ||
d5b69e38 | 283 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
284 | { |
285 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
286 | } | |
287 | ||
3eae90c3 KH |
288 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
289 | struct page_cgroup *pc) | |
6d12e2d8 KH |
290 | { |
291 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
6d12e2d8 KH |
292 | |
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); | |
508b7be0 | 299 | list_del(&pc->lru); |
6d12e2d8 KH |
300 | } |
301 | ||
3eae90c3 KH |
302 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
303 | struct page_cgroup *pc) | |
6d12e2d8 KH |
304 | { |
305 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
6d12e2d8 KH |
306 | |
307 | if (!to) { | |
308 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 309 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
310 | } else { |
311 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 312 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
313 | } |
314 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
315 | } | |
316 | ||
8697d331 | 317 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 318 | { |
6d12e2d8 KH |
319 | 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 | ||
3564c7c4 | 327 | if (active) { |
6d12e2d8 | 328 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 329 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 330 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 331 | } else { |
6d12e2d8 | 332 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 333 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 334 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 335 | } |
66e1707b BS |
336 | } |
337 | ||
4c4a2214 DR |
338 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
339 | { | |
340 | int ret; | |
341 | ||
342 | task_lock(task); | |
bd845e38 | 343 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
344 | task_unlock(task); |
345 | return ret; | |
346 | } | |
347 | ||
66e1707b BS |
348 | /* |
349 | * This routine assumes that the appropriate zone's lru lock is already held | |
350 | */ | |
427d5416 | 351 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 352 | { |
427d5416 | 353 | struct page_cgroup *pc; |
072c56c1 KH |
354 | struct mem_cgroup_per_zone *mz; |
355 | unsigned long flags; | |
356 | ||
2680eed7 HD |
357 | /* |
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)) | |
66e1707b BS |
365 | return; |
366 | ||
2680eed7 HD |
367 | pc = page_get_page_cgroup(page); |
368 | if (pc) { | |
2680eed7 | 369 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 370 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 371 | __mem_cgroup_move_lists(pc, active); |
2680eed7 | 372 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 HT |
373 | } |
374 | unlock_page_cgroup(page); | |
66e1707b BS |
375 | } |
376 | ||
58ae83db KH |
377 | /* |
378 | * Calculate mapped_ratio under memory controller. This will be used in | |
379 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
380 | */ | |
381 | int 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 | } | |
8869b8f6 | 393 | |
5932f367 KH |
394 | /* |
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 | */ | |
400 | long 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 | } | |
58ae83db | 408 | |
6c48a1d0 KH |
409 | /* |
410 | * prev_priority control...this will be used in memory reclaim path. | |
411 | */ | |
412 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
413 | { | |
414 | return mem->prev_priority; | |
415 | } | |
416 | ||
417 | void 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 | ||
423 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
424 | { | |
425 | mem->prev_priority = priority; | |
426 | } | |
427 | ||
cc38108e KH |
428 | /* |
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 | ||
436 | long 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 | ||
448 | long 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); | |
cc38108e KH |
457 | return (nr_inactive >> priority); |
458 | } | |
459 | ||
66e1707b BS |
460 | unsigned 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; | |
ff7283fa | 472 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
473 | int nid = z->zone_pgdat->node_id; |
474 | int zid = zone_idx(z); | |
475 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 476 | |
cf475ad2 | 477 | BUG_ON(!mem_cont); |
1ecaab2b | 478 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 479 | if (active) |
1ecaab2b | 480 | src = &mz->active_list; |
66e1707b | 481 | else |
1ecaab2b KH |
482 | src = &mz->inactive_list; |
483 | ||
66e1707b | 484 | |
072c56c1 | 485 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
486 | scan = 0; |
487 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 488 | if (scan >= nr_to_scan) |
ff7283fa | 489 | break; |
66e1707b | 490 | page = pc->page; |
66e1707b | 491 | |
436c6541 | 492 | if (unlikely(!PageLRU(page))) |
ff7283fa | 493 | continue; |
ff7283fa | 494 | |
66e1707b BS |
495 | if (PageActive(page) && !active) { |
496 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
497 | continue; |
498 | } | |
499 | if (!PageActive(page) && active) { | |
500 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
501 | continue; |
502 | } | |
503 | ||
436c6541 HD |
504 | scan++; |
505 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
506 | |
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); | |
072c56c1 | 514 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
515 | |
516 | *scanned = scan; | |
517 | return nr_taken; | |
518 | } | |
519 | ||
8a9f3ccd BS |
520 | /* |
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 | */ | |
217bc319 | 526 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
e8589cc1 KH |
527 | gfp_t gfp_mask, enum charge_type ctype, |
528 | struct mem_cgroup *memcg) | |
8a9f3ccd BS |
529 | { |
530 | struct mem_cgroup *mem; | |
9175e031 | 531 | struct page_cgroup *pc; |
66e1707b BS |
532 | unsigned long flags; |
533 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 534 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 535 | |
4077960e BS |
536 | if (mem_cgroup_subsys.disabled) |
537 | return 0; | |
538 | ||
8a9f3ccd BS |
539 | /* |
540 | * Should page_cgroup's go to their own slab? | |
541 | * One could optimize the performance of the charging routine | |
542 | * by saving a bit in the page_flags and using it as a lock | |
543 | * to see if the cgroup page already has a page_cgroup associated | |
544 | * with it | |
545 | */ | |
66e1707b | 546 | retry: |
7e924aaf HD |
547 | lock_page_cgroup(page); |
548 | pc = page_get_page_cgroup(page); | |
549 | /* | |
550 | * The page_cgroup exists and | |
551 | * the page has already been accounted. | |
552 | */ | |
553 | if (pc) { | |
b9c565d5 HD |
554 | VM_BUG_ON(pc->page != page); |
555 | VM_BUG_ON(pc->ref_cnt <= 0); | |
556 | ||
557 | pc->ref_cnt++; | |
558 | unlock_page_cgroup(page); | |
559 | goto done; | |
8a9f3ccd | 560 | } |
7e924aaf | 561 | unlock_page_cgroup(page); |
8a9f3ccd | 562 | |
508b7be0 | 563 | pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask); |
8a9f3ccd BS |
564 | if (pc == NULL) |
565 | goto err; | |
566 | ||
8a9f3ccd | 567 | /* |
3be91277 HD |
568 | * We always charge the cgroup the mm_struct belongs to. |
569 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
570 | * thread group leader migrates. It's possible that mm is not |
571 | * set, if so charge the init_mm (happens for pagecache usage). | |
572 | */ | |
e8589cc1 KH |
573 | if (!memcg) { |
574 | if (!mm) | |
575 | mm = &init_mm; | |
8a9f3ccd | 576 | |
e8589cc1 KH |
577 | rcu_read_lock(); |
578 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
579 | /* | |
580 | * For every charge from the cgroup, increment reference count | |
581 | */ | |
582 | css_get(&mem->css); | |
583 | rcu_read_unlock(); | |
584 | } else { | |
585 | mem = memcg; | |
586 | css_get(&memcg->css); | |
587 | } | |
8a9f3ccd | 588 | |
0eea1030 | 589 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
590 | if (!(gfp_mask & __GFP_WAIT)) |
591 | goto out; | |
e1a1cd59 BS |
592 | |
593 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
594 | continue; |
595 | ||
596 | /* | |
8869b8f6 HD |
597 | * try_to_free_mem_cgroup_pages() might not give us a full |
598 | * picture of reclaim. Some pages are reclaimed and might be | |
599 | * moved to swap cache or just unmapped from the cgroup. | |
600 | * Check the limit again to see if the reclaim reduced the | |
601 | * current usage of the cgroup before giving up | |
602 | */ | |
66e1707b BS |
603 | if (res_counter_check_under_limit(&mem->res)) |
604 | continue; | |
3be91277 HD |
605 | |
606 | if (!nr_retries--) { | |
607 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
608 | goto out; | |
66e1707b | 609 | } |
8a9f3ccd BS |
610 | } |
611 | ||
b9c565d5 | 612 | pc->ref_cnt = 1; |
8a9f3ccd BS |
613 | pc->mem_cgroup = mem; |
614 | pc->page = page; | |
508b7be0 KH |
615 | /* |
616 | * If a page is accounted as a page cache, insert to inactive list. | |
617 | * If anon, insert to active list. | |
618 | */ | |
217bc319 | 619 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
4a56d02e | 620 | pc->flags = PAGE_CGROUP_FLAG_CACHE; |
508b7be0 KH |
621 | else |
622 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; | |
3be91277 | 623 | |
7e924aaf HD |
624 | lock_page_cgroup(page); |
625 | if (page_get_page_cgroup(page)) { | |
626 | unlock_page_cgroup(page); | |
9175e031 | 627 | /* |
3be91277 HD |
628 | * Another charge has been added to this page already. |
629 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
630 | * page->cgroup, increment refcnt.... just retry is OK. |
631 | */ | |
632 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
633 | css_put(&mem->css); | |
b6ac57d5 | 634 | kmem_cache_free(page_cgroup_cache, pc); |
9175e031 KH |
635 | goto retry; |
636 | } | |
7e924aaf | 637 | page_assign_page_cgroup(page, pc); |
8a9f3ccd | 638 | |
072c56c1 KH |
639 | mz = page_cgroup_zoneinfo(pc); |
640 | spin_lock_irqsave(&mz->lru_lock, flags); | |
3eae90c3 | 641 | __mem_cgroup_add_list(mz, pc); |
072c56c1 | 642 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 643 | |
fb59e9f1 | 644 | unlock_page_cgroup(page); |
8a9f3ccd | 645 | done: |
8a9f3ccd | 646 | return 0; |
3be91277 HD |
647 | out: |
648 | css_put(&mem->css); | |
b6ac57d5 | 649 | kmem_cache_free(page_cgroup_cache, pc); |
8a9f3ccd | 650 | err: |
8a9f3ccd BS |
651 | return -ENOMEM; |
652 | } | |
653 | ||
8869b8f6 | 654 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 KH |
655 | { |
656 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 657 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
658 | } |
659 | ||
e1a1cd59 BS |
660 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
661 | gfp_t gfp_mask) | |
8697d331 | 662 | { |
8697d331 BS |
663 | if (!mm) |
664 | mm = &init_mm; | |
8869b8f6 | 665 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 KH |
666 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
667 | } | |
668 | ||
669 | int mem_cgroup_getref(struct page *page) | |
670 | { | |
671 | struct page_cgroup *pc; | |
672 | ||
673 | if (mem_cgroup_subsys.disabled) | |
674 | return 0; | |
675 | ||
676 | lock_page_cgroup(page); | |
677 | pc = page_get_page_cgroup(page); | |
678 | VM_BUG_ON(!pc); | |
679 | pc->ref_cnt++; | |
680 | unlock_page_cgroup(page); | |
681 | return 0; | |
8697d331 BS |
682 | } |
683 | ||
8a9f3ccd BS |
684 | /* |
685 | * Uncharging is always a welcome operation, we never complain, simply | |
8289546e | 686 | * uncharge. |
8a9f3ccd | 687 | */ |
8289546e | 688 | void mem_cgroup_uncharge_page(struct page *page) |
8a9f3ccd | 689 | { |
8289546e | 690 | struct page_cgroup *pc; |
8a9f3ccd | 691 | struct mem_cgroup *mem; |
072c56c1 | 692 | struct mem_cgroup_per_zone *mz; |
66e1707b | 693 | unsigned long flags; |
8a9f3ccd | 694 | |
4077960e BS |
695 | if (mem_cgroup_subsys.disabled) |
696 | return; | |
697 | ||
8697d331 | 698 | /* |
3c541e14 | 699 | * Check if our page_cgroup is valid |
8697d331 | 700 | */ |
8289546e HD |
701 | lock_page_cgroup(page); |
702 | pc = page_get_page_cgroup(page); | |
8a9f3ccd | 703 | if (!pc) |
8289546e | 704 | goto unlock; |
8a9f3ccd | 705 | |
b9c565d5 HD |
706 | VM_BUG_ON(pc->page != page); |
707 | VM_BUG_ON(pc->ref_cnt <= 0); | |
708 | ||
709 | if (--(pc->ref_cnt) == 0) { | |
b9c565d5 HD |
710 | mz = page_cgroup_zoneinfo(pc); |
711 | spin_lock_irqsave(&mz->lru_lock, flags); | |
3eae90c3 | 712 | __mem_cgroup_remove_list(mz, pc); |
b9c565d5 HD |
713 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
714 | ||
fb59e9f1 HD |
715 | page_assign_page_cgroup(page, NULL); |
716 | unlock_page_cgroup(page); | |
717 | ||
6d48ff8b HD |
718 | mem = pc->mem_cgroup; |
719 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
720 | css_put(&mem->css); | |
721 | ||
b6ac57d5 | 722 | kmem_cache_free(page_cgroup_cache, pc); |
b9c565d5 | 723 | return; |
8a9f3ccd | 724 | } |
6d12e2d8 | 725 | |
8289546e | 726 | unlock: |
3c541e14 BS |
727 | unlock_page_cgroup(page); |
728 | } | |
729 | ||
ae41be37 | 730 | /* |
e8589cc1 | 731 | * Before starting migration, account against new page. |
ae41be37 | 732 | */ |
e8589cc1 | 733 | int mem_cgroup_prepare_migration(struct page *page, struct page *newpage) |
ae41be37 KH |
734 | { |
735 | struct page_cgroup *pc; | |
e8589cc1 KH |
736 | struct mem_cgroup *mem = NULL; |
737 | enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
738 | int ret = 0; | |
8869b8f6 | 739 | |
4077960e BS |
740 | if (mem_cgroup_subsys.disabled) |
741 | return 0; | |
742 | ||
ae41be37 KH |
743 | lock_page_cgroup(page); |
744 | pc = page_get_page_cgroup(page); | |
e8589cc1 KH |
745 | if (pc) { |
746 | mem = pc->mem_cgroup; | |
747 | css_get(&mem->css); | |
748 | if (pc->flags & PAGE_CGROUP_FLAG_CACHE) | |
749 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
750 | } | |
ae41be37 | 751 | unlock_page_cgroup(page); |
e8589cc1 KH |
752 | if (mem) { |
753 | ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL, | |
754 | ctype, mem); | |
755 | css_put(&mem->css); | |
756 | } | |
757 | return ret; | |
ae41be37 | 758 | } |
8869b8f6 | 759 | |
e8589cc1 KH |
760 | /* remove redundant charge */ |
761 | void mem_cgroup_end_migration(struct page *newpage) | |
ae41be37 | 762 | { |
e8589cc1 | 763 | mem_cgroup_uncharge_page(newpage); |
ae41be37 | 764 | } |
78fb7466 | 765 | |
cc847582 KH |
766 | /* |
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) | |
8869b8f6 | 772 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 KH |
773 | struct mem_cgroup_per_zone *mz, |
774 | int active) | |
cc847582 KH |
775 | { |
776 | struct page_cgroup *pc; | |
777 | struct page *page; | |
9b3c0a07 | 778 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 779 | unsigned long flags; |
072c56c1 KH |
780 | struct list_head *list; |
781 | ||
782 | if (active) | |
783 | list = &mz->active_list; | |
784 | else | |
785 | list = &mz->inactive_list; | |
cc847582 | 786 | |
072c56c1 | 787 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 788 | while (!list_empty(list)) { |
cc847582 KH |
789 | pc = list_entry(list->prev, struct page_cgroup, lru); |
790 | page = pc->page; | |
9b3c0a07 HT |
791 | get_page(page); |
792 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
e8589cc1 KH |
793 | /* |
794 | * Check if this page is on LRU. !LRU page can be found | |
795 | * if it's under page migration. | |
796 | */ | |
797 | if (PageLRU(page)) { | |
798 | mem_cgroup_uncharge_page(page); | |
799 | put_page(page); | |
800 | if (--count <= 0) { | |
801 | count = FORCE_UNCHARGE_BATCH; | |
802 | cond_resched(); | |
803 | } | |
804 | } else | |
9b3c0a07 | 805 | cond_resched(); |
9b3c0a07 | 806 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 807 | } |
072c56c1 | 808 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
809 | } |
810 | ||
811 | /* | |
812 | * make mem_cgroup's charge to be 0 if there is no task. | |
813 | * This enables deleting this mem_cgroup. | |
814 | */ | |
d5b69e38 | 815 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
816 | { |
817 | int ret = -EBUSY; | |
1ecaab2b | 818 | int node, zid; |
8869b8f6 | 819 | |
4077960e BS |
820 | if (mem_cgroup_subsys.disabled) |
821 | return 0; | |
822 | ||
cc847582 KH |
823 | css_get(&mem->css); |
824 | /* | |
825 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 826 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
827 | * So, we have to do loop here until all lists are empty. |
828 | */ | |
1ecaab2b | 829 | while (mem->res.usage > 0) { |
cc847582 KH |
830 | if (atomic_read(&mem->css.cgroup->count) > 0) |
831 | goto out; | |
1ecaab2b KH |
832 | for_each_node_state(node, N_POSSIBLE) |
833 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
834 | struct mem_cgroup_per_zone *mz; | |
835 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
836 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 837 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 838 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 839 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 840 | } |
cc847582 KH |
841 | } |
842 | ret = 0; | |
843 | out: | |
844 | css_put(&mem->css); | |
845 | return ret; | |
846 | } | |
847 | ||
2c3daa72 | 848 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 849 | { |
2c3daa72 PM |
850 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
851 | cft->private); | |
8cdea7c0 BS |
852 | } |
853 | ||
856c13aa PM |
854 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
855 | const char *buffer) | |
8cdea7c0 BS |
856 | { |
857 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
856c13aa PM |
858 | cft->private, buffer, |
859 | res_counter_memparse_write_strategy); | |
8cdea7c0 BS |
860 | } |
861 | ||
29f2a4da | 862 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
863 | { |
864 | struct mem_cgroup *mem; | |
865 | ||
866 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
867 | switch (event) { |
868 | case RES_MAX_USAGE: | |
869 | res_counter_reset_max(&mem->res); | |
870 | break; | |
871 | case RES_FAILCNT: | |
872 | res_counter_reset_failcnt(&mem->res); | |
873 | break; | |
874 | } | |
85cc59db | 875 | return 0; |
c84872e1 PE |
876 | } |
877 | ||
85cc59db | 878 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
cc847582 | 879 | { |
85cc59db | 880 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
cc847582 KH |
881 | } |
882 | ||
d2ceb9b7 KH |
883 | static const struct mem_cgroup_stat_desc { |
884 | const char *msg; | |
885 | u64 unit; | |
886 | } mem_cgroup_stat_desc[] = { | |
887 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
888 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
889 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
890 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
891 | }; |
892 | ||
c64745cf PM |
893 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
894 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 895 | { |
d2ceb9b7 KH |
896 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
897 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
898 | int i; | |
899 | ||
900 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
901 | s64 val; | |
902 | ||
903 | val = mem_cgroup_read_stat(stat, i); | |
904 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 905 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 906 | } |
6d12e2d8 KH |
907 | /* showing # of active pages */ |
908 | { | |
909 | unsigned long active, inactive; | |
910 | ||
911 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
912 | MEM_CGROUP_ZSTAT_INACTIVE); | |
913 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
914 | MEM_CGROUP_ZSTAT_ACTIVE); | |
c64745cf PM |
915 | cb->fill(cb, "active", (active) * PAGE_SIZE); |
916 | cb->fill(cb, "inactive", (inactive) * PAGE_SIZE); | |
6d12e2d8 | 917 | } |
d2ceb9b7 KH |
918 | return 0; |
919 | } | |
920 | ||
8cdea7c0 BS |
921 | static struct cftype mem_cgroup_files[] = { |
922 | { | |
0eea1030 | 923 | .name = "usage_in_bytes", |
8cdea7c0 | 924 | .private = RES_USAGE, |
2c3daa72 | 925 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 926 | }, |
c84872e1 PE |
927 | { |
928 | .name = "max_usage_in_bytes", | |
929 | .private = RES_MAX_USAGE, | |
29f2a4da | 930 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
931 | .read_u64 = mem_cgroup_read, |
932 | }, | |
8cdea7c0 | 933 | { |
0eea1030 | 934 | .name = "limit_in_bytes", |
8cdea7c0 | 935 | .private = RES_LIMIT, |
856c13aa | 936 | .write_string = mem_cgroup_write, |
2c3daa72 | 937 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
938 | }, |
939 | { | |
940 | .name = "failcnt", | |
941 | .private = RES_FAILCNT, | |
29f2a4da | 942 | .trigger = mem_cgroup_reset, |
2c3daa72 | 943 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 944 | }, |
cc847582 KH |
945 | { |
946 | .name = "force_empty", | |
85cc59db | 947 | .trigger = mem_force_empty_write, |
cc847582 | 948 | }, |
d2ceb9b7 KH |
949 | { |
950 | .name = "stat", | |
c64745cf | 951 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 952 | }, |
8cdea7c0 BS |
953 | }; |
954 | ||
6d12e2d8 KH |
955 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
956 | { | |
957 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 958 | struct mem_cgroup_per_zone *mz; |
41e3355d | 959 | int zone, tmp = node; |
1ecaab2b KH |
960 | /* |
961 | * This routine is called against possible nodes. | |
962 | * But it's BUG to call kmalloc() against offline node. | |
963 | * | |
964 | * TODO: this routine can waste much memory for nodes which will | |
965 | * never be onlined. It's better to use memory hotplug callback | |
966 | * function. | |
967 | */ | |
41e3355d KH |
968 | if (!node_state(node, N_NORMAL_MEMORY)) |
969 | tmp = -1; | |
970 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
971 | if (!pn) |
972 | return 1; | |
1ecaab2b | 973 | |
6d12e2d8 KH |
974 | mem->info.nodeinfo[node] = pn; |
975 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
976 | |
977 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
978 | mz = &pn->zoneinfo[zone]; | |
979 | INIT_LIST_HEAD(&mz->active_list); | |
980 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 981 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 982 | } |
6d12e2d8 KH |
983 | return 0; |
984 | } | |
985 | ||
1ecaab2b KH |
986 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
987 | { | |
988 | kfree(mem->info.nodeinfo[node]); | |
989 | } | |
990 | ||
33327948 KH |
991 | static struct mem_cgroup *mem_cgroup_alloc(void) |
992 | { | |
993 | struct mem_cgroup *mem; | |
994 | ||
995 | if (sizeof(*mem) < PAGE_SIZE) | |
996 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); | |
997 | else | |
998 | mem = vmalloc(sizeof(*mem)); | |
999 | ||
1000 | if (mem) | |
1001 | memset(mem, 0, sizeof(*mem)); | |
1002 | return mem; | |
1003 | } | |
1004 | ||
1005 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1006 | { | |
1007 | if (sizeof(*mem) < PAGE_SIZE) | |
1008 | kfree(mem); | |
1009 | else | |
1010 | vfree(mem); | |
1011 | } | |
1012 | ||
1013 | ||
8cdea7c0 BS |
1014 | static struct cgroup_subsys_state * |
1015 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1016 | { | |
1017 | struct mem_cgroup *mem; | |
6d12e2d8 | 1018 | int node; |
8cdea7c0 | 1019 | |
b6ac57d5 | 1020 | if (unlikely((cont->parent) == NULL)) { |
78fb7466 | 1021 | mem = &init_mem_cgroup; |
b6ac57d5 BS |
1022 | page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC); |
1023 | } else { | |
33327948 KH |
1024 | mem = mem_cgroup_alloc(); |
1025 | if (!mem) | |
1026 | return ERR_PTR(-ENOMEM); | |
b6ac57d5 | 1027 | } |
78fb7466 | 1028 | |
8cdea7c0 | 1029 | res_counter_init(&mem->res); |
1ecaab2b | 1030 | |
6d12e2d8 KH |
1031 | for_each_node_state(node, N_POSSIBLE) |
1032 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1033 | goto free_out; | |
1034 | ||
8cdea7c0 | 1035 | return &mem->css; |
6d12e2d8 KH |
1036 | free_out: |
1037 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1038 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1039 | if (cont->parent != NULL) |
33327948 | 1040 | mem_cgroup_free(mem); |
2dda81ca | 1041 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1042 | } |
1043 | ||
df878fb0 KH |
1044 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1045 | struct cgroup *cont) | |
1046 | { | |
1047 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1048 | mem_cgroup_force_empty(mem); | |
1049 | } | |
1050 | ||
8cdea7c0 BS |
1051 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1052 | struct cgroup *cont) | |
1053 | { | |
6d12e2d8 KH |
1054 | int node; |
1055 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1056 | ||
1057 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1058 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1059 | |
33327948 | 1060 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1061 | } |
1062 | ||
1063 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1064 | struct cgroup *cont) | |
1065 | { | |
4077960e BS |
1066 | if (mem_cgroup_subsys.disabled) |
1067 | return 0; | |
8cdea7c0 BS |
1068 | return cgroup_add_files(cont, ss, mem_cgroup_files, |
1069 | ARRAY_SIZE(mem_cgroup_files)); | |
1070 | } | |
1071 | ||
67e465a7 BS |
1072 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1073 | struct cgroup *cont, | |
1074 | struct cgroup *old_cont, | |
1075 | struct task_struct *p) | |
1076 | { | |
1077 | struct mm_struct *mm; | |
1078 | struct mem_cgroup *mem, *old_mem; | |
1079 | ||
4077960e BS |
1080 | if (mem_cgroup_subsys.disabled) |
1081 | return; | |
1082 | ||
67e465a7 BS |
1083 | mm = get_task_mm(p); |
1084 | if (mm == NULL) | |
1085 | return; | |
1086 | ||
1087 | mem = mem_cgroup_from_cont(cont); | |
1088 | old_mem = mem_cgroup_from_cont(old_cont); | |
1089 | ||
1090 | if (mem == old_mem) | |
1091 | goto out; | |
1092 | ||
1093 | /* | |
1094 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1095 | * in effect owned by the leader | |
1096 | */ | |
52ea27eb | 1097 | if (!thread_group_leader(p)) |
67e465a7 BS |
1098 | goto out; |
1099 | ||
67e465a7 BS |
1100 | out: |
1101 | mmput(mm); | |
67e465a7 BS |
1102 | } |
1103 | ||
8cdea7c0 BS |
1104 | struct cgroup_subsys mem_cgroup_subsys = { |
1105 | .name = "memory", | |
1106 | .subsys_id = mem_cgroup_subsys_id, | |
1107 | .create = mem_cgroup_create, | |
df878fb0 | 1108 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1109 | .destroy = mem_cgroup_destroy, |
1110 | .populate = mem_cgroup_populate, | |
67e465a7 | 1111 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1112 | .early_init = 0, |
8cdea7c0 | 1113 | }; |