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> | |
66e1707b BS |
29 | #include <linux/swap.h> |
30 | #include <linux/spinlock.h> | |
31 | #include <linux/fs.h> | |
d2ceb9b7 | 32 | #include <linux/seq_file.h> |
8cdea7c0 | 33 | |
8697d331 BS |
34 | #include <asm/uaccess.h> |
35 | ||
8cdea7c0 | 36 | struct cgroup_subsys mem_cgroup_subsys; |
66e1707b | 37 | static const int MEM_CGROUP_RECLAIM_RETRIES = 5; |
8cdea7c0 | 38 | |
d52aa412 KH |
39 | /* |
40 | * Statistics for memory cgroup. | |
41 | */ | |
42 | enum mem_cgroup_stat_index { | |
43 | /* | |
44 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
45 | */ | |
46 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
47 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
48 | ||
49 | MEM_CGROUP_STAT_NSTATS, | |
50 | }; | |
51 | ||
52 | struct mem_cgroup_stat_cpu { | |
53 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
54 | } ____cacheline_aligned_in_smp; | |
55 | ||
56 | struct mem_cgroup_stat { | |
57 | struct mem_cgroup_stat_cpu cpustat[NR_CPUS]; | |
58 | }; | |
59 | ||
60 | /* | |
61 | * For accounting under irq disable, no need for increment preempt count. | |
62 | */ | |
63 | static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat, | |
64 | enum mem_cgroup_stat_index idx, int val) | |
65 | { | |
66 | int cpu = smp_processor_id(); | |
67 | stat->cpustat[cpu].count[idx] += val; | |
68 | } | |
69 | ||
70 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
71 | enum mem_cgroup_stat_index idx) | |
72 | { | |
73 | int cpu; | |
74 | s64 ret = 0; | |
75 | for_each_possible_cpu(cpu) | |
76 | ret += stat->cpustat[cpu].count[idx]; | |
77 | return ret; | |
78 | } | |
79 | ||
6d12e2d8 KH |
80 | /* |
81 | * per-zone information in memory controller. | |
82 | */ | |
83 | ||
84 | enum mem_cgroup_zstat_index { | |
85 | MEM_CGROUP_ZSTAT_ACTIVE, | |
86 | MEM_CGROUP_ZSTAT_INACTIVE, | |
87 | ||
88 | NR_MEM_CGROUP_ZSTAT, | |
89 | }; | |
90 | ||
91 | struct mem_cgroup_per_zone { | |
072c56c1 KH |
92 | /* |
93 | * spin_lock to protect the per cgroup LRU | |
94 | */ | |
95 | spinlock_t lru_lock; | |
1ecaab2b KH |
96 | struct list_head active_list; |
97 | struct list_head inactive_list; | |
6d12e2d8 KH |
98 | unsigned long count[NR_MEM_CGROUP_ZSTAT]; |
99 | }; | |
100 | /* Macro for accessing counter */ | |
101 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
102 | ||
103 | struct mem_cgroup_per_node { | |
104 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
105 | }; | |
106 | ||
107 | struct mem_cgroup_lru_info { | |
108 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
109 | }; | |
110 | ||
8cdea7c0 BS |
111 | /* |
112 | * The memory controller data structure. The memory controller controls both | |
113 | * page cache and RSS per cgroup. We would eventually like to provide | |
114 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
115 | * to help the administrator determine what knobs to tune. | |
116 | * | |
117 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
118 | * we hit the water mark. May be even add a low water mark, such that |
119 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
120 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
121 | */ |
122 | struct mem_cgroup { | |
123 | struct cgroup_subsys_state css; | |
124 | /* | |
125 | * the counter to account for memory usage | |
126 | */ | |
127 | struct res_counter res; | |
78fb7466 PE |
128 | /* |
129 | * Per cgroup active and inactive list, similar to the | |
130 | * per zone LRU lists. | |
78fb7466 | 131 | */ |
6d12e2d8 | 132 | struct mem_cgroup_lru_info info; |
072c56c1 | 133 | |
6c48a1d0 | 134 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
135 | /* |
136 | * statistics. | |
137 | */ | |
138 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 139 | }; |
8869b8f6 | 140 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 141 | |
8a9f3ccd BS |
142 | /* |
143 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
144 | * lock. We need to ensure that page->page_cgroup is at least two |
145 | * byte aligned (based on comments from Nick Piggin). But since | |
146 | * bit_spin_lock doesn't actually set that lock bit in a non-debug | |
147 | * uniprocessor kernel, we should avoid setting it here too. | |
8a9f3ccd BS |
148 | */ |
149 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
150 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) |
151 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
152 | #else | |
153 | #define PAGE_CGROUP_LOCK 0x0 | |
154 | #endif | |
8a9f3ccd | 155 | |
8cdea7c0 BS |
156 | /* |
157 | * A page_cgroup page is associated with every page descriptor. The | |
158 | * page_cgroup helps us identify information about the cgroup | |
159 | */ | |
160 | struct page_cgroup { | |
161 | struct list_head lru; /* per cgroup LRU list */ | |
162 | struct page *page; | |
163 | struct mem_cgroup *mem_cgroup; | |
b9c565d5 | 164 | int ref_cnt; /* cached, mapped, migrating */ |
8869b8f6 | 165 | int flags; |
8cdea7c0 | 166 | }; |
217bc319 | 167 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 168 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
8cdea7c0 | 169 | |
d5b69e38 | 170 | static int page_cgroup_nid(struct page_cgroup *pc) |
c0149530 KH |
171 | { |
172 | return page_to_nid(pc->page); | |
173 | } | |
174 | ||
d5b69e38 | 175 | static enum zone_type page_cgroup_zid(struct page_cgroup *pc) |
c0149530 KH |
176 | { |
177 | return page_zonenum(pc->page); | |
178 | } | |
179 | ||
217bc319 KH |
180 | enum charge_type { |
181 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
182 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
183 | }; | |
184 | ||
d52aa412 KH |
185 | /* |
186 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
187 | */ | |
188 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
189 | bool charge) | |
190 | { | |
191 | int val = (charge)? 1 : -1; | |
192 | struct mem_cgroup_stat *stat = &mem->stat; | |
d52aa412 | 193 | |
8869b8f6 | 194 | VM_BUG_ON(!irqs_disabled()); |
d52aa412 | 195 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
8869b8f6 | 196 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 KH |
197 | else |
198 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
6d12e2d8 KH |
199 | } |
200 | ||
d5b69e38 | 201 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
202 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
203 | { | |
6d12e2d8 KH |
204 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
205 | } | |
206 | ||
d5b69e38 | 207 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
208 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
209 | { | |
210 | struct mem_cgroup *mem = pc->mem_cgroup; | |
211 | int nid = page_cgroup_nid(pc); | |
212 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 213 | |
6d12e2d8 KH |
214 | return mem_cgroup_zoneinfo(mem, nid, zid); |
215 | } | |
216 | ||
217 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
218 | enum mem_cgroup_zstat_index idx) | |
219 | { | |
220 | int nid, zid; | |
221 | struct mem_cgroup_per_zone *mz; | |
222 | u64 total = 0; | |
223 | ||
224 | for_each_online_node(nid) | |
225 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
226 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
227 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
228 | } | |
229 | return total; | |
d52aa412 KH |
230 | } |
231 | ||
d5b69e38 | 232 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
233 | { |
234 | return container_of(cgroup_subsys_state(cont, | |
235 | mem_cgroup_subsys_id), struct mem_cgroup, | |
236 | css); | |
237 | } | |
238 | ||
d5b69e38 | 239 | static struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 PE |
240 | { |
241 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
242 | struct mem_cgroup, css); | |
243 | } | |
244 | ||
245 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
246 | { | |
247 | struct mem_cgroup *mem; | |
248 | ||
249 | mem = mem_cgroup_from_task(p); | |
250 | css_get(&mem->css); | |
251 | mm->mem_cgroup = mem; | |
252 | } | |
253 | ||
254 | void mm_free_cgroup(struct mm_struct *mm) | |
255 | { | |
256 | css_put(&mm->mem_cgroup->css); | |
257 | } | |
258 | ||
8a9f3ccd BS |
259 | static inline int page_cgroup_locked(struct page *page) |
260 | { | |
8869b8f6 | 261 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
262 | } |
263 | ||
9442ec9d | 264 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 265 | { |
9442ec9d HD |
266 | VM_BUG_ON(!page_cgroup_locked(page)); |
267 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
268 | } |
269 | ||
270 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
271 | { | |
8869b8f6 | 272 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
273 | } |
274 | ||
d5b69e38 | 275 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
276 | { |
277 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
278 | } |
279 | ||
2680eed7 HD |
280 | static int try_lock_page_cgroup(struct page *page) |
281 | { | |
282 | return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
283 | } | |
284 | ||
d5b69e38 | 285 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
286 | { |
287 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
288 | } | |
289 | ||
6d12e2d8 KH |
290 | static void __mem_cgroup_remove_list(struct page_cgroup *pc) |
291 | { | |
292 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
293 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
294 | ||
295 | if (from) | |
296 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
297 | else | |
298 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
299 | ||
300 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
301 | list_del_init(&pc->lru); | |
302 | } | |
303 | ||
304 | static void __mem_cgroup_add_list(struct page_cgroup *pc) | |
305 | { | |
306 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
307 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
308 | ||
309 | if (!to) { | |
310 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 311 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
312 | } else { |
313 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 314 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
315 | } |
316 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
317 | } | |
318 | ||
8697d331 | 319 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 320 | { |
6d12e2d8 KH |
321 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
322 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
323 | ||
324 | if (from) | |
325 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
326 | else | |
327 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
328 | ||
3564c7c4 | 329 | if (active) { |
6d12e2d8 | 330 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 331 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 332 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 333 | } else { |
6d12e2d8 | 334 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 335 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 336 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 337 | } |
66e1707b BS |
338 | } |
339 | ||
4c4a2214 DR |
340 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
341 | { | |
342 | int ret; | |
343 | ||
344 | task_lock(task); | |
bd845e38 | 345 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
346 | task_unlock(task); |
347 | return ret; | |
348 | } | |
349 | ||
66e1707b BS |
350 | /* |
351 | * This routine assumes that the appropriate zone's lru lock is already held | |
352 | */ | |
427d5416 | 353 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 354 | { |
427d5416 | 355 | struct page_cgroup *pc; |
072c56c1 KH |
356 | struct mem_cgroup_per_zone *mz; |
357 | unsigned long flags; | |
358 | ||
2680eed7 HD |
359 | /* |
360 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
361 | * because other holders of lock_page_cgroup can be interrupted | |
362 | * with an attempt to rotate_reclaimable_page. But we cannot | |
363 | * safely get to page_cgroup without it, so just try_lock it: | |
364 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
365 | */ | |
366 | if (!try_lock_page_cgroup(page)) | |
66e1707b BS |
367 | return; |
368 | ||
2680eed7 HD |
369 | pc = page_get_page_cgroup(page); |
370 | if (pc) { | |
2680eed7 | 371 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 372 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 373 | __mem_cgroup_move_lists(pc, active); |
2680eed7 | 374 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 HT |
375 | } |
376 | unlock_page_cgroup(page); | |
66e1707b BS |
377 | } |
378 | ||
58ae83db KH |
379 | /* |
380 | * Calculate mapped_ratio under memory controller. This will be used in | |
381 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
382 | */ | |
383 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
384 | { | |
385 | long total, rss; | |
386 | ||
387 | /* | |
388 | * usage is recorded in bytes. But, here, we assume the number of | |
389 | * physical pages can be represented by "long" on any arch. | |
390 | */ | |
391 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
392 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
393 | return (int)((rss * 100L) / total); | |
394 | } | |
8869b8f6 | 395 | |
5932f367 KH |
396 | /* |
397 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
398 | * between active and inactive list is calculated. For memory controller | |
399 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
400 | * zone's global lru imbalance. | |
401 | */ | |
402 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
403 | { | |
404 | unsigned long active, inactive; | |
405 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
406 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
407 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
408 | return (long) (active / (inactive + 1)); | |
409 | } | |
58ae83db | 410 | |
6c48a1d0 KH |
411 | /* |
412 | * prev_priority control...this will be used in memory reclaim path. | |
413 | */ | |
414 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
415 | { | |
416 | return mem->prev_priority; | |
417 | } | |
418 | ||
419 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
420 | { | |
421 | if (priority < mem->prev_priority) | |
422 | mem->prev_priority = priority; | |
423 | } | |
424 | ||
425 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
426 | { | |
427 | mem->prev_priority = priority; | |
428 | } | |
429 | ||
cc38108e KH |
430 | /* |
431 | * Calculate # of pages to be scanned in this priority/zone. | |
432 | * See also vmscan.c | |
433 | * | |
434 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
435 | * (see include/linux/mmzone.h) | |
436 | */ | |
437 | ||
438 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
439 | struct zone *zone, int priority) | |
440 | { | |
441 | long nr_active; | |
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_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
447 | return (nr_active >> priority); | |
448 | } | |
449 | ||
450 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
451 | struct zone *zone, int priority) | |
452 | { | |
453 | long nr_inactive; | |
454 | int nid = zone->zone_pgdat->node_id; | |
455 | int zid = zone_idx(zone); | |
456 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
457 | ||
458 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
cc38108e KH |
459 | return (nr_inactive >> priority); |
460 | } | |
461 | ||
66e1707b BS |
462 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
463 | struct list_head *dst, | |
464 | unsigned long *scanned, int order, | |
465 | int mode, struct zone *z, | |
466 | struct mem_cgroup *mem_cont, | |
467 | int active) | |
468 | { | |
469 | unsigned long nr_taken = 0; | |
470 | struct page *page; | |
471 | unsigned long scan; | |
472 | LIST_HEAD(pc_list); | |
473 | struct list_head *src; | |
ff7283fa | 474 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
475 | int nid = z->zone_pgdat->node_id; |
476 | int zid = zone_idx(z); | |
477 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 478 | |
1ecaab2b | 479 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 480 | if (active) |
1ecaab2b | 481 | src = &mz->active_list; |
66e1707b | 482 | else |
1ecaab2b KH |
483 | src = &mz->inactive_list; |
484 | ||
66e1707b | 485 | |
072c56c1 | 486 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
487 | scan = 0; |
488 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 489 | if (scan >= nr_to_scan) |
ff7283fa | 490 | break; |
66e1707b | 491 | page = pc->page; |
66e1707b | 492 | |
436c6541 | 493 | if (unlikely(!PageLRU(page))) |
ff7283fa | 494 | continue; |
ff7283fa | 495 | |
66e1707b BS |
496 | if (PageActive(page) && !active) { |
497 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
498 | continue; |
499 | } | |
500 | if (!PageActive(page) && active) { | |
501 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
502 | continue; |
503 | } | |
504 | ||
436c6541 HD |
505 | scan++; |
506 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
507 | |
508 | if (__isolate_lru_page(page, mode) == 0) { | |
509 | list_move(&page->lru, dst); | |
510 | nr_taken++; | |
511 | } | |
512 | } | |
513 | ||
514 | list_splice(&pc_list, src); | |
072c56c1 | 515 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
516 | |
517 | *scanned = scan; | |
518 | return nr_taken; | |
519 | } | |
520 | ||
8a9f3ccd BS |
521 | /* |
522 | * Charge the memory controller for page usage. | |
523 | * Return | |
524 | * 0 if the charge was successful | |
525 | * < 0 if the cgroup is over its limit | |
526 | */ | |
217bc319 KH |
527 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
528 | gfp_t gfp_mask, enum charge_type ctype) | |
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 BS |
535 | |
536 | /* | |
537 | * Should page_cgroup's go to their own slab? | |
538 | * One could optimize the performance of the charging routine | |
539 | * by saving a bit in the page_flags and using it as a lock | |
540 | * to see if the cgroup page already has a page_cgroup associated | |
541 | * with it | |
542 | */ | |
66e1707b | 543 | retry: |
7e924aaf HD |
544 | lock_page_cgroup(page); |
545 | pc = page_get_page_cgroup(page); | |
546 | /* | |
547 | * The page_cgroup exists and | |
548 | * the page has already been accounted. | |
549 | */ | |
550 | if (pc) { | |
b9c565d5 HD |
551 | VM_BUG_ON(pc->page != page); |
552 | VM_BUG_ON(pc->ref_cnt <= 0); | |
553 | ||
554 | pc->ref_cnt++; | |
555 | unlock_page_cgroup(page); | |
556 | goto done; | |
8a9f3ccd | 557 | } |
7e924aaf | 558 | unlock_page_cgroup(page); |
8a9f3ccd | 559 | |
e1a1cd59 | 560 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
561 | if (pc == NULL) |
562 | goto err; | |
563 | ||
8a9f3ccd | 564 | /* |
3be91277 HD |
565 | * We always charge the cgroup the mm_struct belongs to. |
566 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
567 | * thread group leader migrates. It's possible that mm is not |
568 | * set, if so charge the init_mm (happens for pagecache usage). | |
569 | */ | |
570 | if (!mm) | |
571 | mm = &init_mm; | |
572 | ||
3be91277 | 573 | rcu_read_lock(); |
8a9f3ccd BS |
574 | mem = rcu_dereference(mm->mem_cgroup); |
575 | /* | |
8869b8f6 | 576 | * For every charge from the cgroup, increment reference count |
8a9f3ccd BS |
577 | */ |
578 | css_get(&mem->css); | |
579 | rcu_read_unlock(); | |
580 | ||
0eea1030 | 581 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
582 | if (!(gfp_mask & __GFP_WAIT)) |
583 | goto out; | |
e1a1cd59 BS |
584 | |
585 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
586 | continue; |
587 | ||
588 | /* | |
8869b8f6 HD |
589 | * try_to_free_mem_cgroup_pages() might not give us a full |
590 | * picture of reclaim. Some pages are reclaimed and might be | |
591 | * moved to swap cache or just unmapped from the cgroup. | |
592 | * Check the limit again to see if the reclaim reduced the | |
593 | * current usage of the cgroup before giving up | |
594 | */ | |
66e1707b BS |
595 | if (res_counter_check_under_limit(&mem->res)) |
596 | continue; | |
3be91277 HD |
597 | |
598 | if (!nr_retries--) { | |
599 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
600 | goto out; | |
66e1707b | 601 | } |
3be91277 | 602 | congestion_wait(WRITE, HZ/10); |
8a9f3ccd BS |
603 | } |
604 | ||
b9c565d5 | 605 | pc->ref_cnt = 1; |
8a9f3ccd BS |
606 | pc->mem_cgroup = mem; |
607 | pc->page = page; | |
3564c7c4 | 608 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
217bc319 KH |
609 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
610 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
3be91277 | 611 | |
7e924aaf HD |
612 | lock_page_cgroup(page); |
613 | if (page_get_page_cgroup(page)) { | |
614 | unlock_page_cgroup(page); | |
9175e031 | 615 | /* |
3be91277 HD |
616 | * Another charge has been added to this page already. |
617 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
618 | * page->cgroup, increment refcnt.... just retry is OK. |
619 | */ | |
620 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
621 | css_put(&mem->css); | |
622 | kfree(pc); | |
623 | goto retry; | |
624 | } | |
7e924aaf | 625 | page_assign_page_cgroup(page, pc); |
8a9f3ccd | 626 | |
072c56c1 KH |
627 | mz = page_cgroup_zoneinfo(pc); |
628 | spin_lock_irqsave(&mz->lru_lock, flags); | |
6d12e2d8 | 629 | __mem_cgroup_add_list(pc); |
072c56c1 | 630 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 631 | |
fb59e9f1 | 632 | unlock_page_cgroup(page); |
8a9f3ccd | 633 | done: |
8a9f3ccd | 634 | return 0; |
3be91277 HD |
635 | out: |
636 | css_put(&mem->css); | |
8a9f3ccd | 637 | kfree(pc); |
8a9f3ccd | 638 | err: |
8a9f3ccd BS |
639 | return -ENOMEM; |
640 | } | |
641 | ||
8869b8f6 | 642 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 KH |
643 | { |
644 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
8869b8f6 | 645 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
646 | } |
647 | ||
e1a1cd59 BS |
648 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
649 | gfp_t gfp_mask) | |
8697d331 | 650 | { |
8697d331 BS |
651 | if (!mm) |
652 | mm = &init_mm; | |
8869b8f6 | 653 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 654 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
8697d331 BS |
655 | } |
656 | ||
8a9f3ccd BS |
657 | /* |
658 | * Uncharging is always a welcome operation, we never complain, simply | |
8289546e | 659 | * uncharge. |
8a9f3ccd | 660 | */ |
8289546e | 661 | void mem_cgroup_uncharge_page(struct page *page) |
8a9f3ccd | 662 | { |
8289546e | 663 | struct page_cgroup *pc; |
8a9f3ccd | 664 | struct mem_cgroup *mem; |
072c56c1 | 665 | struct mem_cgroup_per_zone *mz; |
66e1707b | 666 | unsigned long flags; |
8a9f3ccd | 667 | |
8697d331 | 668 | /* |
3c541e14 | 669 | * Check if our page_cgroup is valid |
8697d331 | 670 | */ |
8289546e HD |
671 | lock_page_cgroup(page); |
672 | pc = page_get_page_cgroup(page); | |
8a9f3ccd | 673 | if (!pc) |
8289546e | 674 | goto unlock; |
8a9f3ccd | 675 | |
b9c565d5 HD |
676 | VM_BUG_ON(pc->page != page); |
677 | VM_BUG_ON(pc->ref_cnt <= 0); | |
678 | ||
679 | if (--(pc->ref_cnt) == 0) { | |
b9c565d5 HD |
680 | mz = page_cgroup_zoneinfo(pc); |
681 | spin_lock_irqsave(&mz->lru_lock, flags); | |
682 | __mem_cgroup_remove_list(pc); | |
683 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
684 | ||
fb59e9f1 HD |
685 | page_assign_page_cgroup(page, NULL); |
686 | unlock_page_cgroup(page); | |
687 | ||
6d48ff8b HD |
688 | mem = pc->mem_cgroup; |
689 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
690 | css_put(&mem->css); | |
691 | ||
b9c565d5 HD |
692 | kfree(pc); |
693 | return; | |
8a9f3ccd | 694 | } |
6d12e2d8 | 695 | |
8289546e | 696 | unlock: |
3c541e14 BS |
697 | unlock_page_cgroup(page); |
698 | } | |
699 | ||
ae41be37 KH |
700 | /* |
701 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
702 | * Refcnt of page_cgroup is incremented. | |
703 | */ | |
ae41be37 KH |
704 | int mem_cgroup_prepare_migration(struct page *page) |
705 | { | |
706 | struct page_cgroup *pc; | |
8869b8f6 | 707 | |
ae41be37 KH |
708 | lock_page_cgroup(page); |
709 | pc = page_get_page_cgroup(page); | |
b9c565d5 HD |
710 | if (pc) |
711 | pc->ref_cnt++; | |
ae41be37 | 712 | unlock_page_cgroup(page); |
b9c565d5 | 713 | return pc != NULL; |
ae41be37 KH |
714 | } |
715 | ||
716 | void mem_cgroup_end_migration(struct page *page) | |
717 | { | |
8289546e | 718 | mem_cgroup_uncharge_page(page); |
ae41be37 | 719 | } |
8869b8f6 | 720 | |
ae41be37 | 721 | /* |
8869b8f6 | 722 | * We know both *page* and *newpage* are now not-on-LRU and PG_locked. |
ae41be37 KH |
723 | * And no race with uncharge() routines because page_cgroup for *page* |
724 | * has extra one reference by mem_cgroup_prepare_migration. | |
725 | */ | |
ae41be37 KH |
726 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) |
727 | { | |
728 | struct page_cgroup *pc; | |
072c56c1 | 729 | struct mem_cgroup_per_zone *mz; |
d5b69e38 | 730 | unsigned long flags; |
8869b8f6 | 731 | |
b9c565d5 | 732 | lock_page_cgroup(page); |
ae41be37 | 733 | pc = page_get_page_cgroup(page); |
b9c565d5 HD |
734 | if (!pc) { |
735 | unlock_page_cgroup(page); | |
ae41be37 | 736 | return; |
b9c565d5 | 737 | } |
8869b8f6 | 738 | |
b9c565d5 | 739 | mz = page_cgroup_zoneinfo(pc); |
8869b8f6 | 740 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 741 | __mem_cgroup_remove_list(pc); |
072c56c1 KH |
742 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
743 | ||
fb59e9f1 HD |
744 | page_assign_page_cgroup(page, NULL); |
745 | unlock_page_cgroup(page); | |
746 | ||
ae41be37 KH |
747 | pc->page = newpage; |
748 | lock_page_cgroup(newpage); | |
749 | page_assign_page_cgroup(newpage, pc); | |
6d12e2d8 | 750 | |
072c56c1 KH |
751 | mz = page_cgroup_zoneinfo(pc); |
752 | spin_lock_irqsave(&mz->lru_lock, flags); | |
753 | __mem_cgroup_add_list(pc); | |
754 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
fb59e9f1 HD |
755 | |
756 | unlock_page_cgroup(newpage); | |
ae41be37 | 757 | } |
78fb7466 | 758 | |
cc847582 KH |
759 | /* |
760 | * This routine traverse page_cgroup in given list and drop them all. | |
761 | * This routine ignores page_cgroup->ref_cnt. | |
762 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
763 | */ | |
764 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 765 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 KH |
766 | struct mem_cgroup_per_zone *mz, |
767 | int active) | |
cc847582 KH |
768 | { |
769 | struct page_cgroup *pc; | |
770 | struct page *page; | |
9b3c0a07 | 771 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 772 | unsigned long flags; |
072c56c1 KH |
773 | struct list_head *list; |
774 | ||
775 | if (active) | |
776 | list = &mz->active_list; | |
777 | else | |
778 | list = &mz->inactive_list; | |
cc847582 | 779 | |
072c56c1 | 780 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 781 | while (!list_empty(list)) { |
cc847582 KH |
782 | pc = list_entry(list->prev, struct page_cgroup, lru); |
783 | page = pc->page; | |
9b3c0a07 HT |
784 | get_page(page); |
785 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
786 | mem_cgroup_uncharge_page(page); | |
787 | put_page(page); | |
788 | if (--count <= 0) { | |
789 | count = FORCE_UNCHARGE_BATCH; | |
790 | cond_resched(); | |
b9c565d5 | 791 | } |
9b3c0a07 | 792 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 793 | } |
072c56c1 | 794 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
795 | } |
796 | ||
797 | /* | |
798 | * make mem_cgroup's charge to be 0 if there is no task. | |
799 | * This enables deleting this mem_cgroup. | |
800 | */ | |
d5b69e38 | 801 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
802 | { |
803 | int ret = -EBUSY; | |
1ecaab2b | 804 | int node, zid; |
8869b8f6 | 805 | |
cc847582 KH |
806 | css_get(&mem->css); |
807 | /* | |
808 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 809 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
810 | * So, we have to do loop here until all lists are empty. |
811 | */ | |
1ecaab2b | 812 | while (mem->res.usage > 0) { |
cc847582 KH |
813 | if (atomic_read(&mem->css.cgroup->count) > 0) |
814 | goto out; | |
1ecaab2b KH |
815 | for_each_node_state(node, N_POSSIBLE) |
816 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
817 | struct mem_cgroup_per_zone *mz; | |
818 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
819 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 820 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 821 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 822 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 823 | } |
cc847582 KH |
824 | } |
825 | ret = 0; | |
826 | out: | |
827 | css_put(&mem->css); | |
828 | return ret; | |
829 | } | |
830 | ||
d5b69e38 | 831 | static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
0eea1030 BS |
832 | { |
833 | *tmp = memparse(buf, &buf); | |
834 | if (*buf != '\0') | |
835 | return -EINVAL; | |
836 | ||
837 | /* | |
838 | * Round up the value to the closest page size | |
839 | */ | |
840 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
841 | return 0; | |
842 | } | |
843 | ||
844 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
845 | struct cftype *cft, struct file *file, | |
846 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
847 | { |
848 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
849 | cft->private, userbuf, nbytes, ppos, |
850 | NULL); | |
8cdea7c0 BS |
851 | } |
852 | ||
853 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
854 | struct file *file, const char __user *userbuf, | |
855 | size_t nbytes, loff_t *ppos) | |
856 | { | |
857 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
858 | cft->private, userbuf, nbytes, ppos, |
859 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
860 | } |
861 | ||
cc847582 KH |
862 | static ssize_t mem_force_empty_write(struct cgroup *cont, |
863 | struct cftype *cft, struct file *file, | |
864 | const char __user *userbuf, | |
865 | size_t nbytes, loff_t *ppos) | |
866 | { | |
867 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
8869b8f6 | 868 | int ret = mem_cgroup_force_empty(mem); |
cc847582 KH |
869 | if (!ret) |
870 | ret = nbytes; | |
871 | return ret; | |
872 | } | |
873 | ||
874 | /* | |
875 | * Note: This should be removed if cgroup supports write-only file. | |
876 | */ | |
cc847582 KH |
877 | static ssize_t mem_force_empty_read(struct cgroup *cont, |
878 | struct cftype *cft, | |
879 | struct file *file, char __user *userbuf, | |
880 | size_t nbytes, loff_t *ppos) | |
881 | { | |
882 | return -EINVAL; | |
883 | } | |
884 | ||
d2ceb9b7 KH |
885 | static const struct mem_cgroup_stat_desc { |
886 | const char *msg; | |
887 | u64 unit; | |
888 | } mem_cgroup_stat_desc[] = { | |
889 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
890 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
891 | }; | |
892 | ||
893 | static int mem_control_stat_show(struct seq_file *m, void *arg) | |
894 | { | |
895 | struct cgroup *cont = m->private; | |
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; | |
905 | seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg, | |
906 | (long long)val); | |
907 | } | |
6d12e2d8 KH |
908 | /* showing # of active pages */ |
909 | { | |
910 | unsigned long active, inactive; | |
911 | ||
912 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
913 | MEM_CGROUP_ZSTAT_INACTIVE); | |
914 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
915 | MEM_CGROUP_ZSTAT_ACTIVE); | |
916 | seq_printf(m, "active %ld\n", (active) * PAGE_SIZE); | |
917 | seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE); | |
918 | } | |
d2ceb9b7 KH |
919 | return 0; |
920 | } | |
921 | ||
922 | static const struct file_operations mem_control_stat_file_operations = { | |
923 | .read = seq_read, | |
924 | .llseek = seq_lseek, | |
925 | .release = single_release, | |
926 | }; | |
927 | ||
928 | static int mem_control_stat_open(struct inode *unused, struct file *file) | |
929 | { | |
930 | /* XXX __d_cont */ | |
931 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
932 | ||
933 | file->f_op = &mem_control_stat_file_operations; | |
934 | return single_open(file, mem_control_stat_show, cont); | |
935 | } | |
936 | ||
8cdea7c0 BS |
937 | static struct cftype mem_cgroup_files[] = { |
938 | { | |
0eea1030 | 939 | .name = "usage_in_bytes", |
8cdea7c0 BS |
940 | .private = RES_USAGE, |
941 | .read = mem_cgroup_read, | |
942 | }, | |
943 | { | |
0eea1030 | 944 | .name = "limit_in_bytes", |
8cdea7c0 BS |
945 | .private = RES_LIMIT, |
946 | .write = mem_cgroup_write, | |
947 | .read = mem_cgroup_read, | |
948 | }, | |
949 | { | |
950 | .name = "failcnt", | |
951 | .private = RES_FAILCNT, | |
952 | .read = mem_cgroup_read, | |
953 | }, | |
cc847582 KH |
954 | { |
955 | .name = "force_empty", | |
956 | .write = mem_force_empty_write, | |
957 | .read = mem_force_empty_read, | |
958 | }, | |
d2ceb9b7 KH |
959 | { |
960 | .name = "stat", | |
961 | .open = mem_control_stat_open, | |
962 | }, | |
8cdea7c0 BS |
963 | }; |
964 | ||
6d12e2d8 KH |
965 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
966 | { | |
967 | struct mem_cgroup_per_node *pn; | |
1ecaab2b KH |
968 | struct mem_cgroup_per_zone *mz; |
969 | int zone; | |
970 | /* | |
971 | * This routine is called against possible nodes. | |
972 | * But it's BUG to call kmalloc() against offline node. | |
973 | * | |
974 | * TODO: this routine can waste much memory for nodes which will | |
975 | * never be onlined. It's better to use memory hotplug callback | |
976 | * function. | |
977 | */ | |
978 | if (node_state(node, N_HIGH_MEMORY)) | |
979 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
980 | else | |
981 | pn = kmalloc(sizeof(*pn), GFP_KERNEL); | |
6d12e2d8 KH |
982 | if (!pn) |
983 | return 1; | |
1ecaab2b | 984 | |
6d12e2d8 KH |
985 | mem->info.nodeinfo[node] = pn; |
986 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
987 | |
988 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
989 | mz = &pn->zoneinfo[zone]; | |
990 | INIT_LIST_HEAD(&mz->active_list); | |
991 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 992 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 993 | } |
6d12e2d8 KH |
994 | return 0; |
995 | } | |
996 | ||
1ecaab2b KH |
997 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
998 | { | |
999 | kfree(mem->info.nodeinfo[node]); | |
1000 | } | |
1001 | ||
8cdea7c0 BS |
1002 | static struct cgroup_subsys_state * |
1003 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1004 | { | |
1005 | struct mem_cgroup *mem; | |
6d12e2d8 | 1006 | int node; |
8cdea7c0 | 1007 | |
78fb7466 PE |
1008 | if (unlikely((cont->parent) == NULL)) { |
1009 | mem = &init_mem_cgroup; | |
1010 | init_mm.mem_cgroup = mem; | |
1011 | } else | |
1012 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
1013 | ||
1014 | if (mem == NULL) | |
2dda81ca | 1015 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1016 | |
1017 | res_counter_init(&mem->res); | |
1ecaab2b | 1018 | |
6d12e2d8 KH |
1019 | memset(&mem->info, 0, sizeof(mem->info)); |
1020 | ||
1021 | for_each_node_state(node, N_POSSIBLE) | |
1022 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1023 | goto free_out; | |
1024 | ||
8cdea7c0 | 1025 | return &mem->css; |
6d12e2d8 KH |
1026 | free_out: |
1027 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1028 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 KH |
1029 | if (cont->parent != NULL) |
1030 | kfree(mem); | |
2dda81ca | 1031 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1032 | } |
1033 | ||
df878fb0 KH |
1034 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1035 | struct cgroup *cont) | |
1036 | { | |
1037 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1038 | mem_cgroup_force_empty(mem); | |
1039 | } | |
1040 | ||
8cdea7c0 BS |
1041 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1042 | struct cgroup *cont) | |
1043 | { | |
6d12e2d8 KH |
1044 | int node; |
1045 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1046 | ||
1047 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1048 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1049 | |
8cdea7c0 BS |
1050 | kfree(mem_cgroup_from_cont(cont)); |
1051 | } | |
1052 | ||
1053 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1054 | struct cgroup *cont) | |
1055 | { | |
1056 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1057 | ARRAY_SIZE(mem_cgroup_files)); | |
1058 | } | |
1059 | ||
67e465a7 BS |
1060 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1061 | struct cgroup *cont, | |
1062 | struct cgroup *old_cont, | |
1063 | struct task_struct *p) | |
1064 | { | |
1065 | struct mm_struct *mm; | |
1066 | struct mem_cgroup *mem, *old_mem; | |
1067 | ||
1068 | mm = get_task_mm(p); | |
1069 | if (mm == NULL) | |
1070 | return; | |
1071 | ||
1072 | mem = mem_cgroup_from_cont(cont); | |
1073 | old_mem = mem_cgroup_from_cont(old_cont); | |
1074 | ||
1075 | if (mem == old_mem) | |
1076 | goto out; | |
1077 | ||
1078 | /* | |
1079 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1080 | * in effect owned by the leader | |
1081 | */ | |
52ea27eb | 1082 | if (!thread_group_leader(p)) |
67e465a7 BS |
1083 | goto out; |
1084 | ||
1085 | css_get(&mem->css); | |
1086 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
1087 | css_put(&old_mem->css); | |
1088 | ||
1089 | out: | |
1090 | mmput(mm); | |
67e465a7 BS |
1091 | } |
1092 | ||
8cdea7c0 BS |
1093 | struct cgroup_subsys mem_cgroup_subsys = { |
1094 | .name = "memory", | |
1095 | .subsys_id = mem_cgroup_subsys_id, | |
1096 | .create = mem_cgroup_create, | |
df878fb0 | 1097 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1098 | .destroy = mem_cgroup_destroy, |
1099 | .populate = mem_cgroup_populate, | |
67e465a7 | 1100 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1101 | .early_init = 0, |
8cdea7c0 | 1102 | }; |