]>
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 BS |
139 | }; |
140 | ||
8a9f3ccd BS |
141 | /* |
142 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
143 | * lock. We need to ensure that page->page_cgroup is at least two |
144 | * byte aligned (based on comments from Nick Piggin). But since | |
145 | * bit_spin_lock doesn't actually set that lock bit in a non-debug | |
146 | * uniprocessor kernel, we should avoid setting it here too. | |
8a9f3ccd BS |
147 | */ |
148 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
149 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) |
150 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
151 | #else | |
152 | #define PAGE_CGROUP_LOCK 0x0 | |
153 | #endif | |
8a9f3ccd | 154 | |
8cdea7c0 BS |
155 | /* |
156 | * A page_cgroup page is associated with every page descriptor. The | |
157 | * page_cgroup helps us identify information about the cgroup | |
158 | */ | |
159 | struct page_cgroup { | |
160 | struct list_head lru; /* per cgroup LRU list */ | |
161 | struct page *page; | |
162 | struct mem_cgroup *mem_cgroup; | |
8a9f3ccd BS |
163 | atomic_t ref_cnt; /* Helpful when pages move b/w */ |
164 | /* mapped and cached states */ | |
217bc319 | 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 | |
c0149530 KH |
170 | static inline int page_cgroup_nid(struct page_cgroup *pc) |
171 | { | |
172 | return page_to_nid(pc->page); | |
173 | } | |
174 | ||
175 | static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc) | |
176 | { | |
177 | return page_zonenum(pc->page); | |
178 | } | |
179 | ||
8697d331 BS |
180 | enum { |
181 | MEM_CGROUP_TYPE_UNSPEC = 0, | |
182 | MEM_CGROUP_TYPE_MAPPED, | |
183 | MEM_CGROUP_TYPE_CACHED, | |
184 | MEM_CGROUP_TYPE_ALL, | |
185 | MEM_CGROUP_TYPE_MAX, | |
186 | }; | |
187 | ||
217bc319 KH |
188 | enum charge_type { |
189 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
190 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
191 | }; | |
192 | ||
6d12e2d8 | 193 | |
d52aa412 KH |
194 | /* |
195 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
196 | */ | |
197 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
198 | bool charge) | |
199 | { | |
200 | int val = (charge)? 1 : -1; | |
201 | struct mem_cgroup_stat *stat = &mem->stat; | |
202 | VM_BUG_ON(!irqs_disabled()); | |
203 | ||
204 | if (flags & PAGE_CGROUP_FLAG_CACHE) | |
205 | __mem_cgroup_stat_add_safe(stat, | |
206 | MEM_CGROUP_STAT_CACHE, val); | |
207 | else | |
208 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
6d12e2d8 KH |
209 | } |
210 | ||
211 | static inline struct mem_cgroup_per_zone * | |
212 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
213 | { | |
214 | BUG_ON(!mem->info.nodeinfo[nid]); | |
215 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
216 | } | |
217 | ||
218 | static inline struct mem_cgroup_per_zone * | |
219 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
220 | { | |
221 | struct mem_cgroup *mem = pc->mem_cgroup; | |
222 | int nid = page_cgroup_nid(pc); | |
223 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 224 | |
6d12e2d8 KH |
225 | return mem_cgroup_zoneinfo(mem, nid, zid); |
226 | } | |
227 | ||
228 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
229 | enum mem_cgroup_zstat_index idx) | |
230 | { | |
231 | int nid, zid; | |
232 | struct mem_cgroup_per_zone *mz; | |
233 | u64 total = 0; | |
234 | ||
235 | for_each_online_node(nid) | |
236 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
237 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
238 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
239 | } | |
240 | return total; | |
d52aa412 KH |
241 | } |
242 | ||
8697d331 | 243 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 BS |
244 | |
245 | static inline | |
246 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
247 | { | |
248 | return container_of(cgroup_subsys_state(cont, | |
249 | mem_cgroup_subsys_id), struct mem_cgroup, | |
250 | css); | |
251 | } | |
252 | ||
78fb7466 PE |
253 | static inline |
254 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
255 | { | |
256 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
257 | struct mem_cgroup, css); | |
258 | } | |
259 | ||
260 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
261 | { | |
262 | struct mem_cgroup *mem; | |
263 | ||
264 | mem = mem_cgroup_from_task(p); | |
265 | css_get(&mem->css); | |
266 | mm->mem_cgroup = mem; | |
267 | } | |
268 | ||
269 | void mm_free_cgroup(struct mm_struct *mm) | |
270 | { | |
271 | css_put(&mm->mem_cgroup->css); | |
272 | } | |
273 | ||
8a9f3ccd BS |
274 | static inline int page_cgroup_locked(struct page *page) |
275 | { | |
276 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, | |
277 | &page->page_cgroup); | |
278 | } | |
279 | ||
9442ec9d | 280 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 281 | { |
9442ec9d HD |
282 | VM_BUG_ON(!page_cgroup_locked(page)); |
283 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
284 | } |
285 | ||
286 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
287 | { | |
8a9f3ccd BS |
288 | return (struct page_cgroup *) |
289 | (page->page_cgroup & ~PAGE_CGROUP_LOCK); | |
290 | } | |
291 | ||
8697d331 | 292 | static void __always_inline lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
293 | { |
294 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
295 | VM_BUG_ON(!page_cgroup_locked(page)); | |
296 | } | |
297 | ||
8697d331 | 298 | static void __always_inline unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
299 | { |
300 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
301 | } | |
302 | ||
9175e031 KH |
303 | /* |
304 | * Clear page->page_cgroup member under lock_page_cgroup(). | |
305 | * If given "pc" value is different from one page->page_cgroup, | |
306 | * page->cgroup is not cleared. | |
307 | * Returns a value of page->page_cgroup at lock taken. | |
308 | * A can can detect failure of clearing by following | |
309 | * clear_page_cgroup(page, pc) == pc | |
310 | */ | |
311 | ||
d52aa412 KH |
312 | static struct page_cgroup *clear_page_cgroup(struct page *page, |
313 | struct page_cgroup *pc) | |
9175e031 KH |
314 | { |
315 | struct page_cgroup *ret; | |
316 | /* lock and clear */ | |
317 | lock_page_cgroup(page); | |
318 | ret = page_get_page_cgroup(page); | |
319 | if (likely(ret == pc)) | |
320 | page_assign_page_cgroup(page, NULL); | |
321 | unlock_page_cgroup(page); | |
322 | return ret; | |
323 | } | |
324 | ||
6d12e2d8 KH |
325 | static void __mem_cgroup_remove_list(struct page_cgroup *pc) |
326 | { | |
327 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
328 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
329 | ||
330 | if (from) | |
331 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
332 | else | |
333 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
334 | ||
335 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
336 | list_del_init(&pc->lru); | |
337 | } | |
338 | ||
339 | static void __mem_cgroup_add_list(struct page_cgroup *pc) | |
340 | { | |
341 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
342 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
343 | ||
344 | if (!to) { | |
345 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 346 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
347 | } else { |
348 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 349 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
350 | } |
351 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
352 | } | |
353 | ||
8697d331 | 354 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 355 | { |
6d12e2d8 KH |
356 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
357 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
358 | ||
359 | if (from) | |
360 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
361 | else | |
362 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
363 | ||
3564c7c4 | 364 | if (active) { |
6d12e2d8 | 365 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 366 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 367 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 368 | } else { |
6d12e2d8 | 369 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 370 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 371 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 372 | } |
66e1707b BS |
373 | } |
374 | ||
4c4a2214 DR |
375 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
376 | { | |
377 | int ret; | |
378 | ||
379 | task_lock(task); | |
bd845e38 | 380 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
381 | task_unlock(task); |
382 | return ret; | |
383 | } | |
384 | ||
66e1707b BS |
385 | /* |
386 | * This routine assumes that the appropriate zone's lru lock is already held | |
387 | */ | |
427d5416 | 388 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 389 | { |
427d5416 | 390 | struct page_cgroup *pc; |
072c56c1 KH |
391 | struct mem_cgroup_per_zone *mz; |
392 | unsigned long flags; | |
393 | ||
427d5416 | 394 | pc = page_get_page_cgroup(page); |
66e1707b BS |
395 | if (!pc) |
396 | return; | |
397 | ||
072c56c1 KH |
398 | mz = page_cgroup_zoneinfo(pc); |
399 | spin_lock_irqsave(&mz->lru_lock, flags); | |
66e1707b | 400 | __mem_cgroup_move_lists(pc, active); |
072c56c1 | 401 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b BS |
402 | } |
403 | ||
58ae83db KH |
404 | /* |
405 | * Calculate mapped_ratio under memory controller. This will be used in | |
406 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
407 | */ | |
408 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
409 | { | |
410 | long total, rss; | |
411 | ||
412 | /* | |
413 | * usage is recorded in bytes. But, here, we assume the number of | |
414 | * physical pages can be represented by "long" on any arch. | |
415 | */ | |
416 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
417 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
418 | return (int)((rss * 100L) / total); | |
419 | } | |
5932f367 KH |
420 | /* |
421 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
422 | * between active and inactive list is calculated. For memory controller | |
423 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
424 | * zone's global lru imbalance. | |
425 | */ | |
426 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
427 | { | |
428 | unsigned long active, inactive; | |
429 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
430 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
431 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
432 | return (long) (active / (inactive + 1)); | |
433 | } | |
58ae83db | 434 | |
6c48a1d0 KH |
435 | /* |
436 | * prev_priority control...this will be used in memory reclaim path. | |
437 | */ | |
438 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
439 | { | |
440 | return mem->prev_priority; | |
441 | } | |
442 | ||
443 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
444 | { | |
445 | if (priority < mem->prev_priority) | |
446 | mem->prev_priority = priority; | |
447 | } | |
448 | ||
449 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
450 | { | |
451 | mem->prev_priority = priority; | |
452 | } | |
453 | ||
cc38108e KH |
454 | /* |
455 | * Calculate # of pages to be scanned in this priority/zone. | |
456 | * See also vmscan.c | |
457 | * | |
458 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
459 | * (see include/linux/mmzone.h) | |
460 | */ | |
461 | ||
462 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
463 | struct zone *zone, int priority) | |
464 | { | |
465 | long nr_active; | |
466 | int nid = zone->zone_pgdat->node_id; | |
467 | int zid = zone_idx(zone); | |
468 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
469 | ||
470 | nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
471 | return (nr_active >> priority); | |
472 | } | |
473 | ||
474 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
475 | struct zone *zone, int priority) | |
476 | { | |
477 | long nr_inactive; | |
478 | int nid = zone->zone_pgdat->node_id; | |
479 | int zid = zone_idx(zone); | |
480 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
481 | ||
482 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
483 | ||
484 | return (nr_inactive >> priority); | |
485 | } | |
486 | ||
66e1707b BS |
487 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
488 | struct list_head *dst, | |
489 | unsigned long *scanned, int order, | |
490 | int mode, struct zone *z, | |
491 | struct mem_cgroup *mem_cont, | |
492 | int active) | |
493 | { | |
494 | unsigned long nr_taken = 0; | |
495 | struct page *page; | |
496 | unsigned long scan; | |
497 | LIST_HEAD(pc_list); | |
498 | struct list_head *src; | |
ff7283fa | 499 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
500 | int nid = z->zone_pgdat->node_id; |
501 | int zid = zone_idx(z); | |
502 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 503 | |
1ecaab2b | 504 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 505 | if (active) |
1ecaab2b | 506 | src = &mz->active_list; |
66e1707b | 507 | else |
1ecaab2b KH |
508 | src = &mz->inactive_list; |
509 | ||
66e1707b | 510 | |
072c56c1 | 511 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
512 | scan = 0; |
513 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 514 | if (scan >= nr_to_scan) |
ff7283fa | 515 | break; |
66e1707b | 516 | page = pc->page; |
66e1707b | 517 | |
436c6541 | 518 | if (unlikely(!PageLRU(page))) |
ff7283fa | 519 | continue; |
ff7283fa | 520 | |
66e1707b BS |
521 | if (PageActive(page) && !active) { |
522 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
523 | continue; |
524 | } | |
525 | if (!PageActive(page) && active) { | |
526 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
527 | continue; |
528 | } | |
529 | ||
436c6541 HD |
530 | scan++; |
531 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
532 | |
533 | if (__isolate_lru_page(page, mode) == 0) { | |
534 | list_move(&page->lru, dst); | |
535 | nr_taken++; | |
536 | } | |
537 | } | |
538 | ||
539 | list_splice(&pc_list, src); | |
072c56c1 | 540 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
541 | |
542 | *scanned = scan; | |
543 | return nr_taken; | |
544 | } | |
545 | ||
8a9f3ccd BS |
546 | /* |
547 | * Charge the memory controller for page usage. | |
548 | * Return | |
549 | * 0 if the charge was successful | |
550 | * < 0 if the cgroup is over its limit | |
551 | */ | |
217bc319 KH |
552 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
553 | gfp_t gfp_mask, enum charge_type ctype) | |
8a9f3ccd BS |
554 | { |
555 | struct mem_cgroup *mem; | |
9175e031 | 556 | struct page_cgroup *pc; |
66e1707b BS |
557 | unsigned long flags; |
558 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 559 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd BS |
560 | |
561 | /* | |
562 | * Should page_cgroup's go to their own slab? | |
563 | * One could optimize the performance of the charging routine | |
564 | * by saving a bit in the page_flags and using it as a lock | |
565 | * to see if the cgroup page already has a page_cgroup associated | |
566 | * with it | |
567 | */ | |
66e1707b | 568 | retry: |
7e924aaf HD |
569 | lock_page_cgroup(page); |
570 | pc = page_get_page_cgroup(page); | |
571 | /* | |
572 | * The page_cgroup exists and | |
573 | * the page has already been accounted. | |
574 | */ | |
575 | if (pc) { | |
576 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { | |
577 | /* this page is under being uncharged ? */ | |
578 | unlock_page_cgroup(page); | |
579 | cpu_relax(); | |
580 | goto retry; | |
581 | } else { | |
582 | unlock_page_cgroup(page); | |
583 | goto done; | |
9175e031 | 584 | } |
8a9f3ccd | 585 | } |
7e924aaf | 586 | unlock_page_cgroup(page); |
8a9f3ccd | 587 | |
e1a1cd59 | 588 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
589 | if (pc == NULL) |
590 | goto err; | |
591 | ||
8a9f3ccd | 592 | /* |
3be91277 HD |
593 | * We always charge the cgroup the mm_struct belongs to. |
594 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
595 | * thread group leader migrates. It's possible that mm is not |
596 | * set, if so charge the init_mm (happens for pagecache usage). | |
597 | */ | |
598 | if (!mm) | |
599 | mm = &init_mm; | |
600 | ||
3be91277 | 601 | rcu_read_lock(); |
8a9f3ccd BS |
602 | mem = rcu_dereference(mm->mem_cgroup); |
603 | /* | |
604 | * For every charge from the cgroup, increment reference | |
605 | * count | |
606 | */ | |
607 | css_get(&mem->css); | |
608 | rcu_read_unlock(); | |
609 | ||
610 | /* | |
611 | * If we created the page_cgroup, we should free it on exceeding | |
612 | * the cgroup limit. | |
613 | */ | |
0eea1030 | 614 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
615 | if (!(gfp_mask & __GFP_WAIT)) |
616 | goto out; | |
e1a1cd59 BS |
617 | |
618 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
619 | continue; |
620 | ||
621 | /* | |
622 | * try_to_free_mem_cgroup_pages() might not give us a full | |
623 | * picture of reclaim. Some pages are reclaimed and might be | |
624 | * moved to swap cache or just unmapped from the cgroup. | |
625 | * Check the limit again to see if the reclaim reduced the | |
626 | * current usage of the cgroup before giving up | |
627 | */ | |
628 | if (res_counter_check_under_limit(&mem->res)) | |
629 | continue; | |
3be91277 HD |
630 | |
631 | if (!nr_retries--) { | |
632 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
633 | goto out; | |
66e1707b | 634 | } |
3be91277 | 635 | congestion_wait(WRITE, HZ/10); |
8a9f3ccd BS |
636 | } |
637 | ||
8a9f3ccd BS |
638 | atomic_set(&pc->ref_cnt, 1); |
639 | pc->mem_cgroup = mem; | |
640 | pc->page = page; | |
3564c7c4 | 641 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
217bc319 KH |
642 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
643 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
3be91277 | 644 | |
7e924aaf HD |
645 | lock_page_cgroup(page); |
646 | if (page_get_page_cgroup(page)) { | |
647 | unlock_page_cgroup(page); | |
9175e031 | 648 | /* |
3be91277 HD |
649 | * Another charge has been added to this page already. |
650 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
651 | * page->cgroup, increment refcnt.... just retry is OK. |
652 | */ | |
653 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
654 | css_put(&mem->css); | |
655 | kfree(pc); | |
656 | goto retry; | |
657 | } | |
7e924aaf HD |
658 | page_assign_page_cgroup(page, pc); |
659 | unlock_page_cgroup(page); | |
8a9f3ccd | 660 | |
072c56c1 KH |
661 | mz = page_cgroup_zoneinfo(pc); |
662 | spin_lock_irqsave(&mz->lru_lock, flags); | |
d52aa412 | 663 | /* Update statistics vector */ |
6d12e2d8 | 664 | __mem_cgroup_add_list(pc); |
072c56c1 | 665 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 666 | |
8a9f3ccd | 667 | done: |
8a9f3ccd | 668 | return 0; |
3be91277 HD |
669 | out: |
670 | css_put(&mem->css); | |
8a9f3ccd | 671 | kfree(pc); |
8a9f3ccd | 672 | err: |
8a9f3ccd BS |
673 | return -ENOMEM; |
674 | } | |
675 | ||
217bc319 KH |
676 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, |
677 | gfp_t gfp_mask) | |
678 | { | |
679 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
680 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
681 | } | |
682 | ||
8697d331 BS |
683 | /* |
684 | * See if the cached pages should be charged at all? | |
685 | */ | |
e1a1cd59 BS |
686 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
687 | gfp_t gfp_mask) | |
8697d331 | 688 | { |
ac44d354 | 689 | int ret = 0; |
8697d331 BS |
690 | if (!mm) |
691 | mm = &init_mm; | |
692 | ||
3c541e14 | 693 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 694 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
ac44d354 | 695 | return ret; |
8697d331 BS |
696 | } |
697 | ||
8a9f3ccd BS |
698 | /* |
699 | * Uncharging is always a welcome operation, we never complain, simply | |
3c541e14 | 700 | * uncharge. This routine should be called with lock_page_cgroup held |
8a9f3ccd BS |
701 | */ |
702 | void mem_cgroup_uncharge(struct page_cgroup *pc) | |
703 | { | |
704 | struct mem_cgroup *mem; | |
072c56c1 | 705 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 706 | struct page *page; |
66e1707b | 707 | unsigned long flags; |
8a9f3ccd | 708 | |
8697d331 | 709 | /* |
3c541e14 | 710 | * Check if our page_cgroup is valid |
8697d331 | 711 | */ |
8a9f3ccd BS |
712 | if (!pc) |
713 | return; | |
714 | ||
715 | if (atomic_dec_and_test(&pc->ref_cnt)) { | |
716 | page = pc->page; | |
072c56c1 | 717 | mz = page_cgroup_zoneinfo(pc); |
9175e031 KH |
718 | /* |
719 | * get page->cgroup and clear it under lock. | |
cc847582 | 720 | * force_empty can drop page->cgroup without checking refcnt. |
9175e031 | 721 | */ |
3c541e14 | 722 | unlock_page_cgroup(page); |
9175e031 KH |
723 | if (clear_page_cgroup(page, pc) == pc) { |
724 | mem = pc->mem_cgroup; | |
725 | css_put(&mem->css); | |
726 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
072c56c1 | 727 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 728 | __mem_cgroup_remove_list(pc); |
072c56c1 | 729 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9175e031 | 730 | kfree(pc); |
9175e031 | 731 | } |
3c541e14 | 732 | lock_page_cgroup(page); |
8a9f3ccd | 733 | } |
78fb7466 | 734 | } |
6d12e2d8 | 735 | |
3c541e14 BS |
736 | void mem_cgroup_uncharge_page(struct page *page) |
737 | { | |
738 | lock_page_cgroup(page); | |
739 | mem_cgroup_uncharge(page_get_page_cgroup(page)); | |
740 | unlock_page_cgroup(page); | |
741 | } | |
742 | ||
ae41be37 KH |
743 | /* |
744 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
745 | * Refcnt of page_cgroup is incremented. | |
746 | */ | |
747 | ||
748 | int mem_cgroup_prepare_migration(struct page *page) | |
749 | { | |
750 | struct page_cgroup *pc; | |
751 | int ret = 0; | |
752 | lock_page_cgroup(page); | |
753 | pc = page_get_page_cgroup(page); | |
754 | if (pc && atomic_inc_not_zero(&pc->ref_cnt)) | |
755 | ret = 1; | |
756 | unlock_page_cgroup(page); | |
757 | return ret; | |
758 | } | |
759 | ||
760 | void mem_cgroup_end_migration(struct page *page) | |
761 | { | |
3c541e14 BS |
762 | struct page_cgroup *pc; |
763 | ||
764 | lock_page_cgroup(page); | |
765 | pc = page_get_page_cgroup(page); | |
ae41be37 | 766 | mem_cgroup_uncharge(pc); |
3c541e14 | 767 | unlock_page_cgroup(page); |
ae41be37 KH |
768 | } |
769 | /* | |
770 | * We know both *page* and *newpage* are now not-on-LRU and Pg_locked. | |
771 | * And no race with uncharge() routines because page_cgroup for *page* | |
772 | * has extra one reference by mem_cgroup_prepare_migration. | |
773 | */ | |
774 | ||
775 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) | |
776 | { | |
777 | struct page_cgroup *pc; | |
6d12e2d8 KH |
778 | struct mem_cgroup *mem; |
779 | unsigned long flags; | |
072c56c1 | 780 | struct mem_cgroup_per_zone *mz; |
ae41be37 KH |
781 | retry: |
782 | pc = page_get_page_cgroup(page); | |
783 | if (!pc) | |
784 | return; | |
6d12e2d8 | 785 | mem = pc->mem_cgroup; |
072c56c1 | 786 | mz = page_cgroup_zoneinfo(pc); |
ae41be37 KH |
787 | if (clear_page_cgroup(page, pc) != pc) |
788 | goto retry; | |
072c56c1 | 789 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 KH |
790 | |
791 | __mem_cgroup_remove_list(pc); | |
072c56c1 KH |
792 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
793 | ||
ae41be37 KH |
794 | pc->page = newpage; |
795 | lock_page_cgroup(newpage); | |
796 | page_assign_page_cgroup(newpage, pc); | |
797 | unlock_page_cgroup(newpage); | |
6d12e2d8 | 798 | |
072c56c1 KH |
799 | mz = page_cgroup_zoneinfo(pc); |
800 | spin_lock_irqsave(&mz->lru_lock, flags); | |
801 | __mem_cgroup_add_list(pc); | |
802 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
ae41be37 KH |
803 | return; |
804 | } | |
78fb7466 | 805 | |
cc847582 KH |
806 | /* |
807 | * This routine traverse page_cgroup in given list and drop them all. | |
808 | * This routine ignores page_cgroup->ref_cnt. | |
809 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
810 | */ | |
811 | #define FORCE_UNCHARGE_BATCH (128) | |
812 | static void | |
072c56c1 KH |
813 | mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
814 | struct mem_cgroup_per_zone *mz, | |
815 | int active) | |
cc847582 KH |
816 | { |
817 | struct page_cgroup *pc; | |
818 | struct page *page; | |
819 | int count; | |
820 | unsigned long flags; | |
072c56c1 KH |
821 | struct list_head *list; |
822 | ||
823 | if (active) | |
824 | list = &mz->active_list; | |
825 | else | |
826 | list = &mz->inactive_list; | |
cc847582 | 827 | |
1ecaab2b KH |
828 | if (list_empty(list)) |
829 | return; | |
cc847582 KH |
830 | retry: |
831 | count = FORCE_UNCHARGE_BATCH; | |
072c56c1 | 832 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 KH |
833 | |
834 | while (--count && !list_empty(list)) { | |
835 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
836 | page = pc->page; | |
837 | /* Avoid race with charge */ | |
838 | atomic_set(&pc->ref_cnt, 0); | |
839 | if (clear_page_cgroup(page, pc) == pc) { | |
840 | css_put(&mem->css); | |
841 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
6d12e2d8 | 842 | __mem_cgroup_remove_list(pc); |
cc847582 KH |
843 | kfree(pc); |
844 | } else /* being uncharged ? ...do relax */ | |
845 | break; | |
846 | } | |
072c56c1 | 847 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
848 | if (!list_empty(list)) { |
849 | cond_resched(); | |
850 | goto retry; | |
851 | } | |
852 | return; | |
853 | } | |
854 | ||
855 | /* | |
856 | * make mem_cgroup's charge to be 0 if there is no task. | |
857 | * This enables deleting this mem_cgroup. | |
858 | */ | |
859 | ||
860 | int mem_cgroup_force_empty(struct mem_cgroup *mem) | |
861 | { | |
862 | int ret = -EBUSY; | |
1ecaab2b | 863 | int node, zid; |
cc847582 KH |
864 | css_get(&mem->css); |
865 | /* | |
866 | * page reclaim code (kswapd etc..) will move pages between | |
867 | ` * active_list <-> inactive_list while we don't take a lock. | |
868 | * So, we have to do loop here until all lists are empty. | |
869 | */ | |
1ecaab2b | 870 | while (mem->res.usage > 0) { |
cc847582 KH |
871 | if (atomic_read(&mem->css.cgroup->count) > 0) |
872 | goto out; | |
1ecaab2b KH |
873 | for_each_node_state(node, N_POSSIBLE) |
874 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
875 | struct mem_cgroup_per_zone *mz; | |
876 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
877 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 878 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 879 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 880 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 881 | } |
cc847582 KH |
882 | } |
883 | ret = 0; | |
884 | out: | |
885 | css_put(&mem->css); | |
886 | return ret; | |
887 | } | |
888 | ||
889 | ||
890 | ||
0eea1030 BS |
891 | int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
892 | { | |
893 | *tmp = memparse(buf, &buf); | |
894 | if (*buf != '\0') | |
895 | return -EINVAL; | |
896 | ||
897 | /* | |
898 | * Round up the value to the closest page size | |
899 | */ | |
900 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
901 | return 0; | |
902 | } | |
903 | ||
904 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
905 | struct cftype *cft, struct file *file, | |
906 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
907 | { |
908 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
909 | cft->private, userbuf, nbytes, ppos, |
910 | NULL); | |
8cdea7c0 BS |
911 | } |
912 | ||
913 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
914 | struct file *file, const char __user *userbuf, | |
915 | size_t nbytes, loff_t *ppos) | |
916 | { | |
917 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
918 | cft->private, userbuf, nbytes, ppos, |
919 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
920 | } |
921 | ||
cc847582 KH |
922 | static ssize_t mem_force_empty_write(struct cgroup *cont, |
923 | struct cftype *cft, struct file *file, | |
924 | const char __user *userbuf, | |
925 | size_t nbytes, loff_t *ppos) | |
926 | { | |
927 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
928 | int ret; | |
929 | ret = mem_cgroup_force_empty(mem); | |
930 | if (!ret) | |
931 | ret = nbytes; | |
932 | return ret; | |
933 | } | |
934 | ||
935 | /* | |
936 | * Note: This should be removed if cgroup supports write-only file. | |
937 | */ | |
938 | ||
939 | static ssize_t mem_force_empty_read(struct cgroup *cont, | |
940 | struct cftype *cft, | |
941 | struct file *file, char __user *userbuf, | |
942 | size_t nbytes, loff_t *ppos) | |
943 | { | |
944 | return -EINVAL; | |
945 | } | |
946 | ||
947 | ||
d2ceb9b7 KH |
948 | static const struct mem_cgroup_stat_desc { |
949 | const char *msg; | |
950 | u64 unit; | |
951 | } mem_cgroup_stat_desc[] = { | |
952 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
953 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
954 | }; | |
955 | ||
956 | static int mem_control_stat_show(struct seq_file *m, void *arg) | |
957 | { | |
958 | struct cgroup *cont = m->private; | |
959 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
960 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
961 | int i; | |
962 | ||
963 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
964 | s64 val; | |
965 | ||
966 | val = mem_cgroup_read_stat(stat, i); | |
967 | val *= mem_cgroup_stat_desc[i].unit; | |
968 | seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg, | |
969 | (long long)val); | |
970 | } | |
6d12e2d8 KH |
971 | /* showing # of active pages */ |
972 | { | |
973 | unsigned long active, inactive; | |
974 | ||
975 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
976 | MEM_CGROUP_ZSTAT_INACTIVE); | |
977 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
978 | MEM_CGROUP_ZSTAT_ACTIVE); | |
979 | seq_printf(m, "active %ld\n", (active) * PAGE_SIZE); | |
980 | seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE); | |
981 | } | |
d2ceb9b7 KH |
982 | return 0; |
983 | } | |
984 | ||
985 | static const struct file_operations mem_control_stat_file_operations = { | |
986 | .read = seq_read, | |
987 | .llseek = seq_lseek, | |
988 | .release = single_release, | |
989 | }; | |
990 | ||
991 | static int mem_control_stat_open(struct inode *unused, struct file *file) | |
992 | { | |
993 | /* XXX __d_cont */ | |
994 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
995 | ||
996 | file->f_op = &mem_control_stat_file_operations; | |
997 | return single_open(file, mem_control_stat_show, cont); | |
998 | } | |
999 | ||
1000 | ||
1001 | ||
8cdea7c0 BS |
1002 | static struct cftype mem_cgroup_files[] = { |
1003 | { | |
0eea1030 | 1004 | .name = "usage_in_bytes", |
8cdea7c0 BS |
1005 | .private = RES_USAGE, |
1006 | .read = mem_cgroup_read, | |
1007 | }, | |
1008 | { | |
0eea1030 | 1009 | .name = "limit_in_bytes", |
8cdea7c0 BS |
1010 | .private = RES_LIMIT, |
1011 | .write = mem_cgroup_write, | |
1012 | .read = mem_cgroup_read, | |
1013 | }, | |
1014 | { | |
1015 | .name = "failcnt", | |
1016 | .private = RES_FAILCNT, | |
1017 | .read = mem_cgroup_read, | |
1018 | }, | |
cc847582 KH |
1019 | { |
1020 | .name = "force_empty", | |
1021 | .write = mem_force_empty_write, | |
1022 | .read = mem_force_empty_read, | |
1023 | }, | |
d2ceb9b7 KH |
1024 | { |
1025 | .name = "stat", | |
1026 | .open = mem_control_stat_open, | |
1027 | }, | |
8cdea7c0 BS |
1028 | }; |
1029 | ||
6d12e2d8 KH |
1030 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1031 | { | |
1032 | struct mem_cgroup_per_node *pn; | |
1ecaab2b KH |
1033 | struct mem_cgroup_per_zone *mz; |
1034 | int zone; | |
1035 | /* | |
1036 | * This routine is called against possible nodes. | |
1037 | * But it's BUG to call kmalloc() against offline node. | |
1038 | * | |
1039 | * TODO: this routine can waste much memory for nodes which will | |
1040 | * never be onlined. It's better to use memory hotplug callback | |
1041 | * function. | |
1042 | */ | |
1043 | if (node_state(node, N_HIGH_MEMORY)) | |
1044 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
1045 | else | |
1046 | pn = kmalloc(sizeof(*pn), GFP_KERNEL); | |
6d12e2d8 KH |
1047 | if (!pn) |
1048 | return 1; | |
1ecaab2b | 1049 | |
6d12e2d8 KH |
1050 | mem->info.nodeinfo[node] = pn; |
1051 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1052 | |
1053 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1054 | mz = &pn->zoneinfo[zone]; | |
1055 | INIT_LIST_HEAD(&mz->active_list); | |
1056 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 1057 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 1058 | } |
6d12e2d8 KH |
1059 | return 0; |
1060 | } | |
1061 | ||
1ecaab2b KH |
1062 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1063 | { | |
1064 | kfree(mem->info.nodeinfo[node]); | |
1065 | } | |
1066 | ||
1067 | ||
78fb7466 PE |
1068 | static struct mem_cgroup init_mem_cgroup; |
1069 | ||
8cdea7c0 BS |
1070 | static struct cgroup_subsys_state * |
1071 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1072 | { | |
1073 | struct mem_cgroup *mem; | |
6d12e2d8 | 1074 | int node; |
8cdea7c0 | 1075 | |
78fb7466 PE |
1076 | if (unlikely((cont->parent) == NULL)) { |
1077 | mem = &init_mem_cgroup; | |
1078 | init_mm.mem_cgroup = mem; | |
1079 | } else | |
1080 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
1081 | ||
1082 | if (mem == NULL) | |
2dda81ca | 1083 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1084 | |
1085 | res_counter_init(&mem->res); | |
1ecaab2b | 1086 | |
6d12e2d8 KH |
1087 | memset(&mem->info, 0, sizeof(mem->info)); |
1088 | ||
1089 | for_each_node_state(node, N_POSSIBLE) | |
1090 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1091 | goto free_out; | |
1092 | ||
8cdea7c0 | 1093 | return &mem->css; |
6d12e2d8 KH |
1094 | free_out: |
1095 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1096 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 KH |
1097 | if (cont->parent != NULL) |
1098 | kfree(mem); | |
2dda81ca | 1099 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1100 | } |
1101 | ||
df878fb0 KH |
1102 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1103 | struct cgroup *cont) | |
1104 | { | |
1105 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1106 | mem_cgroup_force_empty(mem); | |
1107 | } | |
1108 | ||
8cdea7c0 BS |
1109 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1110 | struct cgroup *cont) | |
1111 | { | |
6d12e2d8 KH |
1112 | int node; |
1113 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1114 | ||
1115 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1116 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1117 | |
8cdea7c0 BS |
1118 | kfree(mem_cgroup_from_cont(cont)); |
1119 | } | |
1120 | ||
1121 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1122 | struct cgroup *cont) | |
1123 | { | |
1124 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1125 | ARRAY_SIZE(mem_cgroup_files)); | |
1126 | } | |
1127 | ||
67e465a7 BS |
1128 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1129 | struct cgroup *cont, | |
1130 | struct cgroup *old_cont, | |
1131 | struct task_struct *p) | |
1132 | { | |
1133 | struct mm_struct *mm; | |
1134 | struct mem_cgroup *mem, *old_mem; | |
1135 | ||
1136 | mm = get_task_mm(p); | |
1137 | if (mm == NULL) | |
1138 | return; | |
1139 | ||
1140 | mem = mem_cgroup_from_cont(cont); | |
1141 | old_mem = mem_cgroup_from_cont(old_cont); | |
1142 | ||
1143 | if (mem == old_mem) | |
1144 | goto out; | |
1145 | ||
1146 | /* | |
1147 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1148 | * in effect owned by the leader | |
1149 | */ | |
1150 | if (p->tgid != p->pid) | |
1151 | goto out; | |
1152 | ||
1153 | css_get(&mem->css); | |
1154 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
1155 | css_put(&old_mem->css); | |
1156 | ||
1157 | out: | |
1158 | mmput(mm); | |
1159 | return; | |
1160 | } | |
1161 | ||
8cdea7c0 BS |
1162 | struct cgroup_subsys mem_cgroup_subsys = { |
1163 | .name = "memory", | |
1164 | .subsys_id = mem_cgroup_subsys_id, | |
1165 | .create = mem_cgroup_create, | |
df878fb0 | 1166 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1167 | .destroy = mem_cgroup_destroy, |
1168 | .populate = mem_cgroup_populate, | |
67e465a7 | 1169 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1170 | .early_init = 0, |
8cdea7c0 | 1171 | }; |