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 | |
8697d331 | 134 | unsigned long control_type; /* control RSS or RSS+Pagecache */ |
6c48a1d0 | 135 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
136 | /* |
137 | * statistics. | |
138 | */ | |
139 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
140 | }; |
141 | ||
8a9f3ccd BS |
142 | /* |
143 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
144 | * lock. We need to ensure that page->page_cgroup is atleast two | |
145 | * byte aligned (based on comments from Nick Piggin) | |
146 | */ | |
147 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
148 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
149 | ||
8cdea7c0 BS |
150 | /* |
151 | * A page_cgroup page is associated with every page descriptor. The | |
152 | * page_cgroup helps us identify information about the cgroup | |
153 | */ | |
154 | struct page_cgroup { | |
155 | struct list_head lru; /* per cgroup LRU list */ | |
156 | struct page *page; | |
157 | struct mem_cgroup *mem_cgroup; | |
8a9f3ccd BS |
158 | atomic_t ref_cnt; /* Helpful when pages move b/w */ |
159 | /* mapped and cached states */ | |
217bc319 | 160 | int flags; |
8cdea7c0 | 161 | }; |
217bc319 | 162 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 163 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
8cdea7c0 | 164 | |
c0149530 KH |
165 | static inline int page_cgroup_nid(struct page_cgroup *pc) |
166 | { | |
167 | return page_to_nid(pc->page); | |
168 | } | |
169 | ||
170 | static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc) | |
171 | { | |
172 | return page_zonenum(pc->page); | |
173 | } | |
174 | ||
8697d331 BS |
175 | enum { |
176 | MEM_CGROUP_TYPE_UNSPEC = 0, | |
177 | MEM_CGROUP_TYPE_MAPPED, | |
178 | MEM_CGROUP_TYPE_CACHED, | |
179 | MEM_CGROUP_TYPE_ALL, | |
180 | MEM_CGROUP_TYPE_MAX, | |
181 | }; | |
182 | ||
217bc319 KH |
183 | enum charge_type { |
184 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
185 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
186 | }; | |
187 | ||
6d12e2d8 | 188 | |
d52aa412 KH |
189 | /* |
190 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
191 | */ | |
192 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
193 | bool charge) | |
194 | { | |
195 | int val = (charge)? 1 : -1; | |
196 | struct mem_cgroup_stat *stat = &mem->stat; | |
197 | VM_BUG_ON(!irqs_disabled()); | |
198 | ||
199 | if (flags & PAGE_CGROUP_FLAG_CACHE) | |
200 | __mem_cgroup_stat_add_safe(stat, | |
201 | MEM_CGROUP_STAT_CACHE, val); | |
202 | else | |
203 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
6d12e2d8 KH |
204 | } |
205 | ||
206 | static inline struct mem_cgroup_per_zone * | |
207 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
208 | { | |
209 | BUG_ON(!mem->info.nodeinfo[nid]); | |
210 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
211 | } | |
212 | ||
213 | static inline struct mem_cgroup_per_zone * | |
214 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
215 | { | |
216 | struct mem_cgroup *mem = pc->mem_cgroup; | |
217 | int nid = page_cgroup_nid(pc); | |
218 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 219 | |
6d12e2d8 KH |
220 | return mem_cgroup_zoneinfo(mem, nid, zid); |
221 | } | |
222 | ||
223 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
224 | enum mem_cgroup_zstat_index idx) | |
225 | { | |
226 | int nid, zid; | |
227 | struct mem_cgroup_per_zone *mz; | |
228 | u64 total = 0; | |
229 | ||
230 | for_each_online_node(nid) | |
231 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
232 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
233 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
234 | } | |
235 | return total; | |
d52aa412 KH |
236 | } |
237 | ||
8697d331 | 238 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 BS |
239 | |
240 | static inline | |
241 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
242 | { | |
243 | return container_of(cgroup_subsys_state(cont, | |
244 | mem_cgroup_subsys_id), struct mem_cgroup, | |
245 | css); | |
246 | } | |
247 | ||
78fb7466 PE |
248 | static inline |
249 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
250 | { | |
251 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
252 | struct mem_cgroup, css); | |
253 | } | |
254 | ||
255 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
256 | { | |
257 | struct mem_cgroup *mem; | |
258 | ||
259 | mem = mem_cgroup_from_task(p); | |
260 | css_get(&mem->css); | |
261 | mm->mem_cgroup = mem; | |
262 | } | |
263 | ||
264 | void mm_free_cgroup(struct mm_struct *mm) | |
265 | { | |
266 | css_put(&mm->mem_cgroup->css); | |
267 | } | |
268 | ||
8a9f3ccd BS |
269 | static inline int page_cgroup_locked(struct page *page) |
270 | { | |
271 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, | |
272 | &page->page_cgroup); | |
273 | } | |
274 | ||
78fb7466 PE |
275 | void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
276 | { | |
8a9f3ccd BS |
277 | int locked; |
278 | ||
279 | /* | |
280 | * While resetting the page_cgroup we might not hold the | |
281 | * page_cgroup lock. free_hot_cold_page() is an example | |
282 | * of such a scenario | |
283 | */ | |
284 | if (pc) | |
285 | VM_BUG_ON(!page_cgroup_locked(page)); | |
286 | locked = (page->page_cgroup & PAGE_CGROUP_LOCK); | |
287 | page->page_cgroup = ((unsigned long)pc | locked); | |
78fb7466 PE |
288 | } |
289 | ||
290 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
291 | { | |
8a9f3ccd BS |
292 | return (struct page_cgroup *) |
293 | (page->page_cgroup & ~PAGE_CGROUP_LOCK); | |
294 | } | |
295 | ||
8697d331 | 296 | static void __always_inline lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
297 | { |
298 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
299 | VM_BUG_ON(!page_cgroup_locked(page)); | |
300 | } | |
301 | ||
8697d331 | 302 | static void __always_inline unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
303 | { |
304 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
305 | } | |
306 | ||
9175e031 KH |
307 | /* |
308 | * Tie new page_cgroup to struct page under lock_page_cgroup() | |
309 | * This can fail if the page has been tied to a page_cgroup. | |
310 | * If success, returns 0. | |
311 | */ | |
d52aa412 KH |
312 | static int page_cgroup_assign_new_page_cgroup(struct page *page, |
313 | struct page_cgroup *pc) | |
9175e031 KH |
314 | { |
315 | int ret = 0; | |
316 | ||
317 | lock_page_cgroup(page); | |
318 | if (!page_get_page_cgroup(page)) | |
319 | page_assign_page_cgroup(page, pc); | |
320 | else /* A page is tied to other pc. */ | |
321 | ret = 1; | |
322 | unlock_page_cgroup(page); | |
323 | return ret; | |
324 | } | |
325 | ||
326 | /* | |
327 | * Clear page->page_cgroup member under lock_page_cgroup(). | |
328 | * If given "pc" value is different from one page->page_cgroup, | |
329 | * page->cgroup is not cleared. | |
330 | * Returns a value of page->page_cgroup at lock taken. | |
331 | * A can can detect failure of clearing by following | |
332 | * clear_page_cgroup(page, pc) == pc | |
333 | */ | |
334 | ||
d52aa412 KH |
335 | static struct page_cgroup *clear_page_cgroup(struct page *page, |
336 | struct page_cgroup *pc) | |
9175e031 KH |
337 | { |
338 | struct page_cgroup *ret; | |
339 | /* lock and clear */ | |
340 | lock_page_cgroup(page); | |
341 | ret = page_get_page_cgroup(page); | |
342 | if (likely(ret == pc)) | |
343 | page_assign_page_cgroup(page, NULL); | |
344 | unlock_page_cgroup(page); | |
345 | return ret; | |
346 | } | |
347 | ||
6d12e2d8 KH |
348 | static void __mem_cgroup_remove_list(struct page_cgroup *pc) |
349 | { | |
350 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
351 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
352 | ||
353 | if (from) | |
354 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
355 | else | |
356 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
357 | ||
358 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
359 | list_del_init(&pc->lru); | |
360 | } | |
361 | ||
362 | static void __mem_cgroup_add_list(struct page_cgroup *pc) | |
363 | { | |
364 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
365 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
366 | ||
367 | if (!to) { | |
368 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 369 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
370 | } else { |
371 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 372 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
373 | } |
374 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
375 | } | |
376 | ||
8697d331 | 377 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 378 | { |
6d12e2d8 KH |
379 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
380 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
381 | ||
382 | if (from) | |
383 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
384 | else | |
385 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
386 | ||
3564c7c4 | 387 | if (active) { |
6d12e2d8 | 388 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 389 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 390 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 391 | } else { |
6d12e2d8 | 392 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 393 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 394 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 395 | } |
66e1707b BS |
396 | } |
397 | ||
4c4a2214 DR |
398 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
399 | { | |
400 | int ret; | |
401 | ||
402 | task_lock(task); | |
403 | ret = task->mm && mm_cgroup(task->mm) == mem; | |
404 | task_unlock(task); | |
405 | return ret; | |
406 | } | |
407 | ||
66e1707b BS |
408 | /* |
409 | * This routine assumes that the appropriate zone's lru lock is already held | |
410 | */ | |
411 | void mem_cgroup_move_lists(struct page_cgroup *pc, bool active) | |
412 | { | |
072c56c1 KH |
413 | struct mem_cgroup_per_zone *mz; |
414 | unsigned long flags; | |
415 | ||
66e1707b BS |
416 | if (!pc) |
417 | return; | |
418 | ||
072c56c1 KH |
419 | mz = page_cgroup_zoneinfo(pc); |
420 | spin_lock_irqsave(&mz->lru_lock, flags); | |
66e1707b | 421 | __mem_cgroup_move_lists(pc, active); |
072c56c1 | 422 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b BS |
423 | } |
424 | ||
58ae83db KH |
425 | /* |
426 | * Calculate mapped_ratio under memory controller. This will be used in | |
427 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
428 | */ | |
429 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
430 | { | |
431 | long total, rss; | |
432 | ||
433 | /* | |
434 | * usage is recorded in bytes. But, here, we assume the number of | |
435 | * physical pages can be represented by "long" on any arch. | |
436 | */ | |
437 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
438 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
439 | return (int)((rss * 100L) / total); | |
440 | } | |
5932f367 KH |
441 | /* |
442 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
443 | * between active and inactive list is calculated. For memory controller | |
444 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
445 | * zone's global lru imbalance. | |
446 | */ | |
447 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
448 | { | |
449 | unsigned long active, inactive; | |
450 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
451 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
452 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
453 | return (long) (active / (inactive + 1)); | |
454 | } | |
58ae83db | 455 | |
6c48a1d0 KH |
456 | /* |
457 | * prev_priority control...this will be used in memory reclaim path. | |
458 | */ | |
459 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
460 | { | |
461 | return mem->prev_priority; | |
462 | } | |
463 | ||
464 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
465 | { | |
466 | if (priority < mem->prev_priority) | |
467 | mem->prev_priority = priority; | |
468 | } | |
469 | ||
470 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
471 | { | |
472 | mem->prev_priority = priority; | |
473 | } | |
474 | ||
cc38108e KH |
475 | /* |
476 | * Calculate # of pages to be scanned in this priority/zone. | |
477 | * See also vmscan.c | |
478 | * | |
479 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
480 | * (see include/linux/mmzone.h) | |
481 | */ | |
482 | ||
483 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
484 | struct zone *zone, int priority) | |
485 | { | |
486 | long nr_active; | |
487 | int nid = zone->zone_pgdat->node_id; | |
488 | int zid = zone_idx(zone); | |
489 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
490 | ||
491 | nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
492 | return (nr_active >> priority); | |
493 | } | |
494 | ||
495 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
496 | struct zone *zone, int priority) | |
497 | { | |
498 | long nr_inactive; | |
499 | int nid = zone->zone_pgdat->node_id; | |
500 | int zid = zone_idx(zone); | |
501 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
502 | ||
503 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
504 | ||
505 | return (nr_inactive >> priority); | |
506 | } | |
507 | ||
66e1707b BS |
508 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
509 | struct list_head *dst, | |
510 | unsigned long *scanned, int order, | |
511 | int mode, struct zone *z, | |
512 | struct mem_cgroup *mem_cont, | |
513 | int active) | |
514 | { | |
515 | unsigned long nr_taken = 0; | |
516 | struct page *page; | |
517 | unsigned long scan; | |
518 | LIST_HEAD(pc_list); | |
519 | struct list_head *src; | |
ff7283fa | 520 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
521 | int nid = z->zone_pgdat->node_id; |
522 | int zid = zone_idx(z); | |
523 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 524 | |
1ecaab2b | 525 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 526 | if (active) |
1ecaab2b | 527 | src = &mz->active_list; |
66e1707b | 528 | else |
1ecaab2b KH |
529 | src = &mz->inactive_list; |
530 | ||
66e1707b | 531 | |
072c56c1 | 532 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
533 | scan = 0; |
534 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 535 | if (scan >= nr_to_scan) |
ff7283fa | 536 | break; |
66e1707b BS |
537 | page = pc->page; |
538 | VM_BUG_ON(!pc); | |
539 | ||
436c6541 | 540 | if (unlikely(!PageLRU(page))) |
ff7283fa | 541 | continue; |
ff7283fa | 542 | |
66e1707b BS |
543 | if (PageActive(page) && !active) { |
544 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
545 | continue; |
546 | } | |
547 | if (!PageActive(page) && active) { | |
548 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
549 | continue; |
550 | } | |
551 | ||
436c6541 HD |
552 | scan++; |
553 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
554 | |
555 | if (__isolate_lru_page(page, mode) == 0) { | |
556 | list_move(&page->lru, dst); | |
557 | nr_taken++; | |
558 | } | |
559 | } | |
560 | ||
561 | list_splice(&pc_list, src); | |
072c56c1 | 562 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
563 | |
564 | *scanned = scan; | |
565 | return nr_taken; | |
566 | } | |
567 | ||
8a9f3ccd BS |
568 | /* |
569 | * Charge the memory controller for page usage. | |
570 | * Return | |
571 | * 0 if the charge was successful | |
572 | * < 0 if the cgroup is over its limit | |
573 | */ | |
217bc319 KH |
574 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
575 | gfp_t gfp_mask, enum charge_type ctype) | |
8a9f3ccd BS |
576 | { |
577 | struct mem_cgroup *mem; | |
9175e031 | 578 | struct page_cgroup *pc; |
66e1707b BS |
579 | unsigned long flags; |
580 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 581 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd BS |
582 | |
583 | /* | |
584 | * Should page_cgroup's go to their own slab? | |
585 | * One could optimize the performance of the charging routine | |
586 | * by saving a bit in the page_flags and using it as a lock | |
587 | * to see if the cgroup page already has a page_cgroup associated | |
588 | * with it | |
589 | */ | |
66e1707b | 590 | retry: |
82369553 HD |
591 | if (page) { |
592 | lock_page_cgroup(page); | |
593 | pc = page_get_page_cgroup(page); | |
594 | /* | |
595 | * The page_cgroup exists and | |
596 | * the page has already been accounted. | |
597 | */ | |
598 | if (pc) { | |
599 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { | |
600 | /* this page is under being uncharged ? */ | |
601 | unlock_page_cgroup(page); | |
602 | cpu_relax(); | |
603 | goto retry; | |
604 | } else { | |
605 | unlock_page_cgroup(page); | |
606 | goto done; | |
607 | } | |
9175e031 | 608 | } |
82369553 | 609 | unlock_page_cgroup(page); |
8a9f3ccd | 610 | } |
8a9f3ccd | 611 | |
e1a1cd59 | 612 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
613 | if (pc == NULL) |
614 | goto err; | |
615 | ||
8a9f3ccd | 616 | /* |
3be91277 HD |
617 | * We always charge the cgroup the mm_struct belongs to. |
618 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
619 | * thread group leader migrates. It's possible that mm is not |
620 | * set, if so charge the init_mm (happens for pagecache usage). | |
621 | */ | |
622 | if (!mm) | |
623 | mm = &init_mm; | |
624 | ||
3be91277 | 625 | rcu_read_lock(); |
8a9f3ccd BS |
626 | mem = rcu_dereference(mm->mem_cgroup); |
627 | /* | |
628 | * For every charge from the cgroup, increment reference | |
629 | * count | |
630 | */ | |
631 | css_get(&mem->css); | |
632 | rcu_read_unlock(); | |
633 | ||
634 | /* | |
635 | * If we created the page_cgroup, we should free it on exceeding | |
636 | * the cgroup limit. | |
637 | */ | |
0eea1030 | 638 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
639 | if (!(gfp_mask & __GFP_WAIT)) |
640 | goto out; | |
e1a1cd59 BS |
641 | |
642 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
643 | continue; |
644 | ||
645 | /* | |
646 | * try_to_free_mem_cgroup_pages() might not give us a full | |
647 | * picture of reclaim. Some pages are reclaimed and might be | |
648 | * moved to swap cache or just unmapped from the cgroup. | |
649 | * Check the limit again to see if the reclaim reduced the | |
650 | * current usage of the cgroup before giving up | |
651 | */ | |
652 | if (res_counter_check_under_limit(&mem->res)) | |
653 | continue; | |
3be91277 HD |
654 | |
655 | if (!nr_retries--) { | |
656 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
657 | goto out; | |
66e1707b | 658 | } |
3be91277 | 659 | congestion_wait(WRITE, HZ/10); |
8a9f3ccd BS |
660 | } |
661 | ||
8a9f3ccd BS |
662 | atomic_set(&pc->ref_cnt, 1); |
663 | pc->mem_cgroup = mem; | |
664 | pc->page = page; | |
3564c7c4 | 665 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
217bc319 KH |
666 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
667 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
3be91277 | 668 | |
82369553 | 669 | if (!page || page_cgroup_assign_new_page_cgroup(page, pc)) { |
9175e031 | 670 | /* |
3be91277 HD |
671 | * Another charge has been added to this page already. |
672 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
673 | * page->cgroup, increment refcnt.... just retry is OK. |
674 | */ | |
675 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
676 | css_put(&mem->css); | |
677 | kfree(pc); | |
82369553 HD |
678 | if (!page) |
679 | goto done; | |
9175e031 KH |
680 | goto retry; |
681 | } | |
8a9f3ccd | 682 | |
072c56c1 KH |
683 | mz = page_cgroup_zoneinfo(pc); |
684 | spin_lock_irqsave(&mz->lru_lock, flags); | |
d52aa412 | 685 | /* Update statistics vector */ |
6d12e2d8 | 686 | __mem_cgroup_add_list(pc); |
072c56c1 | 687 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 688 | |
8a9f3ccd | 689 | done: |
8a9f3ccd | 690 | return 0; |
3be91277 HD |
691 | out: |
692 | css_put(&mem->css); | |
8a9f3ccd | 693 | kfree(pc); |
8a9f3ccd | 694 | err: |
8a9f3ccd BS |
695 | return -ENOMEM; |
696 | } | |
697 | ||
217bc319 KH |
698 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, |
699 | gfp_t gfp_mask) | |
700 | { | |
701 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
702 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
703 | } | |
704 | ||
8697d331 BS |
705 | /* |
706 | * See if the cached pages should be charged at all? | |
707 | */ | |
e1a1cd59 BS |
708 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
709 | gfp_t gfp_mask) | |
8697d331 | 710 | { |
ac44d354 | 711 | int ret = 0; |
8697d331 BS |
712 | struct mem_cgroup *mem; |
713 | if (!mm) | |
714 | mm = &init_mm; | |
715 | ||
ac44d354 | 716 | rcu_read_lock(); |
8697d331 | 717 | mem = rcu_dereference(mm->mem_cgroup); |
ac44d354 BS |
718 | css_get(&mem->css); |
719 | rcu_read_unlock(); | |
8697d331 | 720 | if (mem->control_type == MEM_CGROUP_TYPE_ALL) |
ac44d354 | 721 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 722 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
ac44d354 BS |
723 | css_put(&mem->css); |
724 | return ret; | |
8697d331 BS |
725 | } |
726 | ||
8a9f3ccd BS |
727 | /* |
728 | * Uncharging is always a welcome operation, we never complain, simply | |
729 | * uncharge. | |
730 | */ | |
731 | void mem_cgroup_uncharge(struct page_cgroup *pc) | |
732 | { | |
733 | struct mem_cgroup *mem; | |
072c56c1 | 734 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 735 | struct page *page; |
66e1707b | 736 | unsigned long flags; |
8a9f3ccd | 737 | |
8697d331 BS |
738 | /* |
739 | * This can handle cases when a page is not charged at all and we | |
740 | * are switching between handling the control_type. | |
741 | */ | |
8a9f3ccd BS |
742 | if (!pc) |
743 | return; | |
744 | ||
745 | if (atomic_dec_and_test(&pc->ref_cnt)) { | |
746 | page = pc->page; | |
072c56c1 | 747 | mz = page_cgroup_zoneinfo(pc); |
9175e031 KH |
748 | /* |
749 | * get page->cgroup and clear it under lock. | |
cc847582 | 750 | * force_empty can drop page->cgroup without checking refcnt. |
9175e031 KH |
751 | */ |
752 | if (clear_page_cgroup(page, pc) == pc) { | |
753 | mem = pc->mem_cgroup; | |
754 | css_put(&mem->css); | |
755 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
072c56c1 | 756 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 757 | __mem_cgroup_remove_list(pc); |
072c56c1 | 758 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9175e031 | 759 | kfree(pc); |
9175e031 | 760 | } |
8a9f3ccd | 761 | } |
78fb7466 | 762 | } |
6d12e2d8 | 763 | |
ae41be37 KH |
764 | /* |
765 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
766 | * Refcnt of page_cgroup is incremented. | |
767 | */ | |
768 | ||
769 | int mem_cgroup_prepare_migration(struct page *page) | |
770 | { | |
771 | struct page_cgroup *pc; | |
772 | int ret = 0; | |
773 | lock_page_cgroup(page); | |
774 | pc = page_get_page_cgroup(page); | |
775 | if (pc && atomic_inc_not_zero(&pc->ref_cnt)) | |
776 | ret = 1; | |
777 | unlock_page_cgroup(page); | |
778 | return ret; | |
779 | } | |
780 | ||
781 | void mem_cgroup_end_migration(struct page *page) | |
782 | { | |
783 | struct page_cgroup *pc = page_get_page_cgroup(page); | |
784 | mem_cgroup_uncharge(pc); | |
785 | } | |
786 | /* | |
787 | * We know both *page* and *newpage* are now not-on-LRU and Pg_locked. | |
788 | * And no race with uncharge() routines because page_cgroup for *page* | |
789 | * has extra one reference by mem_cgroup_prepare_migration. | |
790 | */ | |
791 | ||
792 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) | |
793 | { | |
794 | struct page_cgroup *pc; | |
6d12e2d8 KH |
795 | struct mem_cgroup *mem; |
796 | unsigned long flags; | |
072c56c1 | 797 | struct mem_cgroup_per_zone *mz; |
ae41be37 KH |
798 | retry: |
799 | pc = page_get_page_cgroup(page); | |
800 | if (!pc) | |
801 | return; | |
6d12e2d8 | 802 | mem = pc->mem_cgroup; |
072c56c1 | 803 | mz = page_cgroup_zoneinfo(pc); |
ae41be37 KH |
804 | if (clear_page_cgroup(page, pc) != pc) |
805 | goto retry; | |
072c56c1 | 806 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 KH |
807 | |
808 | __mem_cgroup_remove_list(pc); | |
072c56c1 KH |
809 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
810 | ||
ae41be37 KH |
811 | pc->page = newpage; |
812 | lock_page_cgroup(newpage); | |
813 | page_assign_page_cgroup(newpage, pc); | |
814 | unlock_page_cgroup(newpage); | |
6d12e2d8 | 815 | |
072c56c1 KH |
816 | mz = page_cgroup_zoneinfo(pc); |
817 | spin_lock_irqsave(&mz->lru_lock, flags); | |
818 | __mem_cgroup_add_list(pc); | |
819 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
ae41be37 KH |
820 | return; |
821 | } | |
78fb7466 | 822 | |
cc847582 KH |
823 | /* |
824 | * This routine traverse page_cgroup in given list and drop them all. | |
825 | * This routine ignores page_cgroup->ref_cnt. | |
826 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
827 | */ | |
828 | #define FORCE_UNCHARGE_BATCH (128) | |
829 | static void | |
072c56c1 KH |
830 | mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
831 | struct mem_cgroup_per_zone *mz, | |
832 | int active) | |
cc847582 KH |
833 | { |
834 | struct page_cgroup *pc; | |
835 | struct page *page; | |
836 | int count; | |
837 | unsigned long flags; | |
072c56c1 KH |
838 | struct list_head *list; |
839 | ||
840 | if (active) | |
841 | list = &mz->active_list; | |
842 | else | |
843 | list = &mz->inactive_list; | |
cc847582 | 844 | |
1ecaab2b KH |
845 | if (list_empty(list)) |
846 | return; | |
cc847582 KH |
847 | retry: |
848 | count = FORCE_UNCHARGE_BATCH; | |
072c56c1 | 849 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 KH |
850 | |
851 | while (--count && !list_empty(list)) { | |
852 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
853 | page = pc->page; | |
854 | /* Avoid race with charge */ | |
855 | atomic_set(&pc->ref_cnt, 0); | |
856 | if (clear_page_cgroup(page, pc) == pc) { | |
857 | css_put(&mem->css); | |
858 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
6d12e2d8 | 859 | __mem_cgroup_remove_list(pc); |
cc847582 KH |
860 | kfree(pc); |
861 | } else /* being uncharged ? ...do relax */ | |
862 | break; | |
863 | } | |
072c56c1 | 864 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
865 | if (!list_empty(list)) { |
866 | cond_resched(); | |
867 | goto retry; | |
868 | } | |
869 | return; | |
870 | } | |
871 | ||
872 | /* | |
873 | * make mem_cgroup's charge to be 0 if there is no task. | |
874 | * This enables deleting this mem_cgroup. | |
875 | */ | |
876 | ||
877 | int mem_cgroup_force_empty(struct mem_cgroup *mem) | |
878 | { | |
879 | int ret = -EBUSY; | |
1ecaab2b | 880 | int node, zid; |
cc847582 KH |
881 | css_get(&mem->css); |
882 | /* | |
883 | * page reclaim code (kswapd etc..) will move pages between | |
884 | ` * active_list <-> inactive_list while we don't take a lock. | |
885 | * So, we have to do loop here until all lists are empty. | |
886 | */ | |
1ecaab2b | 887 | while (mem->res.usage > 0) { |
cc847582 KH |
888 | if (atomic_read(&mem->css.cgroup->count) > 0) |
889 | goto out; | |
1ecaab2b KH |
890 | for_each_node_state(node, N_POSSIBLE) |
891 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
892 | struct mem_cgroup_per_zone *mz; | |
893 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
894 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 895 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 896 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 897 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 898 | } |
cc847582 KH |
899 | } |
900 | ret = 0; | |
901 | out: | |
902 | css_put(&mem->css); | |
903 | return ret; | |
904 | } | |
905 | ||
906 | ||
907 | ||
0eea1030 BS |
908 | int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
909 | { | |
910 | *tmp = memparse(buf, &buf); | |
911 | if (*buf != '\0') | |
912 | return -EINVAL; | |
913 | ||
914 | /* | |
915 | * Round up the value to the closest page size | |
916 | */ | |
917 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
918 | return 0; | |
919 | } | |
920 | ||
921 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
922 | struct cftype *cft, struct file *file, | |
923 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
924 | { |
925 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
926 | cft->private, userbuf, nbytes, ppos, |
927 | NULL); | |
8cdea7c0 BS |
928 | } |
929 | ||
930 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
931 | struct file *file, const char __user *userbuf, | |
932 | size_t nbytes, loff_t *ppos) | |
933 | { | |
934 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
935 | cft->private, userbuf, nbytes, ppos, |
936 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
937 | } |
938 | ||
8697d331 BS |
939 | static ssize_t mem_control_type_write(struct cgroup *cont, |
940 | struct cftype *cft, struct file *file, | |
941 | const char __user *userbuf, | |
942 | size_t nbytes, loff_t *pos) | |
943 | { | |
944 | int ret; | |
945 | char *buf, *end; | |
946 | unsigned long tmp; | |
947 | struct mem_cgroup *mem; | |
948 | ||
949 | mem = mem_cgroup_from_cont(cont); | |
950 | buf = kmalloc(nbytes + 1, GFP_KERNEL); | |
951 | ret = -ENOMEM; | |
952 | if (buf == NULL) | |
953 | goto out; | |
954 | ||
955 | buf[nbytes] = 0; | |
956 | ret = -EFAULT; | |
957 | if (copy_from_user(buf, userbuf, nbytes)) | |
958 | goto out_free; | |
959 | ||
960 | ret = -EINVAL; | |
961 | tmp = simple_strtoul(buf, &end, 10); | |
962 | if (*end != '\0') | |
963 | goto out_free; | |
964 | ||
965 | if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX) | |
966 | goto out_free; | |
967 | ||
968 | mem->control_type = tmp; | |
969 | ret = nbytes; | |
970 | out_free: | |
971 | kfree(buf); | |
972 | out: | |
973 | return ret; | |
974 | } | |
975 | ||
976 | static ssize_t mem_control_type_read(struct cgroup *cont, | |
977 | struct cftype *cft, | |
978 | struct file *file, char __user *userbuf, | |
979 | size_t nbytes, loff_t *ppos) | |
980 | { | |
981 | unsigned long val; | |
982 | char buf[64], *s; | |
983 | struct mem_cgroup *mem; | |
984 | ||
985 | mem = mem_cgroup_from_cont(cont); | |
986 | s = buf; | |
987 | val = mem->control_type; | |
988 | s += sprintf(s, "%lu\n", val); | |
989 | return simple_read_from_buffer((void __user *)userbuf, nbytes, | |
990 | ppos, buf, s - buf); | |
991 | } | |
992 | ||
cc847582 KH |
993 | |
994 | static ssize_t mem_force_empty_write(struct cgroup *cont, | |
995 | struct cftype *cft, struct file *file, | |
996 | const char __user *userbuf, | |
997 | size_t nbytes, loff_t *ppos) | |
998 | { | |
999 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1000 | int ret; | |
1001 | ret = mem_cgroup_force_empty(mem); | |
1002 | if (!ret) | |
1003 | ret = nbytes; | |
1004 | return ret; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Note: This should be removed if cgroup supports write-only file. | |
1009 | */ | |
1010 | ||
1011 | static ssize_t mem_force_empty_read(struct cgroup *cont, | |
1012 | struct cftype *cft, | |
1013 | struct file *file, char __user *userbuf, | |
1014 | size_t nbytes, loff_t *ppos) | |
1015 | { | |
1016 | return -EINVAL; | |
1017 | } | |
1018 | ||
1019 | ||
d2ceb9b7 KH |
1020 | static const struct mem_cgroup_stat_desc { |
1021 | const char *msg; | |
1022 | u64 unit; | |
1023 | } mem_cgroup_stat_desc[] = { | |
1024 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1025 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
1026 | }; | |
1027 | ||
1028 | static int mem_control_stat_show(struct seq_file *m, void *arg) | |
1029 | { | |
1030 | struct cgroup *cont = m->private; | |
1031 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
1032 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1033 | int i; | |
1034 | ||
1035 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1036 | s64 val; | |
1037 | ||
1038 | val = mem_cgroup_read_stat(stat, i); | |
1039 | val *= mem_cgroup_stat_desc[i].unit; | |
1040 | seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg, | |
1041 | (long long)val); | |
1042 | } | |
6d12e2d8 KH |
1043 | /* showing # of active pages */ |
1044 | { | |
1045 | unsigned long active, inactive; | |
1046 | ||
1047 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
1048 | MEM_CGROUP_ZSTAT_INACTIVE); | |
1049 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
1050 | MEM_CGROUP_ZSTAT_ACTIVE); | |
1051 | seq_printf(m, "active %ld\n", (active) * PAGE_SIZE); | |
1052 | seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE); | |
1053 | } | |
d2ceb9b7 KH |
1054 | return 0; |
1055 | } | |
1056 | ||
1057 | static const struct file_operations mem_control_stat_file_operations = { | |
1058 | .read = seq_read, | |
1059 | .llseek = seq_lseek, | |
1060 | .release = single_release, | |
1061 | }; | |
1062 | ||
1063 | static int mem_control_stat_open(struct inode *unused, struct file *file) | |
1064 | { | |
1065 | /* XXX __d_cont */ | |
1066 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
1067 | ||
1068 | file->f_op = &mem_control_stat_file_operations; | |
1069 | return single_open(file, mem_control_stat_show, cont); | |
1070 | } | |
1071 | ||
1072 | ||
1073 | ||
8cdea7c0 BS |
1074 | static struct cftype mem_cgroup_files[] = { |
1075 | { | |
0eea1030 | 1076 | .name = "usage_in_bytes", |
8cdea7c0 BS |
1077 | .private = RES_USAGE, |
1078 | .read = mem_cgroup_read, | |
1079 | }, | |
1080 | { | |
0eea1030 | 1081 | .name = "limit_in_bytes", |
8cdea7c0 BS |
1082 | .private = RES_LIMIT, |
1083 | .write = mem_cgroup_write, | |
1084 | .read = mem_cgroup_read, | |
1085 | }, | |
1086 | { | |
1087 | .name = "failcnt", | |
1088 | .private = RES_FAILCNT, | |
1089 | .read = mem_cgroup_read, | |
1090 | }, | |
8697d331 BS |
1091 | { |
1092 | .name = "control_type", | |
1093 | .write = mem_control_type_write, | |
1094 | .read = mem_control_type_read, | |
1095 | }, | |
cc847582 KH |
1096 | { |
1097 | .name = "force_empty", | |
1098 | .write = mem_force_empty_write, | |
1099 | .read = mem_force_empty_read, | |
1100 | }, | |
d2ceb9b7 KH |
1101 | { |
1102 | .name = "stat", | |
1103 | .open = mem_control_stat_open, | |
1104 | }, | |
8cdea7c0 BS |
1105 | }; |
1106 | ||
6d12e2d8 KH |
1107 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1108 | { | |
1109 | struct mem_cgroup_per_node *pn; | |
1ecaab2b KH |
1110 | struct mem_cgroup_per_zone *mz; |
1111 | int zone; | |
1112 | /* | |
1113 | * This routine is called against possible nodes. | |
1114 | * But it's BUG to call kmalloc() against offline node. | |
1115 | * | |
1116 | * TODO: this routine can waste much memory for nodes which will | |
1117 | * never be onlined. It's better to use memory hotplug callback | |
1118 | * function. | |
1119 | */ | |
1120 | if (node_state(node, N_HIGH_MEMORY)) | |
1121 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
1122 | else | |
1123 | pn = kmalloc(sizeof(*pn), GFP_KERNEL); | |
6d12e2d8 KH |
1124 | if (!pn) |
1125 | return 1; | |
1ecaab2b | 1126 | |
6d12e2d8 KH |
1127 | mem->info.nodeinfo[node] = pn; |
1128 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1129 | |
1130 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1131 | mz = &pn->zoneinfo[zone]; | |
1132 | INIT_LIST_HEAD(&mz->active_list); | |
1133 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 1134 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 1135 | } |
6d12e2d8 KH |
1136 | return 0; |
1137 | } | |
1138 | ||
1ecaab2b KH |
1139 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1140 | { | |
1141 | kfree(mem->info.nodeinfo[node]); | |
1142 | } | |
1143 | ||
1144 | ||
78fb7466 PE |
1145 | static struct mem_cgroup init_mem_cgroup; |
1146 | ||
8cdea7c0 BS |
1147 | static struct cgroup_subsys_state * |
1148 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1149 | { | |
1150 | struct mem_cgroup *mem; | |
6d12e2d8 | 1151 | int node; |
8cdea7c0 | 1152 | |
78fb7466 PE |
1153 | if (unlikely((cont->parent) == NULL)) { |
1154 | mem = &init_mem_cgroup; | |
1155 | init_mm.mem_cgroup = mem; | |
1156 | } else | |
1157 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
1158 | ||
1159 | if (mem == NULL) | |
1160 | return NULL; | |
8cdea7c0 BS |
1161 | |
1162 | res_counter_init(&mem->res); | |
1ecaab2b | 1163 | |
8697d331 | 1164 | mem->control_type = MEM_CGROUP_TYPE_ALL; |
6d12e2d8 KH |
1165 | memset(&mem->info, 0, sizeof(mem->info)); |
1166 | ||
1167 | for_each_node_state(node, N_POSSIBLE) | |
1168 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1169 | goto free_out; | |
1170 | ||
8cdea7c0 | 1171 | return &mem->css; |
6d12e2d8 KH |
1172 | free_out: |
1173 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1174 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 KH |
1175 | if (cont->parent != NULL) |
1176 | kfree(mem); | |
1177 | return NULL; | |
8cdea7c0 BS |
1178 | } |
1179 | ||
df878fb0 KH |
1180 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1181 | struct cgroup *cont) | |
1182 | { | |
1183 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1184 | mem_cgroup_force_empty(mem); | |
1185 | } | |
1186 | ||
8cdea7c0 BS |
1187 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1188 | struct cgroup *cont) | |
1189 | { | |
6d12e2d8 KH |
1190 | int node; |
1191 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1192 | ||
1193 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1194 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1195 | |
8cdea7c0 BS |
1196 | kfree(mem_cgroup_from_cont(cont)); |
1197 | } | |
1198 | ||
1199 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1200 | struct cgroup *cont) | |
1201 | { | |
1202 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1203 | ARRAY_SIZE(mem_cgroup_files)); | |
1204 | } | |
1205 | ||
67e465a7 BS |
1206 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1207 | struct cgroup *cont, | |
1208 | struct cgroup *old_cont, | |
1209 | struct task_struct *p) | |
1210 | { | |
1211 | struct mm_struct *mm; | |
1212 | struct mem_cgroup *mem, *old_mem; | |
1213 | ||
1214 | mm = get_task_mm(p); | |
1215 | if (mm == NULL) | |
1216 | return; | |
1217 | ||
1218 | mem = mem_cgroup_from_cont(cont); | |
1219 | old_mem = mem_cgroup_from_cont(old_cont); | |
1220 | ||
1221 | if (mem == old_mem) | |
1222 | goto out; | |
1223 | ||
1224 | /* | |
1225 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1226 | * in effect owned by the leader | |
1227 | */ | |
1228 | if (p->tgid != p->pid) | |
1229 | goto out; | |
1230 | ||
1231 | css_get(&mem->css); | |
1232 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
1233 | css_put(&old_mem->css); | |
1234 | ||
1235 | out: | |
1236 | mmput(mm); | |
1237 | return; | |
1238 | } | |
1239 | ||
8cdea7c0 BS |
1240 | struct cgroup_subsys mem_cgroup_subsys = { |
1241 | .name = "memory", | |
1242 | .subsys_id = mem_cgroup_subsys_id, | |
1243 | .create = mem_cgroup_create, | |
df878fb0 | 1244 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1245 | .destroy = mem_cgroup_destroy, |
1246 | .populate = mem_cgroup_populate, | |
67e465a7 | 1247 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1248 | .early_init = 0, |
8cdea7c0 | 1249 | }; |