]>
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; | |
8a9f3ccd BS |
164 | atomic_t ref_cnt; /* Helpful when pages move b/w */ |
165 | /* mapped and cached states */ | |
8869b8f6 | 166 | int flags; |
8cdea7c0 | 167 | }; |
217bc319 | 168 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 169 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
8cdea7c0 | 170 | |
c0149530 KH |
171 | static inline int page_cgroup_nid(struct page_cgroup *pc) |
172 | { | |
173 | return page_to_nid(pc->page); | |
174 | } | |
175 | ||
176 | static inline enum zone_type page_cgroup_zid(struct page_cgroup *pc) | |
177 | { | |
178 | return page_zonenum(pc->page); | |
179 | } | |
180 | ||
217bc319 KH |
181 | enum charge_type { |
182 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
183 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
184 | }; | |
185 | ||
d52aa412 KH |
186 | /* |
187 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
188 | */ | |
189 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
190 | bool charge) | |
191 | { | |
192 | int val = (charge)? 1 : -1; | |
193 | struct mem_cgroup_stat *stat = &mem->stat; | |
d52aa412 | 194 | |
8869b8f6 | 195 | VM_BUG_ON(!irqs_disabled()); |
d52aa412 | 196 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
8869b8f6 | 197 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 KH |
198 | else |
199 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
6d12e2d8 KH |
200 | } |
201 | ||
202 | static inline struct mem_cgroup_per_zone * | |
203 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
204 | { | |
205 | BUG_ON(!mem->info.nodeinfo[nid]); | |
206 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
207 | } | |
208 | ||
209 | static inline struct mem_cgroup_per_zone * | |
210 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
211 | { | |
212 | struct mem_cgroup *mem = pc->mem_cgroup; | |
213 | int nid = page_cgroup_nid(pc); | |
214 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 215 | |
6d12e2d8 KH |
216 | return mem_cgroup_zoneinfo(mem, nid, zid); |
217 | } | |
218 | ||
219 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
220 | enum mem_cgroup_zstat_index idx) | |
221 | { | |
222 | int nid, zid; | |
223 | struct mem_cgroup_per_zone *mz; | |
224 | u64 total = 0; | |
225 | ||
226 | for_each_online_node(nid) | |
227 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
228 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
229 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
230 | } | |
231 | return total; | |
d52aa412 KH |
232 | } |
233 | ||
8cdea7c0 BS |
234 | static inline |
235 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
236 | { | |
237 | return container_of(cgroup_subsys_state(cont, | |
238 | mem_cgroup_subsys_id), struct mem_cgroup, | |
239 | css); | |
240 | } | |
241 | ||
78fb7466 PE |
242 | static inline |
243 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
244 | { | |
245 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
246 | struct mem_cgroup, css); | |
247 | } | |
248 | ||
249 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
250 | { | |
251 | struct mem_cgroup *mem; | |
252 | ||
253 | mem = mem_cgroup_from_task(p); | |
254 | css_get(&mem->css); | |
255 | mm->mem_cgroup = mem; | |
256 | } | |
257 | ||
258 | void mm_free_cgroup(struct mm_struct *mm) | |
259 | { | |
260 | css_put(&mm->mem_cgroup->css); | |
261 | } | |
262 | ||
8a9f3ccd BS |
263 | static inline int page_cgroup_locked(struct page *page) |
264 | { | |
8869b8f6 | 265 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
266 | } |
267 | ||
9442ec9d | 268 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 269 | { |
9442ec9d HD |
270 | VM_BUG_ON(!page_cgroup_locked(page)); |
271 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
272 | } |
273 | ||
274 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
275 | { | |
8869b8f6 | 276 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
277 | } |
278 | ||
8697d331 | 279 | static void __always_inline lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
280 | { |
281 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
282 | VM_BUG_ON(!page_cgroup_locked(page)); | |
283 | } | |
284 | ||
8697d331 | 285 | static void __always_inline unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
286 | { |
287 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
288 | } | |
289 | ||
9175e031 KH |
290 | /* |
291 | * Clear page->page_cgroup member under lock_page_cgroup(). | |
292 | * If given "pc" value is different from one page->page_cgroup, | |
293 | * page->cgroup is not cleared. | |
294 | * Returns a value of page->page_cgroup at lock taken. | |
295 | * A can can detect failure of clearing by following | |
296 | * clear_page_cgroup(page, pc) == pc | |
297 | */ | |
d52aa412 KH |
298 | static struct page_cgroup *clear_page_cgroup(struct page *page, |
299 | struct page_cgroup *pc) | |
9175e031 KH |
300 | { |
301 | struct page_cgroup *ret; | |
302 | /* lock and clear */ | |
303 | lock_page_cgroup(page); | |
304 | ret = page_get_page_cgroup(page); | |
305 | if (likely(ret == pc)) | |
306 | page_assign_page_cgroup(page, NULL); | |
307 | unlock_page_cgroup(page); | |
308 | return ret; | |
309 | } | |
310 | ||
6d12e2d8 KH |
311 | static void __mem_cgroup_remove_list(struct page_cgroup *pc) |
312 | { | |
313 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
314 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
315 | ||
316 | if (from) | |
317 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
318 | else | |
319 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
320 | ||
321 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
322 | list_del_init(&pc->lru); | |
323 | } | |
324 | ||
325 | static void __mem_cgroup_add_list(struct page_cgroup *pc) | |
326 | { | |
327 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
328 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
329 | ||
330 | if (!to) { | |
331 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 332 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
333 | } else { |
334 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 335 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
336 | } |
337 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
338 | } | |
339 | ||
8697d331 | 340 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 341 | { |
6d12e2d8 KH |
342 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
343 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
344 | ||
345 | if (from) | |
346 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
347 | else | |
348 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
349 | ||
3564c7c4 | 350 | if (active) { |
6d12e2d8 | 351 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 352 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 353 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 354 | } else { |
6d12e2d8 | 355 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 356 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 357 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 358 | } |
66e1707b BS |
359 | } |
360 | ||
4c4a2214 DR |
361 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
362 | { | |
363 | int ret; | |
364 | ||
365 | task_lock(task); | |
bd845e38 | 366 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
367 | task_unlock(task); |
368 | return ret; | |
369 | } | |
370 | ||
66e1707b BS |
371 | /* |
372 | * This routine assumes that the appropriate zone's lru lock is already held | |
373 | */ | |
427d5416 | 374 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 375 | { |
427d5416 | 376 | struct page_cgroup *pc; |
072c56c1 KH |
377 | struct mem_cgroup_per_zone *mz; |
378 | unsigned long flags; | |
379 | ||
427d5416 | 380 | pc = page_get_page_cgroup(page); |
66e1707b BS |
381 | if (!pc) |
382 | return; | |
383 | ||
072c56c1 KH |
384 | mz = page_cgroup_zoneinfo(pc); |
385 | spin_lock_irqsave(&mz->lru_lock, flags); | |
66e1707b | 386 | __mem_cgroup_move_lists(pc, active); |
072c56c1 | 387 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b BS |
388 | } |
389 | ||
58ae83db KH |
390 | /* |
391 | * Calculate mapped_ratio under memory controller. This will be used in | |
392 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
393 | */ | |
394 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
395 | { | |
396 | long total, rss; | |
397 | ||
398 | /* | |
399 | * usage is recorded in bytes. But, here, we assume the number of | |
400 | * physical pages can be represented by "long" on any arch. | |
401 | */ | |
402 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
403 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
404 | return (int)((rss * 100L) / total); | |
405 | } | |
8869b8f6 | 406 | |
5932f367 KH |
407 | /* |
408 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
409 | * between active and inactive list is calculated. For memory controller | |
410 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
411 | * zone's global lru imbalance. | |
412 | */ | |
413 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
414 | { | |
415 | unsigned long active, inactive; | |
416 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
417 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
418 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
419 | return (long) (active / (inactive + 1)); | |
420 | } | |
58ae83db | 421 | |
6c48a1d0 KH |
422 | /* |
423 | * prev_priority control...this will be used in memory reclaim path. | |
424 | */ | |
425 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
426 | { | |
427 | return mem->prev_priority; | |
428 | } | |
429 | ||
430 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
431 | { | |
432 | if (priority < mem->prev_priority) | |
433 | mem->prev_priority = priority; | |
434 | } | |
435 | ||
436 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
437 | { | |
438 | mem->prev_priority = priority; | |
439 | } | |
440 | ||
cc38108e KH |
441 | /* |
442 | * Calculate # of pages to be scanned in this priority/zone. | |
443 | * See also vmscan.c | |
444 | * | |
445 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
446 | * (see include/linux/mmzone.h) | |
447 | */ | |
448 | ||
449 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
450 | struct zone *zone, int priority) | |
451 | { | |
452 | long nr_active; | |
453 | int nid = zone->zone_pgdat->node_id; | |
454 | int zid = zone_idx(zone); | |
455 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
456 | ||
457 | nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
458 | return (nr_active >> priority); | |
459 | } | |
460 | ||
461 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
462 | struct zone *zone, int priority) | |
463 | { | |
464 | long nr_inactive; | |
465 | int nid = zone->zone_pgdat->node_id; | |
466 | int zid = zone_idx(zone); | |
467 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
468 | ||
469 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
cc38108e KH |
470 | return (nr_inactive >> priority); |
471 | } | |
472 | ||
66e1707b BS |
473 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
474 | struct list_head *dst, | |
475 | unsigned long *scanned, int order, | |
476 | int mode, struct zone *z, | |
477 | struct mem_cgroup *mem_cont, | |
478 | int active) | |
479 | { | |
480 | unsigned long nr_taken = 0; | |
481 | struct page *page; | |
482 | unsigned long scan; | |
483 | LIST_HEAD(pc_list); | |
484 | struct list_head *src; | |
ff7283fa | 485 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
486 | int nid = z->zone_pgdat->node_id; |
487 | int zid = zone_idx(z); | |
488 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 489 | |
1ecaab2b | 490 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 491 | if (active) |
1ecaab2b | 492 | src = &mz->active_list; |
66e1707b | 493 | else |
1ecaab2b KH |
494 | src = &mz->inactive_list; |
495 | ||
66e1707b | 496 | |
072c56c1 | 497 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
498 | scan = 0; |
499 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 500 | if (scan >= nr_to_scan) |
ff7283fa | 501 | break; |
66e1707b | 502 | page = pc->page; |
66e1707b | 503 | |
436c6541 | 504 | if (unlikely(!PageLRU(page))) |
ff7283fa | 505 | continue; |
ff7283fa | 506 | |
66e1707b BS |
507 | if (PageActive(page) && !active) { |
508 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
509 | continue; |
510 | } | |
511 | if (!PageActive(page) && active) { | |
512 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
513 | continue; |
514 | } | |
515 | ||
436c6541 HD |
516 | scan++; |
517 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
518 | |
519 | if (__isolate_lru_page(page, mode) == 0) { | |
520 | list_move(&page->lru, dst); | |
521 | nr_taken++; | |
522 | } | |
523 | } | |
524 | ||
525 | list_splice(&pc_list, src); | |
072c56c1 | 526 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
527 | |
528 | *scanned = scan; | |
529 | return nr_taken; | |
530 | } | |
531 | ||
8a9f3ccd BS |
532 | /* |
533 | * Charge the memory controller for page usage. | |
534 | * Return | |
535 | * 0 if the charge was successful | |
536 | * < 0 if the cgroup is over its limit | |
537 | */ | |
217bc319 KH |
538 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
539 | gfp_t gfp_mask, enum charge_type ctype) | |
8a9f3ccd BS |
540 | { |
541 | struct mem_cgroup *mem; | |
9175e031 | 542 | struct page_cgroup *pc; |
66e1707b BS |
543 | unsigned long flags; |
544 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 545 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd BS |
546 | |
547 | /* | |
548 | * Should page_cgroup's go to their own slab? | |
549 | * One could optimize the performance of the charging routine | |
550 | * by saving a bit in the page_flags and using it as a lock | |
551 | * to see if the cgroup page already has a page_cgroup associated | |
552 | * with it | |
553 | */ | |
66e1707b | 554 | retry: |
7e924aaf HD |
555 | lock_page_cgroup(page); |
556 | pc = page_get_page_cgroup(page); | |
557 | /* | |
558 | * The page_cgroup exists and | |
559 | * the page has already been accounted. | |
560 | */ | |
561 | if (pc) { | |
562 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { | |
563 | /* this page is under being uncharged ? */ | |
564 | unlock_page_cgroup(page); | |
565 | cpu_relax(); | |
566 | goto retry; | |
567 | } else { | |
568 | unlock_page_cgroup(page); | |
569 | goto done; | |
9175e031 | 570 | } |
8a9f3ccd | 571 | } |
7e924aaf | 572 | unlock_page_cgroup(page); |
8a9f3ccd | 573 | |
e1a1cd59 | 574 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
575 | if (pc == NULL) |
576 | goto err; | |
577 | ||
8a9f3ccd | 578 | /* |
3be91277 HD |
579 | * We always charge the cgroup the mm_struct belongs to. |
580 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
581 | * thread group leader migrates. It's possible that mm is not |
582 | * set, if so charge the init_mm (happens for pagecache usage). | |
583 | */ | |
584 | if (!mm) | |
585 | mm = &init_mm; | |
586 | ||
3be91277 | 587 | rcu_read_lock(); |
8a9f3ccd BS |
588 | mem = rcu_dereference(mm->mem_cgroup); |
589 | /* | |
8869b8f6 | 590 | * For every charge from the cgroup, increment reference count |
8a9f3ccd BS |
591 | */ |
592 | css_get(&mem->css); | |
593 | rcu_read_unlock(); | |
594 | ||
0eea1030 | 595 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
596 | if (!(gfp_mask & __GFP_WAIT)) |
597 | goto out; | |
e1a1cd59 BS |
598 | |
599 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
600 | continue; |
601 | ||
602 | /* | |
8869b8f6 HD |
603 | * try_to_free_mem_cgroup_pages() might not give us a full |
604 | * picture of reclaim. Some pages are reclaimed and might be | |
605 | * moved to swap cache or just unmapped from the cgroup. | |
606 | * Check the limit again to see if the reclaim reduced the | |
607 | * current usage of the cgroup before giving up | |
608 | */ | |
66e1707b BS |
609 | if (res_counter_check_under_limit(&mem->res)) |
610 | continue; | |
3be91277 HD |
611 | |
612 | if (!nr_retries--) { | |
613 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
614 | goto out; | |
66e1707b | 615 | } |
3be91277 | 616 | congestion_wait(WRITE, HZ/10); |
8a9f3ccd BS |
617 | } |
618 | ||
8a9f3ccd BS |
619 | atomic_set(&pc->ref_cnt, 1); |
620 | pc->mem_cgroup = mem; | |
621 | pc->page = page; | |
3564c7c4 | 622 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
217bc319 KH |
623 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
624 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
3be91277 | 625 | |
7e924aaf HD |
626 | lock_page_cgroup(page); |
627 | if (page_get_page_cgroup(page)) { | |
628 | unlock_page_cgroup(page); | |
9175e031 | 629 | /* |
3be91277 HD |
630 | * Another charge has been added to this page already. |
631 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
632 | * page->cgroup, increment refcnt.... just retry is OK. |
633 | */ | |
634 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
635 | css_put(&mem->css); | |
636 | kfree(pc); | |
637 | goto retry; | |
638 | } | |
7e924aaf HD |
639 | page_assign_page_cgroup(page, pc); |
640 | unlock_page_cgroup(page); | |
8a9f3ccd | 641 | |
072c56c1 KH |
642 | mz = page_cgroup_zoneinfo(pc); |
643 | spin_lock_irqsave(&mz->lru_lock, flags); | |
6d12e2d8 | 644 | __mem_cgroup_add_list(pc); |
072c56c1 | 645 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 646 | |
8a9f3ccd | 647 | done: |
8a9f3ccd | 648 | return 0; |
3be91277 HD |
649 | out: |
650 | css_put(&mem->css); | |
8a9f3ccd | 651 | kfree(pc); |
8a9f3ccd | 652 | err: |
8a9f3ccd BS |
653 | return -ENOMEM; |
654 | } | |
655 | ||
8869b8f6 | 656 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 KH |
657 | { |
658 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
8869b8f6 | 659 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
660 | } |
661 | ||
e1a1cd59 BS |
662 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
663 | gfp_t gfp_mask) | |
8697d331 | 664 | { |
8697d331 BS |
665 | if (!mm) |
666 | mm = &init_mm; | |
8869b8f6 | 667 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 668 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
8697d331 BS |
669 | } |
670 | ||
8a9f3ccd BS |
671 | /* |
672 | * Uncharging is always a welcome operation, we never complain, simply | |
8289546e | 673 | * uncharge. |
8a9f3ccd | 674 | */ |
8289546e | 675 | void mem_cgroup_uncharge_page(struct page *page) |
8a9f3ccd | 676 | { |
8289546e | 677 | struct page_cgroup *pc; |
8a9f3ccd | 678 | struct mem_cgroup *mem; |
072c56c1 | 679 | struct mem_cgroup_per_zone *mz; |
66e1707b | 680 | unsigned long flags; |
8a9f3ccd | 681 | |
8697d331 | 682 | /* |
3c541e14 | 683 | * Check if our page_cgroup is valid |
8697d331 | 684 | */ |
8289546e HD |
685 | lock_page_cgroup(page); |
686 | pc = page_get_page_cgroup(page); | |
8a9f3ccd | 687 | if (!pc) |
8289546e | 688 | goto unlock; |
8a9f3ccd BS |
689 | |
690 | if (atomic_dec_and_test(&pc->ref_cnt)) { | |
691 | page = pc->page; | |
072c56c1 | 692 | mz = page_cgroup_zoneinfo(pc); |
9175e031 KH |
693 | /* |
694 | * get page->cgroup and clear it under lock. | |
cc847582 | 695 | * force_empty can drop page->cgroup without checking refcnt. |
9175e031 | 696 | */ |
3c541e14 | 697 | unlock_page_cgroup(page); |
9175e031 KH |
698 | if (clear_page_cgroup(page, pc) == pc) { |
699 | mem = pc->mem_cgroup; | |
700 | css_put(&mem->css); | |
701 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
072c56c1 | 702 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 703 | __mem_cgroup_remove_list(pc); |
072c56c1 | 704 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9175e031 | 705 | kfree(pc); |
9175e031 | 706 | } |
3c541e14 | 707 | lock_page_cgroup(page); |
8a9f3ccd | 708 | } |
6d12e2d8 | 709 | |
8289546e | 710 | unlock: |
3c541e14 BS |
711 | unlock_page_cgroup(page); |
712 | } | |
713 | ||
ae41be37 KH |
714 | /* |
715 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
716 | * Refcnt of page_cgroup is incremented. | |
717 | */ | |
ae41be37 KH |
718 | int mem_cgroup_prepare_migration(struct page *page) |
719 | { | |
720 | struct page_cgroup *pc; | |
721 | int ret = 0; | |
8869b8f6 | 722 | |
ae41be37 KH |
723 | lock_page_cgroup(page); |
724 | pc = page_get_page_cgroup(page); | |
725 | if (pc && atomic_inc_not_zero(&pc->ref_cnt)) | |
726 | ret = 1; | |
727 | unlock_page_cgroup(page); | |
728 | return ret; | |
729 | } | |
730 | ||
731 | void mem_cgroup_end_migration(struct page *page) | |
732 | { | |
8289546e | 733 | mem_cgroup_uncharge_page(page); |
ae41be37 | 734 | } |
8869b8f6 | 735 | |
ae41be37 | 736 | /* |
8869b8f6 | 737 | * We know both *page* and *newpage* are now not-on-LRU and PG_locked. |
ae41be37 KH |
738 | * And no race with uncharge() routines because page_cgroup for *page* |
739 | * has extra one reference by mem_cgroup_prepare_migration. | |
740 | */ | |
ae41be37 KH |
741 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) |
742 | { | |
743 | struct page_cgroup *pc; | |
6d12e2d8 KH |
744 | struct mem_cgroup *mem; |
745 | unsigned long flags; | |
072c56c1 | 746 | struct mem_cgroup_per_zone *mz; |
8869b8f6 | 747 | |
ae41be37 KH |
748 | retry: |
749 | pc = page_get_page_cgroup(page); | |
750 | if (!pc) | |
751 | return; | |
8869b8f6 | 752 | |
6d12e2d8 | 753 | mem = pc->mem_cgroup; |
072c56c1 | 754 | mz = page_cgroup_zoneinfo(pc); |
ae41be37 KH |
755 | if (clear_page_cgroup(page, pc) != pc) |
756 | goto retry; | |
6d12e2d8 | 757 | |
8869b8f6 | 758 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 759 | __mem_cgroup_remove_list(pc); |
072c56c1 KH |
760 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
761 | ||
ae41be37 KH |
762 | pc->page = newpage; |
763 | lock_page_cgroup(newpage); | |
764 | page_assign_page_cgroup(newpage, pc); | |
765 | unlock_page_cgroup(newpage); | |
6d12e2d8 | 766 | |
072c56c1 KH |
767 | mz = page_cgroup_zoneinfo(pc); |
768 | spin_lock_irqsave(&mz->lru_lock, flags); | |
769 | __mem_cgroup_add_list(pc); | |
770 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
ae41be37 | 771 | } |
78fb7466 | 772 | |
cc847582 KH |
773 | /* |
774 | * This routine traverse page_cgroup in given list and drop them all. | |
775 | * This routine ignores page_cgroup->ref_cnt. | |
776 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
777 | */ | |
778 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 779 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 KH |
780 | struct mem_cgroup_per_zone *mz, |
781 | int active) | |
cc847582 KH |
782 | { |
783 | struct page_cgroup *pc; | |
784 | struct page *page; | |
785 | int count; | |
786 | unsigned long flags; | |
072c56c1 KH |
787 | struct list_head *list; |
788 | ||
789 | if (active) | |
790 | list = &mz->active_list; | |
791 | else | |
792 | list = &mz->inactive_list; | |
cc847582 | 793 | |
1ecaab2b KH |
794 | if (list_empty(list)) |
795 | return; | |
cc847582 KH |
796 | retry: |
797 | count = FORCE_UNCHARGE_BATCH; | |
072c56c1 | 798 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 KH |
799 | |
800 | while (--count && !list_empty(list)) { | |
801 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
802 | page = pc->page; | |
803 | /* Avoid race with charge */ | |
804 | atomic_set(&pc->ref_cnt, 0); | |
805 | if (clear_page_cgroup(page, pc) == pc) { | |
806 | css_put(&mem->css); | |
807 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
6d12e2d8 | 808 | __mem_cgroup_remove_list(pc); |
cc847582 KH |
809 | kfree(pc); |
810 | } else /* being uncharged ? ...do relax */ | |
811 | break; | |
812 | } | |
8869b8f6 | 813 | |
072c56c1 | 814 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
815 | if (!list_empty(list)) { |
816 | cond_resched(); | |
817 | goto retry; | |
818 | } | |
cc847582 KH |
819 | } |
820 | ||
821 | /* | |
822 | * make mem_cgroup's charge to be 0 if there is no task. | |
823 | * This enables deleting this mem_cgroup. | |
824 | */ | |
cc847582 KH |
825 | int mem_cgroup_force_empty(struct mem_cgroup *mem) |
826 | { | |
827 | int ret = -EBUSY; | |
1ecaab2b | 828 | int node, zid; |
8869b8f6 | 829 | |
cc847582 KH |
830 | css_get(&mem->css); |
831 | /* | |
832 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 833 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
834 | * So, we have to do loop here until all lists are empty. |
835 | */ | |
1ecaab2b | 836 | while (mem->res.usage > 0) { |
cc847582 KH |
837 | if (atomic_read(&mem->css.cgroup->count) > 0) |
838 | goto out; | |
1ecaab2b KH |
839 | for_each_node_state(node, N_POSSIBLE) |
840 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
841 | struct mem_cgroup_per_zone *mz; | |
842 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
843 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 844 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 845 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 846 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 847 | } |
cc847582 KH |
848 | } |
849 | ret = 0; | |
850 | out: | |
851 | css_put(&mem->css); | |
852 | return ret; | |
853 | } | |
854 | ||
0eea1030 BS |
855 | int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
856 | { | |
857 | *tmp = memparse(buf, &buf); | |
858 | if (*buf != '\0') | |
859 | return -EINVAL; | |
860 | ||
861 | /* | |
862 | * Round up the value to the closest page size | |
863 | */ | |
864 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
865 | return 0; | |
866 | } | |
867 | ||
868 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
869 | struct cftype *cft, struct file *file, | |
870 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
871 | { |
872 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
873 | cft->private, userbuf, nbytes, ppos, |
874 | NULL); | |
8cdea7c0 BS |
875 | } |
876 | ||
877 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
878 | struct file *file, const char __user *userbuf, | |
879 | size_t nbytes, loff_t *ppos) | |
880 | { | |
881 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
882 | cft->private, userbuf, nbytes, ppos, |
883 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
884 | } |
885 | ||
cc847582 KH |
886 | static ssize_t mem_force_empty_write(struct cgroup *cont, |
887 | struct cftype *cft, struct file *file, | |
888 | const char __user *userbuf, | |
889 | size_t nbytes, loff_t *ppos) | |
890 | { | |
891 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
8869b8f6 | 892 | int ret = mem_cgroup_force_empty(mem); |
cc847582 KH |
893 | if (!ret) |
894 | ret = nbytes; | |
895 | return ret; | |
896 | } | |
897 | ||
898 | /* | |
899 | * Note: This should be removed if cgroup supports write-only file. | |
900 | */ | |
cc847582 KH |
901 | static ssize_t mem_force_empty_read(struct cgroup *cont, |
902 | struct cftype *cft, | |
903 | struct file *file, char __user *userbuf, | |
904 | size_t nbytes, loff_t *ppos) | |
905 | { | |
906 | return -EINVAL; | |
907 | } | |
908 | ||
d2ceb9b7 KH |
909 | static const struct mem_cgroup_stat_desc { |
910 | const char *msg; | |
911 | u64 unit; | |
912 | } mem_cgroup_stat_desc[] = { | |
913 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
914 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
915 | }; | |
916 | ||
917 | static int mem_control_stat_show(struct seq_file *m, void *arg) | |
918 | { | |
919 | struct cgroup *cont = m->private; | |
920 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
921 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
922 | int i; | |
923 | ||
924 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
925 | s64 val; | |
926 | ||
927 | val = mem_cgroup_read_stat(stat, i); | |
928 | val *= mem_cgroup_stat_desc[i].unit; | |
929 | seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg, | |
930 | (long long)val); | |
931 | } | |
6d12e2d8 KH |
932 | /* showing # of active pages */ |
933 | { | |
934 | unsigned long active, inactive; | |
935 | ||
936 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
937 | MEM_CGROUP_ZSTAT_INACTIVE); | |
938 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
939 | MEM_CGROUP_ZSTAT_ACTIVE); | |
940 | seq_printf(m, "active %ld\n", (active) * PAGE_SIZE); | |
941 | seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE); | |
942 | } | |
d2ceb9b7 KH |
943 | return 0; |
944 | } | |
945 | ||
946 | static const struct file_operations mem_control_stat_file_operations = { | |
947 | .read = seq_read, | |
948 | .llseek = seq_lseek, | |
949 | .release = single_release, | |
950 | }; | |
951 | ||
952 | static int mem_control_stat_open(struct inode *unused, struct file *file) | |
953 | { | |
954 | /* XXX __d_cont */ | |
955 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
956 | ||
957 | file->f_op = &mem_control_stat_file_operations; | |
958 | return single_open(file, mem_control_stat_show, cont); | |
959 | } | |
960 | ||
8cdea7c0 BS |
961 | static struct cftype mem_cgroup_files[] = { |
962 | { | |
0eea1030 | 963 | .name = "usage_in_bytes", |
8cdea7c0 BS |
964 | .private = RES_USAGE, |
965 | .read = mem_cgroup_read, | |
966 | }, | |
967 | { | |
0eea1030 | 968 | .name = "limit_in_bytes", |
8cdea7c0 BS |
969 | .private = RES_LIMIT, |
970 | .write = mem_cgroup_write, | |
971 | .read = mem_cgroup_read, | |
972 | }, | |
973 | { | |
974 | .name = "failcnt", | |
975 | .private = RES_FAILCNT, | |
976 | .read = mem_cgroup_read, | |
977 | }, | |
cc847582 KH |
978 | { |
979 | .name = "force_empty", | |
980 | .write = mem_force_empty_write, | |
981 | .read = mem_force_empty_read, | |
982 | }, | |
d2ceb9b7 KH |
983 | { |
984 | .name = "stat", | |
985 | .open = mem_control_stat_open, | |
986 | }, | |
8cdea7c0 BS |
987 | }; |
988 | ||
6d12e2d8 KH |
989 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
990 | { | |
991 | struct mem_cgroup_per_node *pn; | |
1ecaab2b KH |
992 | struct mem_cgroup_per_zone *mz; |
993 | int zone; | |
994 | /* | |
995 | * This routine is called against possible nodes. | |
996 | * But it's BUG to call kmalloc() against offline node. | |
997 | * | |
998 | * TODO: this routine can waste much memory for nodes which will | |
999 | * never be onlined. It's better to use memory hotplug callback | |
1000 | * function. | |
1001 | */ | |
1002 | if (node_state(node, N_HIGH_MEMORY)) | |
1003 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
1004 | else | |
1005 | pn = kmalloc(sizeof(*pn), GFP_KERNEL); | |
6d12e2d8 KH |
1006 | if (!pn) |
1007 | return 1; | |
1ecaab2b | 1008 | |
6d12e2d8 KH |
1009 | mem->info.nodeinfo[node] = pn; |
1010 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1011 | |
1012 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1013 | mz = &pn->zoneinfo[zone]; | |
1014 | INIT_LIST_HEAD(&mz->active_list); | |
1015 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 1016 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 1017 | } |
6d12e2d8 KH |
1018 | return 0; |
1019 | } | |
1020 | ||
1ecaab2b KH |
1021 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1022 | { | |
1023 | kfree(mem->info.nodeinfo[node]); | |
1024 | } | |
1025 | ||
8cdea7c0 BS |
1026 | static struct cgroup_subsys_state * |
1027 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1028 | { | |
1029 | struct mem_cgroup *mem; | |
6d12e2d8 | 1030 | int node; |
8cdea7c0 | 1031 | |
78fb7466 PE |
1032 | if (unlikely((cont->parent) == NULL)) { |
1033 | mem = &init_mem_cgroup; | |
1034 | init_mm.mem_cgroup = mem; | |
1035 | } else | |
1036 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
1037 | ||
1038 | if (mem == NULL) | |
2dda81ca | 1039 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1040 | |
1041 | res_counter_init(&mem->res); | |
1ecaab2b | 1042 | |
6d12e2d8 KH |
1043 | memset(&mem->info, 0, sizeof(mem->info)); |
1044 | ||
1045 | for_each_node_state(node, N_POSSIBLE) | |
1046 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1047 | goto free_out; | |
1048 | ||
8cdea7c0 | 1049 | return &mem->css; |
6d12e2d8 KH |
1050 | free_out: |
1051 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1052 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 KH |
1053 | if (cont->parent != NULL) |
1054 | kfree(mem); | |
2dda81ca | 1055 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1056 | } |
1057 | ||
df878fb0 KH |
1058 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1059 | struct cgroup *cont) | |
1060 | { | |
1061 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1062 | mem_cgroup_force_empty(mem); | |
1063 | } | |
1064 | ||
8cdea7c0 BS |
1065 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1066 | struct cgroup *cont) | |
1067 | { | |
6d12e2d8 KH |
1068 | int node; |
1069 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1070 | ||
1071 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1072 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1073 | |
8cdea7c0 BS |
1074 | kfree(mem_cgroup_from_cont(cont)); |
1075 | } | |
1076 | ||
1077 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1078 | struct cgroup *cont) | |
1079 | { | |
1080 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1081 | ARRAY_SIZE(mem_cgroup_files)); | |
1082 | } | |
1083 | ||
67e465a7 BS |
1084 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1085 | struct cgroup *cont, | |
1086 | struct cgroup *old_cont, | |
1087 | struct task_struct *p) | |
1088 | { | |
1089 | struct mm_struct *mm; | |
1090 | struct mem_cgroup *mem, *old_mem; | |
1091 | ||
1092 | mm = get_task_mm(p); | |
1093 | if (mm == NULL) | |
1094 | return; | |
1095 | ||
1096 | mem = mem_cgroup_from_cont(cont); | |
1097 | old_mem = mem_cgroup_from_cont(old_cont); | |
1098 | ||
1099 | if (mem == old_mem) | |
1100 | goto out; | |
1101 | ||
1102 | /* | |
1103 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1104 | * in effect owned by the leader | |
1105 | */ | |
1106 | if (p->tgid != p->pid) | |
1107 | goto out; | |
1108 | ||
1109 | css_get(&mem->css); | |
1110 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
1111 | css_put(&old_mem->css); | |
1112 | ||
1113 | out: | |
1114 | mmput(mm); | |
67e465a7 BS |
1115 | } |
1116 | ||
8cdea7c0 BS |
1117 | struct cgroup_subsys mem_cgroup_subsys = { |
1118 | .name = "memory", | |
1119 | .subsys_id = mem_cgroup_subsys_id, | |
1120 | .create = mem_cgroup_create, | |
df878fb0 | 1121 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1122 | .destroy = mem_cgroup_destroy, |
1123 | .populate = mem_cgroup_populate, | |
67e465a7 | 1124 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1125 | .early_init = 0, |
8cdea7c0 | 1126 | }; |