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 | |
d5b69e38 | 171 | static int page_cgroup_nid(struct page_cgroup *pc) |
c0149530 KH |
172 | { |
173 | return page_to_nid(pc->page); | |
174 | } | |
175 | ||
d5b69e38 | 176 | static enum zone_type page_cgroup_zid(struct page_cgroup *pc) |
c0149530 KH |
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 | ||
d5b69e38 | 202 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
203 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
204 | { | |
6d12e2d8 KH |
205 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
206 | } | |
207 | ||
d5b69e38 | 208 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
209 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
210 | { | |
211 | struct mem_cgroup *mem = pc->mem_cgroup; | |
212 | int nid = page_cgroup_nid(pc); | |
213 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 214 | |
6d12e2d8 KH |
215 | return mem_cgroup_zoneinfo(mem, nid, zid); |
216 | } | |
217 | ||
218 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
219 | enum mem_cgroup_zstat_index idx) | |
220 | { | |
221 | int nid, zid; | |
222 | struct mem_cgroup_per_zone *mz; | |
223 | u64 total = 0; | |
224 | ||
225 | for_each_online_node(nid) | |
226 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
227 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
228 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
229 | } | |
230 | return total; | |
d52aa412 KH |
231 | } |
232 | ||
d5b69e38 | 233 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
234 | { |
235 | return container_of(cgroup_subsys_state(cont, | |
236 | mem_cgroup_subsys_id), struct mem_cgroup, | |
237 | css); | |
238 | } | |
239 | ||
d5b69e38 | 240 | static struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 PE |
241 | { |
242 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
243 | struct mem_cgroup, css); | |
244 | } | |
245 | ||
246 | void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p) | |
247 | { | |
248 | struct mem_cgroup *mem; | |
249 | ||
250 | mem = mem_cgroup_from_task(p); | |
251 | css_get(&mem->css); | |
252 | mm->mem_cgroup = mem; | |
253 | } | |
254 | ||
255 | void mm_free_cgroup(struct mm_struct *mm) | |
256 | { | |
257 | css_put(&mm->mem_cgroup->css); | |
258 | } | |
259 | ||
8a9f3ccd BS |
260 | static inline int page_cgroup_locked(struct page *page) |
261 | { | |
8869b8f6 | 262 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
263 | } |
264 | ||
9442ec9d | 265 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 266 | { |
9442ec9d HD |
267 | VM_BUG_ON(!page_cgroup_locked(page)); |
268 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
269 | } |
270 | ||
271 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
272 | { | |
8869b8f6 | 273 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
274 | } |
275 | ||
d5b69e38 | 276 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
277 | { |
278 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
279 | } |
280 | ||
d5b69e38 | 281 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
282 | { |
283 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
284 | } | |
285 | ||
9175e031 KH |
286 | /* |
287 | * Clear page->page_cgroup member under lock_page_cgroup(). | |
288 | * If given "pc" value is different from one page->page_cgroup, | |
289 | * page->cgroup is not cleared. | |
290 | * Returns a value of page->page_cgroup at lock taken. | |
291 | * A can can detect failure of clearing by following | |
292 | * clear_page_cgroup(page, pc) == pc | |
293 | */ | |
d52aa412 KH |
294 | static struct page_cgroup *clear_page_cgroup(struct page *page, |
295 | struct page_cgroup *pc) | |
9175e031 KH |
296 | { |
297 | struct page_cgroup *ret; | |
298 | /* lock and clear */ | |
299 | lock_page_cgroup(page); | |
300 | ret = page_get_page_cgroup(page); | |
301 | if (likely(ret == pc)) | |
302 | page_assign_page_cgroup(page, NULL); | |
303 | unlock_page_cgroup(page); | |
304 | return ret; | |
305 | } | |
306 | ||
6d12e2d8 KH |
307 | static void __mem_cgroup_remove_list(struct page_cgroup *pc) |
308 | { | |
309 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
310 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
311 | ||
312 | if (from) | |
313 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
314 | else | |
315 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
316 | ||
317 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
318 | list_del_init(&pc->lru); | |
319 | } | |
320 | ||
321 | static void __mem_cgroup_add_list(struct page_cgroup *pc) | |
322 | { | |
323 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
324 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
325 | ||
326 | if (!to) { | |
327 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 328 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
329 | } else { |
330 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 331 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
332 | } |
333 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
334 | } | |
335 | ||
8697d331 | 336 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 337 | { |
6d12e2d8 KH |
338 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
339 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
340 | ||
341 | if (from) | |
342 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
343 | else | |
344 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
345 | ||
3564c7c4 | 346 | if (active) { |
6d12e2d8 | 347 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 348 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 349 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 350 | } else { |
6d12e2d8 | 351 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 352 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 353 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 354 | } |
66e1707b BS |
355 | } |
356 | ||
4c4a2214 DR |
357 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
358 | { | |
359 | int ret; | |
360 | ||
361 | task_lock(task); | |
bd845e38 | 362 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
363 | task_unlock(task); |
364 | return ret; | |
365 | } | |
366 | ||
66e1707b BS |
367 | /* |
368 | * This routine assumes that the appropriate zone's lru lock is already held | |
369 | */ | |
427d5416 | 370 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 371 | { |
427d5416 | 372 | struct page_cgroup *pc; |
072c56c1 KH |
373 | struct mem_cgroup_per_zone *mz; |
374 | unsigned long flags; | |
375 | ||
427d5416 | 376 | pc = page_get_page_cgroup(page); |
66e1707b BS |
377 | if (!pc) |
378 | return; | |
379 | ||
072c56c1 KH |
380 | mz = page_cgroup_zoneinfo(pc); |
381 | spin_lock_irqsave(&mz->lru_lock, flags); | |
66e1707b | 382 | __mem_cgroup_move_lists(pc, active); |
072c56c1 | 383 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b BS |
384 | } |
385 | ||
58ae83db KH |
386 | /* |
387 | * Calculate mapped_ratio under memory controller. This will be used in | |
388 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
389 | */ | |
390 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
391 | { | |
392 | long total, rss; | |
393 | ||
394 | /* | |
395 | * usage is recorded in bytes. But, here, we assume the number of | |
396 | * physical pages can be represented by "long" on any arch. | |
397 | */ | |
398 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
399 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
400 | return (int)((rss * 100L) / total); | |
401 | } | |
8869b8f6 | 402 | |
5932f367 KH |
403 | /* |
404 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
405 | * between active and inactive list is calculated. For memory controller | |
406 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
407 | * zone's global lru imbalance. | |
408 | */ | |
409 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
410 | { | |
411 | unsigned long active, inactive; | |
412 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
413 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
414 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
415 | return (long) (active / (inactive + 1)); | |
416 | } | |
58ae83db | 417 | |
6c48a1d0 KH |
418 | /* |
419 | * prev_priority control...this will be used in memory reclaim path. | |
420 | */ | |
421 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
422 | { | |
423 | return mem->prev_priority; | |
424 | } | |
425 | ||
426 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
427 | { | |
428 | if (priority < mem->prev_priority) | |
429 | mem->prev_priority = priority; | |
430 | } | |
431 | ||
432 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
433 | { | |
434 | mem->prev_priority = priority; | |
435 | } | |
436 | ||
cc38108e KH |
437 | /* |
438 | * Calculate # of pages to be scanned in this priority/zone. | |
439 | * See also vmscan.c | |
440 | * | |
441 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
442 | * (see include/linux/mmzone.h) | |
443 | */ | |
444 | ||
445 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
446 | struct zone *zone, int priority) | |
447 | { | |
448 | long nr_active; | |
449 | int nid = zone->zone_pgdat->node_id; | |
450 | int zid = zone_idx(zone); | |
451 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
452 | ||
453 | nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
454 | return (nr_active >> priority); | |
455 | } | |
456 | ||
457 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
458 | struct zone *zone, int priority) | |
459 | { | |
460 | long nr_inactive; | |
461 | int nid = zone->zone_pgdat->node_id; | |
462 | int zid = zone_idx(zone); | |
463 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
464 | ||
465 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
cc38108e KH |
466 | return (nr_inactive >> priority); |
467 | } | |
468 | ||
66e1707b BS |
469 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
470 | struct list_head *dst, | |
471 | unsigned long *scanned, int order, | |
472 | int mode, struct zone *z, | |
473 | struct mem_cgroup *mem_cont, | |
474 | int active) | |
475 | { | |
476 | unsigned long nr_taken = 0; | |
477 | struct page *page; | |
478 | unsigned long scan; | |
479 | LIST_HEAD(pc_list); | |
480 | struct list_head *src; | |
ff7283fa | 481 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
482 | int nid = z->zone_pgdat->node_id; |
483 | int zid = zone_idx(z); | |
484 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 485 | |
1ecaab2b | 486 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 487 | if (active) |
1ecaab2b | 488 | src = &mz->active_list; |
66e1707b | 489 | else |
1ecaab2b KH |
490 | src = &mz->inactive_list; |
491 | ||
66e1707b | 492 | |
072c56c1 | 493 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
494 | scan = 0; |
495 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 496 | if (scan >= nr_to_scan) |
ff7283fa | 497 | break; |
66e1707b | 498 | page = pc->page; |
66e1707b | 499 | |
436c6541 | 500 | if (unlikely(!PageLRU(page))) |
ff7283fa | 501 | continue; |
ff7283fa | 502 | |
66e1707b BS |
503 | if (PageActive(page) && !active) { |
504 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
505 | continue; |
506 | } | |
507 | if (!PageActive(page) && active) { | |
508 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
509 | continue; |
510 | } | |
511 | ||
436c6541 HD |
512 | scan++; |
513 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
514 | |
515 | if (__isolate_lru_page(page, mode) == 0) { | |
516 | list_move(&page->lru, dst); | |
517 | nr_taken++; | |
518 | } | |
519 | } | |
520 | ||
521 | list_splice(&pc_list, src); | |
072c56c1 | 522 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
523 | |
524 | *scanned = scan; | |
525 | return nr_taken; | |
526 | } | |
527 | ||
8a9f3ccd BS |
528 | /* |
529 | * Charge the memory controller for page usage. | |
530 | * Return | |
531 | * 0 if the charge was successful | |
532 | * < 0 if the cgroup is over its limit | |
533 | */ | |
217bc319 KH |
534 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
535 | gfp_t gfp_mask, enum charge_type ctype) | |
8a9f3ccd BS |
536 | { |
537 | struct mem_cgroup *mem; | |
9175e031 | 538 | struct page_cgroup *pc; |
66e1707b BS |
539 | unsigned long flags; |
540 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 541 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd BS |
542 | |
543 | /* | |
544 | * Should page_cgroup's go to their own slab? | |
545 | * One could optimize the performance of the charging routine | |
546 | * by saving a bit in the page_flags and using it as a lock | |
547 | * to see if the cgroup page already has a page_cgroup associated | |
548 | * with it | |
549 | */ | |
66e1707b | 550 | retry: |
7e924aaf HD |
551 | lock_page_cgroup(page); |
552 | pc = page_get_page_cgroup(page); | |
553 | /* | |
554 | * The page_cgroup exists and | |
555 | * the page has already been accounted. | |
556 | */ | |
557 | if (pc) { | |
558 | if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) { | |
559 | /* this page is under being uncharged ? */ | |
560 | unlock_page_cgroup(page); | |
561 | cpu_relax(); | |
562 | goto retry; | |
563 | } else { | |
564 | unlock_page_cgroup(page); | |
565 | goto done; | |
9175e031 | 566 | } |
8a9f3ccd | 567 | } |
7e924aaf | 568 | unlock_page_cgroup(page); |
8a9f3ccd | 569 | |
e1a1cd59 | 570 | pc = kzalloc(sizeof(struct page_cgroup), gfp_mask); |
8a9f3ccd BS |
571 | if (pc == NULL) |
572 | goto err; | |
573 | ||
8a9f3ccd | 574 | /* |
3be91277 HD |
575 | * We always charge the cgroup the mm_struct belongs to. |
576 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
577 | * thread group leader migrates. It's possible that mm is not |
578 | * set, if so charge the init_mm (happens for pagecache usage). | |
579 | */ | |
580 | if (!mm) | |
581 | mm = &init_mm; | |
582 | ||
3be91277 | 583 | rcu_read_lock(); |
8a9f3ccd BS |
584 | mem = rcu_dereference(mm->mem_cgroup); |
585 | /* | |
8869b8f6 | 586 | * For every charge from the cgroup, increment reference count |
8a9f3ccd BS |
587 | */ |
588 | css_get(&mem->css); | |
589 | rcu_read_unlock(); | |
590 | ||
0eea1030 | 591 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
592 | if (!(gfp_mask & __GFP_WAIT)) |
593 | goto out; | |
e1a1cd59 BS |
594 | |
595 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
596 | continue; |
597 | ||
598 | /* | |
8869b8f6 HD |
599 | * try_to_free_mem_cgroup_pages() might not give us a full |
600 | * picture of reclaim. Some pages are reclaimed and might be | |
601 | * moved to swap cache or just unmapped from the cgroup. | |
602 | * Check the limit again to see if the reclaim reduced the | |
603 | * current usage of the cgroup before giving up | |
604 | */ | |
66e1707b BS |
605 | if (res_counter_check_under_limit(&mem->res)) |
606 | continue; | |
3be91277 HD |
607 | |
608 | if (!nr_retries--) { | |
609 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
610 | goto out; | |
66e1707b | 611 | } |
3be91277 | 612 | congestion_wait(WRITE, HZ/10); |
8a9f3ccd BS |
613 | } |
614 | ||
8a9f3ccd BS |
615 | atomic_set(&pc->ref_cnt, 1); |
616 | pc->mem_cgroup = mem; | |
617 | pc->page = page; | |
3564c7c4 | 618 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
217bc319 KH |
619 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
620 | pc->flags |= PAGE_CGROUP_FLAG_CACHE; | |
3be91277 | 621 | |
7e924aaf HD |
622 | lock_page_cgroup(page); |
623 | if (page_get_page_cgroup(page)) { | |
624 | unlock_page_cgroup(page); | |
9175e031 | 625 | /* |
3be91277 HD |
626 | * Another charge has been added to this page already. |
627 | * We take lock_page_cgroup(page) again and read | |
9175e031 KH |
628 | * page->cgroup, increment refcnt.... just retry is OK. |
629 | */ | |
630 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
631 | css_put(&mem->css); | |
632 | kfree(pc); | |
633 | goto retry; | |
634 | } | |
7e924aaf HD |
635 | page_assign_page_cgroup(page, pc); |
636 | unlock_page_cgroup(page); | |
8a9f3ccd | 637 | |
072c56c1 KH |
638 | mz = page_cgroup_zoneinfo(pc); |
639 | spin_lock_irqsave(&mz->lru_lock, flags); | |
6d12e2d8 | 640 | __mem_cgroup_add_list(pc); |
072c56c1 | 641 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 642 | |
8a9f3ccd | 643 | done: |
8a9f3ccd | 644 | return 0; |
3be91277 HD |
645 | out: |
646 | css_put(&mem->css); | |
8a9f3ccd | 647 | kfree(pc); |
8a9f3ccd | 648 | err: |
8a9f3ccd BS |
649 | return -ENOMEM; |
650 | } | |
651 | ||
8869b8f6 | 652 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 KH |
653 | { |
654 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
8869b8f6 | 655 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
656 | } |
657 | ||
e1a1cd59 BS |
658 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
659 | gfp_t gfp_mask) | |
8697d331 | 660 | { |
8697d331 BS |
661 | if (!mm) |
662 | mm = &init_mm; | |
8869b8f6 | 663 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
217bc319 | 664 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
8697d331 BS |
665 | } |
666 | ||
8a9f3ccd BS |
667 | /* |
668 | * Uncharging is always a welcome operation, we never complain, simply | |
8289546e | 669 | * uncharge. |
8a9f3ccd | 670 | */ |
8289546e | 671 | void mem_cgroup_uncharge_page(struct page *page) |
8a9f3ccd | 672 | { |
8289546e | 673 | struct page_cgroup *pc; |
8a9f3ccd | 674 | struct mem_cgroup *mem; |
072c56c1 | 675 | struct mem_cgroup_per_zone *mz; |
66e1707b | 676 | unsigned long flags; |
8a9f3ccd | 677 | |
8697d331 | 678 | /* |
3c541e14 | 679 | * Check if our page_cgroup is valid |
8697d331 | 680 | */ |
8289546e HD |
681 | lock_page_cgroup(page); |
682 | pc = page_get_page_cgroup(page); | |
8a9f3ccd | 683 | if (!pc) |
8289546e | 684 | goto unlock; |
8a9f3ccd BS |
685 | |
686 | if (atomic_dec_and_test(&pc->ref_cnt)) { | |
687 | page = pc->page; | |
072c56c1 | 688 | mz = page_cgroup_zoneinfo(pc); |
9175e031 KH |
689 | /* |
690 | * get page->cgroup and clear it under lock. | |
cc847582 | 691 | * force_empty can drop page->cgroup without checking refcnt. |
9175e031 | 692 | */ |
3c541e14 | 693 | unlock_page_cgroup(page); |
9175e031 KH |
694 | if (clear_page_cgroup(page, pc) == pc) { |
695 | mem = pc->mem_cgroup; | |
696 | css_put(&mem->css); | |
697 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
072c56c1 | 698 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 699 | __mem_cgroup_remove_list(pc); |
072c56c1 | 700 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9175e031 | 701 | kfree(pc); |
9175e031 | 702 | } |
3c541e14 | 703 | lock_page_cgroup(page); |
8a9f3ccd | 704 | } |
6d12e2d8 | 705 | |
8289546e | 706 | unlock: |
3c541e14 BS |
707 | unlock_page_cgroup(page); |
708 | } | |
709 | ||
ae41be37 KH |
710 | /* |
711 | * Returns non-zero if a page (under migration) has valid page_cgroup member. | |
712 | * Refcnt of page_cgroup is incremented. | |
713 | */ | |
ae41be37 KH |
714 | int mem_cgroup_prepare_migration(struct page *page) |
715 | { | |
716 | struct page_cgroup *pc; | |
717 | int ret = 0; | |
8869b8f6 | 718 | |
ae41be37 KH |
719 | lock_page_cgroup(page); |
720 | pc = page_get_page_cgroup(page); | |
721 | if (pc && atomic_inc_not_zero(&pc->ref_cnt)) | |
722 | ret = 1; | |
723 | unlock_page_cgroup(page); | |
724 | return ret; | |
725 | } | |
726 | ||
727 | void mem_cgroup_end_migration(struct page *page) | |
728 | { | |
8289546e | 729 | mem_cgroup_uncharge_page(page); |
ae41be37 | 730 | } |
8869b8f6 | 731 | |
ae41be37 | 732 | /* |
8869b8f6 | 733 | * We know both *page* and *newpage* are now not-on-LRU and PG_locked. |
ae41be37 KH |
734 | * And no race with uncharge() routines because page_cgroup for *page* |
735 | * has extra one reference by mem_cgroup_prepare_migration. | |
736 | */ | |
ae41be37 KH |
737 | void mem_cgroup_page_migration(struct page *page, struct page *newpage) |
738 | { | |
739 | struct page_cgroup *pc; | |
072c56c1 | 740 | struct mem_cgroup_per_zone *mz; |
d5b69e38 | 741 | unsigned long flags; |
8869b8f6 | 742 | |
ae41be37 KH |
743 | retry: |
744 | pc = page_get_page_cgroup(page); | |
745 | if (!pc) | |
746 | return; | |
8869b8f6 | 747 | |
072c56c1 | 748 | mz = page_cgroup_zoneinfo(pc); |
ae41be37 KH |
749 | if (clear_page_cgroup(page, pc) != pc) |
750 | goto retry; | |
6d12e2d8 | 751 | |
8869b8f6 | 752 | spin_lock_irqsave(&mz->lru_lock, flags); |
6d12e2d8 | 753 | __mem_cgroup_remove_list(pc); |
072c56c1 KH |
754 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
755 | ||
ae41be37 KH |
756 | pc->page = newpage; |
757 | lock_page_cgroup(newpage); | |
758 | page_assign_page_cgroup(newpage, pc); | |
759 | unlock_page_cgroup(newpage); | |
6d12e2d8 | 760 | |
072c56c1 KH |
761 | mz = page_cgroup_zoneinfo(pc); |
762 | spin_lock_irqsave(&mz->lru_lock, flags); | |
763 | __mem_cgroup_add_list(pc); | |
764 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
ae41be37 | 765 | } |
78fb7466 | 766 | |
cc847582 KH |
767 | /* |
768 | * This routine traverse page_cgroup in given list and drop them all. | |
769 | * This routine ignores page_cgroup->ref_cnt. | |
770 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
771 | */ | |
772 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 773 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 KH |
774 | struct mem_cgroup_per_zone *mz, |
775 | int active) | |
cc847582 KH |
776 | { |
777 | struct page_cgroup *pc; | |
778 | struct page *page; | |
779 | int count; | |
780 | unsigned long flags; | |
072c56c1 KH |
781 | struct list_head *list; |
782 | ||
783 | if (active) | |
784 | list = &mz->active_list; | |
785 | else | |
786 | list = &mz->inactive_list; | |
cc847582 | 787 | |
1ecaab2b KH |
788 | if (list_empty(list)) |
789 | return; | |
cc847582 KH |
790 | retry: |
791 | count = FORCE_UNCHARGE_BATCH; | |
072c56c1 | 792 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 KH |
793 | |
794 | while (--count && !list_empty(list)) { | |
795 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
796 | page = pc->page; | |
797 | /* Avoid race with charge */ | |
798 | atomic_set(&pc->ref_cnt, 0); | |
799 | if (clear_page_cgroup(page, pc) == pc) { | |
800 | css_put(&mem->css); | |
801 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
6d12e2d8 | 802 | __mem_cgroup_remove_list(pc); |
cc847582 KH |
803 | kfree(pc); |
804 | } else /* being uncharged ? ...do relax */ | |
805 | break; | |
806 | } | |
8869b8f6 | 807 | |
072c56c1 | 808 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
809 | if (!list_empty(list)) { |
810 | cond_resched(); | |
811 | goto retry; | |
812 | } | |
cc847582 KH |
813 | } |
814 | ||
815 | /* | |
816 | * make mem_cgroup's charge to be 0 if there is no task. | |
817 | * This enables deleting this mem_cgroup. | |
818 | */ | |
d5b69e38 | 819 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
820 | { |
821 | int ret = -EBUSY; | |
1ecaab2b | 822 | int node, zid; |
8869b8f6 | 823 | |
cc847582 KH |
824 | css_get(&mem->css); |
825 | /* | |
826 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 827 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
828 | * So, we have to do loop here until all lists are empty. |
829 | */ | |
1ecaab2b | 830 | while (mem->res.usage > 0) { |
cc847582 KH |
831 | if (atomic_read(&mem->css.cgroup->count) > 0) |
832 | goto out; | |
1ecaab2b KH |
833 | for_each_node_state(node, N_POSSIBLE) |
834 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
835 | struct mem_cgroup_per_zone *mz; | |
836 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
837 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 838 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 839 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 840 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 841 | } |
cc847582 KH |
842 | } |
843 | ret = 0; | |
844 | out: | |
845 | css_put(&mem->css); | |
846 | return ret; | |
847 | } | |
848 | ||
d5b69e38 | 849 | static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp) |
0eea1030 BS |
850 | { |
851 | *tmp = memparse(buf, &buf); | |
852 | if (*buf != '\0') | |
853 | return -EINVAL; | |
854 | ||
855 | /* | |
856 | * Round up the value to the closest page size | |
857 | */ | |
858 | *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT; | |
859 | return 0; | |
860 | } | |
861 | ||
862 | static ssize_t mem_cgroup_read(struct cgroup *cont, | |
863 | struct cftype *cft, struct file *file, | |
864 | char __user *userbuf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 BS |
865 | { |
866 | return res_counter_read(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
867 | cft->private, userbuf, nbytes, ppos, |
868 | NULL); | |
8cdea7c0 BS |
869 | } |
870 | ||
871 | static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
872 | struct file *file, const char __user *userbuf, | |
873 | size_t nbytes, loff_t *ppos) | |
874 | { | |
875 | return res_counter_write(&mem_cgroup_from_cont(cont)->res, | |
0eea1030 BS |
876 | cft->private, userbuf, nbytes, ppos, |
877 | mem_cgroup_write_strategy); | |
8cdea7c0 BS |
878 | } |
879 | ||
cc847582 KH |
880 | static ssize_t mem_force_empty_write(struct cgroup *cont, |
881 | struct cftype *cft, struct file *file, | |
882 | const char __user *userbuf, | |
883 | size_t nbytes, loff_t *ppos) | |
884 | { | |
885 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
8869b8f6 | 886 | int ret = mem_cgroup_force_empty(mem); |
cc847582 KH |
887 | if (!ret) |
888 | ret = nbytes; | |
889 | return ret; | |
890 | } | |
891 | ||
892 | /* | |
893 | * Note: This should be removed if cgroup supports write-only file. | |
894 | */ | |
cc847582 KH |
895 | static ssize_t mem_force_empty_read(struct cgroup *cont, |
896 | struct cftype *cft, | |
897 | struct file *file, char __user *userbuf, | |
898 | size_t nbytes, loff_t *ppos) | |
899 | { | |
900 | return -EINVAL; | |
901 | } | |
902 | ||
d2ceb9b7 KH |
903 | static const struct mem_cgroup_stat_desc { |
904 | const char *msg; | |
905 | u64 unit; | |
906 | } mem_cgroup_stat_desc[] = { | |
907 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
908 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
909 | }; | |
910 | ||
911 | static int mem_control_stat_show(struct seq_file *m, void *arg) | |
912 | { | |
913 | struct cgroup *cont = m->private; | |
914 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
915 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
916 | int i; | |
917 | ||
918 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
919 | s64 val; | |
920 | ||
921 | val = mem_cgroup_read_stat(stat, i); | |
922 | val *= mem_cgroup_stat_desc[i].unit; | |
923 | seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg, | |
924 | (long long)val); | |
925 | } | |
6d12e2d8 KH |
926 | /* showing # of active pages */ |
927 | { | |
928 | unsigned long active, inactive; | |
929 | ||
930 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
931 | MEM_CGROUP_ZSTAT_INACTIVE); | |
932 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
933 | MEM_CGROUP_ZSTAT_ACTIVE); | |
934 | seq_printf(m, "active %ld\n", (active) * PAGE_SIZE); | |
935 | seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE); | |
936 | } | |
d2ceb9b7 KH |
937 | return 0; |
938 | } | |
939 | ||
940 | static const struct file_operations mem_control_stat_file_operations = { | |
941 | .read = seq_read, | |
942 | .llseek = seq_lseek, | |
943 | .release = single_release, | |
944 | }; | |
945 | ||
946 | static int mem_control_stat_open(struct inode *unused, struct file *file) | |
947 | { | |
948 | /* XXX __d_cont */ | |
949 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
950 | ||
951 | file->f_op = &mem_control_stat_file_operations; | |
952 | return single_open(file, mem_control_stat_show, cont); | |
953 | } | |
954 | ||
8cdea7c0 BS |
955 | static struct cftype mem_cgroup_files[] = { |
956 | { | |
0eea1030 | 957 | .name = "usage_in_bytes", |
8cdea7c0 BS |
958 | .private = RES_USAGE, |
959 | .read = mem_cgroup_read, | |
960 | }, | |
961 | { | |
0eea1030 | 962 | .name = "limit_in_bytes", |
8cdea7c0 BS |
963 | .private = RES_LIMIT, |
964 | .write = mem_cgroup_write, | |
965 | .read = mem_cgroup_read, | |
966 | }, | |
967 | { | |
968 | .name = "failcnt", | |
969 | .private = RES_FAILCNT, | |
970 | .read = mem_cgroup_read, | |
971 | }, | |
cc847582 KH |
972 | { |
973 | .name = "force_empty", | |
974 | .write = mem_force_empty_write, | |
975 | .read = mem_force_empty_read, | |
976 | }, | |
d2ceb9b7 KH |
977 | { |
978 | .name = "stat", | |
979 | .open = mem_control_stat_open, | |
980 | }, | |
8cdea7c0 BS |
981 | }; |
982 | ||
6d12e2d8 KH |
983 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
984 | { | |
985 | struct mem_cgroup_per_node *pn; | |
1ecaab2b KH |
986 | struct mem_cgroup_per_zone *mz; |
987 | int zone; | |
988 | /* | |
989 | * This routine is called against possible nodes. | |
990 | * But it's BUG to call kmalloc() against offline node. | |
991 | * | |
992 | * TODO: this routine can waste much memory for nodes which will | |
993 | * never be onlined. It's better to use memory hotplug callback | |
994 | * function. | |
995 | */ | |
996 | if (node_state(node, N_HIGH_MEMORY)) | |
997 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
998 | else | |
999 | pn = kmalloc(sizeof(*pn), GFP_KERNEL); | |
6d12e2d8 KH |
1000 | if (!pn) |
1001 | return 1; | |
1ecaab2b | 1002 | |
6d12e2d8 KH |
1003 | mem->info.nodeinfo[node] = pn; |
1004 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1005 | |
1006 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1007 | mz = &pn->zoneinfo[zone]; | |
1008 | INIT_LIST_HEAD(&mz->active_list); | |
1009 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 1010 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 1011 | } |
6d12e2d8 KH |
1012 | return 0; |
1013 | } | |
1014 | ||
1ecaab2b KH |
1015 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1016 | { | |
1017 | kfree(mem->info.nodeinfo[node]); | |
1018 | } | |
1019 | ||
8cdea7c0 BS |
1020 | static struct cgroup_subsys_state * |
1021 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1022 | { | |
1023 | struct mem_cgroup *mem; | |
6d12e2d8 | 1024 | int node; |
8cdea7c0 | 1025 | |
78fb7466 PE |
1026 | if (unlikely((cont->parent) == NULL)) { |
1027 | mem = &init_mem_cgroup; | |
1028 | init_mm.mem_cgroup = mem; | |
1029 | } else | |
1030 | mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL); | |
1031 | ||
1032 | if (mem == NULL) | |
2dda81ca | 1033 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1034 | |
1035 | res_counter_init(&mem->res); | |
1ecaab2b | 1036 | |
6d12e2d8 KH |
1037 | memset(&mem->info, 0, sizeof(mem->info)); |
1038 | ||
1039 | for_each_node_state(node, N_POSSIBLE) | |
1040 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1041 | goto free_out; | |
1042 | ||
8cdea7c0 | 1043 | return &mem->css; |
6d12e2d8 KH |
1044 | free_out: |
1045 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1046 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 KH |
1047 | if (cont->parent != NULL) |
1048 | kfree(mem); | |
2dda81ca | 1049 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1050 | } |
1051 | ||
df878fb0 KH |
1052 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1053 | struct cgroup *cont) | |
1054 | { | |
1055 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1056 | mem_cgroup_force_empty(mem); | |
1057 | } | |
1058 | ||
8cdea7c0 BS |
1059 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1060 | struct cgroup *cont) | |
1061 | { | |
6d12e2d8 KH |
1062 | int node; |
1063 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1064 | ||
1065 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1066 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1067 | |
8cdea7c0 BS |
1068 | kfree(mem_cgroup_from_cont(cont)); |
1069 | } | |
1070 | ||
1071 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1072 | struct cgroup *cont) | |
1073 | { | |
1074 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1075 | ARRAY_SIZE(mem_cgroup_files)); | |
1076 | } | |
1077 | ||
67e465a7 BS |
1078 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1079 | struct cgroup *cont, | |
1080 | struct cgroup *old_cont, | |
1081 | struct task_struct *p) | |
1082 | { | |
1083 | struct mm_struct *mm; | |
1084 | struct mem_cgroup *mem, *old_mem; | |
1085 | ||
1086 | mm = get_task_mm(p); | |
1087 | if (mm == NULL) | |
1088 | return; | |
1089 | ||
1090 | mem = mem_cgroup_from_cont(cont); | |
1091 | old_mem = mem_cgroup_from_cont(old_cont); | |
1092 | ||
1093 | if (mem == old_mem) | |
1094 | goto out; | |
1095 | ||
1096 | /* | |
1097 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1098 | * in effect owned by the leader | |
1099 | */ | |
1100 | if (p->tgid != p->pid) | |
1101 | goto out; | |
1102 | ||
1103 | css_get(&mem->css); | |
1104 | rcu_assign_pointer(mm->mem_cgroup, mem); | |
1105 | css_put(&old_mem->css); | |
1106 | ||
1107 | out: | |
1108 | mmput(mm); | |
67e465a7 BS |
1109 | } |
1110 | ||
8cdea7c0 BS |
1111 | struct cgroup_subsys mem_cgroup_subsys = { |
1112 | .name = "memory", | |
1113 | .subsys_id = mem_cgroup_subsys_id, | |
1114 | .create = mem_cgroup_create, | |
df878fb0 | 1115 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1116 | .destroy = mem_cgroup_destroy, |
1117 | .populate = mem_cgroup_populate, | |
67e465a7 | 1118 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1119 | .early_init = 0, |
8cdea7c0 | 1120 | }; |