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> |
d13d1443 | 24 | #include <linux/pagemap.h> |
d52aa412 | 25 | #include <linux/smp.h> |
8a9f3ccd | 26 | #include <linux/page-flags.h> |
66e1707b | 27 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
28 | #include <linux/bit_spinlock.h> |
29 | #include <linux/rcupdate.h> | |
b6ac57d5 | 30 | #include <linux/slab.h> |
66e1707b BS |
31 | #include <linux/swap.h> |
32 | #include <linux/spinlock.h> | |
33 | #include <linux/fs.h> | |
d2ceb9b7 | 34 | #include <linux/seq_file.h> |
33327948 | 35 | #include <linux/vmalloc.h> |
b69408e8 | 36 | #include <linux/mm_inline.h> |
52d4b9ac | 37 | #include <linux/page_cgroup.h> |
8cdea7c0 | 38 | |
8697d331 BS |
39 | #include <asm/uaccess.h> |
40 | ||
a181b0e8 | 41 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 42 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 43 | |
d52aa412 KH |
44 | /* |
45 | * Statistics for memory cgroup. | |
46 | */ | |
47 | enum mem_cgroup_stat_index { | |
48 | /* | |
49 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
50 | */ | |
51 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
52 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
53 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
54 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
55 | |
56 | MEM_CGROUP_STAT_NSTATS, | |
57 | }; | |
58 | ||
59 | struct mem_cgroup_stat_cpu { | |
60 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
61 | } ____cacheline_aligned_in_smp; | |
62 | ||
63 | struct mem_cgroup_stat { | |
c8dad2bb | 64 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
65 | }; |
66 | ||
67 | /* | |
68 | * For accounting under irq disable, no need for increment preempt count. | |
69 | */ | |
addb9efe | 70 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
71 | enum mem_cgroup_stat_index idx, int val) |
72 | { | |
addb9efe | 73 | stat->count[idx] += val; |
d52aa412 KH |
74 | } |
75 | ||
76 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
77 | enum mem_cgroup_stat_index idx) | |
78 | { | |
79 | int cpu; | |
80 | s64 ret = 0; | |
81 | for_each_possible_cpu(cpu) | |
82 | ret += stat->cpustat[cpu].count[idx]; | |
83 | return ret; | |
84 | } | |
85 | ||
6d12e2d8 KH |
86 | /* |
87 | * per-zone information in memory controller. | |
88 | */ | |
6d12e2d8 | 89 | struct mem_cgroup_per_zone { |
072c56c1 KH |
90 | /* |
91 | * spin_lock to protect the per cgroup LRU | |
92 | */ | |
93 | spinlock_t lru_lock; | |
b69408e8 CL |
94 | struct list_head lists[NR_LRU_LISTS]; |
95 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
96 | }; |
97 | /* Macro for accessing counter */ | |
98 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
99 | ||
100 | struct mem_cgroup_per_node { | |
101 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
102 | }; | |
103 | ||
104 | struct mem_cgroup_lru_info { | |
105 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
106 | }; | |
107 | ||
8cdea7c0 BS |
108 | /* |
109 | * The memory controller data structure. The memory controller controls both | |
110 | * page cache and RSS per cgroup. We would eventually like to provide | |
111 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
112 | * to help the administrator determine what knobs to tune. | |
113 | * | |
114 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
115 | * we hit the water mark. May be even add a low water mark, such that |
116 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
117 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
118 | */ |
119 | struct mem_cgroup { | |
120 | struct cgroup_subsys_state css; | |
121 | /* | |
122 | * the counter to account for memory usage | |
123 | */ | |
124 | struct res_counter res; | |
78fb7466 PE |
125 | /* |
126 | * Per cgroup active and inactive list, similar to the | |
127 | * per zone LRU lists. | |
78fb7466 | 128 | */ |
6d12e2d8 | 129 | struct mem_cgroup_lru_info info; |
072c56c1 | 130 | |
6c48a1d0 | 131 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 | 132 | /* |
c8dad2bb | 133 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
134 | */ |
135 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
136 | }; |
137 | ||
217bc319 KH |
138 | enum charge_type { |
139 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
140 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 141 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 142 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 143 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
c05555b5 KH |
144 | NR_CHARGE_TYPE, |
145 | }; | |
146 | ||
52d4b9ac KH |
147 | /* only for here (for easy reading.) */ |
148 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
149 | #define PCGF_USED (1UL << PCG_USED) | |
150 | #define PCGF_ACTIVE (1UL << PCG_ACTIVE) | |
151 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
152 | #define PCGF_FILE (1UL << PCG_FILE) | |
c05555b5 KH |
153 | static const unsigned long |
154 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
52d4b9ac KH |
155 | PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
156 | PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ | |
157 | PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
158 | 0, /* FORCE */ | |
217bc319 KH |
159 | }; |
160 | ||
d52aa412 KH |
161 | /* |
162 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
163 | */ | |
c05555b5 KH |
164 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
165 | struct page_cgroup *pc, | |
166 | bool charge) | |
d52aa412 KH |
167 | { |
168 | int val = (charge)? 1 : -1; | |
169 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 170 | struct mem_cgroup_stat_cpu *cpustat; |
d52aa412 | 171 | |
8869b8f6 | 172 | VM_BUG_ON(!irqs_disabled()); |
addb9efe KH |
173 | |
174 | cpustat = &stat->cpustat[smp_processor_id()]; | |
c05555b5 | 175 | if (PageCgroupCache(pc)) |
addb9efe | 176 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 177 | else |
addb9efe | 178 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
179 | |
180 | if (charge) | |
addb9efe | 181 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
182 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
183 | else | |
addb9efe | 184 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 185 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
6d12e2d8 KH |
186 | } |
187 | ||
d5b69e38 | 188 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
189 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
190 | { | |
6d12e2d8 KH |
191 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
192 | } | |
193 | ||
d5b69e38 | 194 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
195 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
196 | { | |
197 | struct mem_cgroup *mem = pc->mem_cgroup; | |
198 | int nid = page_cgroup_nid(pc); | |
199 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 200 | |
6d12e2d8 KH |
201 | return mem_cgroup_zoneinfo(mem, nid, zid); |
202 | } | |
203 | ||
204 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 205 | enum lru_list idx) |
6d12e2d8 KH |
206 | { |
207 | int nid, zid; | |
208 | struct mem_cgroup_per_zone *mz; | |
209 | u64 total = 0; | |
210 | ||
211 | for_each_online_node(nid) | |
212 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
213 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
214 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
215 | } | |
216 | return total; | |
d52aa412 KH |
217 | } |
218 | ||
d5b69e38 | 219 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
220 | { |
221 | return container_of(cgroup_subsys_state(cont, | |
222 | mem_cgroup_subsys_id), struct mem_cgroup, | |
223 | css); | |
224 | } | |
225 | ||
cf475ad2 | 226 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 227 | { |
31a78f23 BS |
228 | /* |
229 | * mm_update_next_owner() may clear mm->owner to NULL | |
230 | * if it races with swapoff, page migration, etc. | |
231 | * So this can be called with p == NULL. | |
232 | */ | |
233 | if (unlikely(!p)) | |
234 | return NULL; | |
235 | ||
78fb7466 PE |
236 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
237 | struct mem_cgroup, css); | |
238 | } | |
239 | ||
3eae90c3 KH |
240 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
241 | struct page_cgroup *pc) | |
6d12e2d8 | 242 | { |
4f98a2fe RR |
243 | int lru = LRU_BASE; |
244 | ||
c05555b5 | 245 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
246 | lru = LRU_UNEVICTABLE; |
247 | else { | |
c05555b5 | 248 | if (PageCgroupActive(pc)) |
894bc310 | 249 | lru += LRU_ACTIVE; |
c05555b5 | 250 | if (PageCgroupFile(pc)) |
894bc310 LS |
251 | lru += LRU_FILE; |
252 | } | |
6d12e2d8 | 253 | |
b69408e8 | 254 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 | 255 | |
c05555b5 | 256 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); |
508b7be0 | 257 | list_del(&pc->lru); |
6d12e2d8 KH |
258 | } |
259 | ||
3eae90c3 | 260 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
f817ed48 | 261 | struct page_cgroup *pc, bool hot) |
6d12e2d8 | 262 | { |
4f98a2fe | 263 | int lru = LRU_BASE; |
b69408e8 | 264 | |
c05555b5 | 265 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
266 | lru = LRU_UNEVICTABLE; |
267 | else { | |
c05555b5 | 268 | if (PageCgroupActive(pc)) |
894bc310 | 269 | lru += LRU_ACTIVE; |
c05555b5 | 270 | if (PageCgroupFile(pc)) |
894bc310 LS |
271 | lru += LRU_FILE; |
272 | } | |
b69408e8 CL |
273 | |
274 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
f817ed48 KH |
275 | if (hot) |
276 | list_add(&pc->lru, &mz->lists[lru]); | |
277 | else | |
278 | list_add_tail(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 279 | |
c05555b5 | 280 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); |
6d12e2d8 KH |
281 | } |
282 | ||
894bc310 | 283 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
66e1707b | 284 | { |
6d12e2d8 | 285 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
c05555b5 KH |
286 | int active = PageCgroupActive(pc); |
287 | int file = PageCgroupFile(pc); | |
288 | int unevictable = PageCgroupUnevictable(pc); | |
894bc310 LS |
289 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : |
290 | (LRU_FILE * !!file + !!active); | |
6d12e2d8 | 291 | |
894bc310 LS |
292 | if (lru == from) |
293 | return; | |
b69408e8 | 294 | |
894bc310 | 295 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
c05555b5 KH |
296 | /* |
297 | * However this is done under mz->lru_lock, another flags, which | |
298 | * are not related to LRU, will be modified from out-of-lock. | |
299 | * We have to use atomic set/clear flags. | |
300 | */ | |
894bc310 | 301 | if (is_unevictable_lru(lru)) { |
c05555b5 KH |
302 | ClearPageCgroupActive(pc); |
303 | SetPageCgroupUnevictable(pc); | |
894bc310 LS |
304 | } else { |
305 | if (is_active_lru(lru)) | |
c05555b5 | 306 | SetPageCgroupActive(pc); |
894bc310 | 307 | else |
c05555b5 KH |
308 | ClearPageCgroupActive(pc); |
309 | ClearPageCgroupUnevictable(pc); | |
894bc310 | 310 | } |
b69408e8 | 311 | |
b69408e8 CL |
312 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
313 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
314 | } |
315 | ||
4c4a2214 DR |
316 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
317 | { | |
318 | int ret; | |
319 | ||
320 | task_lock(task); | |
bd845e38 | 321 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
322 | task_unlock(task); |
323 | return ret; | |
324 | } | |
325 | ||
66e1707b BS |
326 | /* |
327 | * This routine assumes that the appropriate zone's lru lock is already held | |
328 | */ | |
894bc310 | 329 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
66e1707b | 330 | { |
427d5416 | 331 | struct page_cgroup *pc; |
072c56c1 KH |
332 | struct mem_cgroup_per_zone *mz; |
333 | unsigned long flags; | |
334 | ||
cede86ac LZ |
335 | if (mem_cgroup_subsys.disabled) |
336 | return; | |
337 | ||
2680eed7 HD |
338 | /* |
339 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
340 | * because other holders of lock_page_cgroup can be interrupted | |
341 | * with an attempt to rotate_reclaimable_page. But we cannot | |
342 | * safely get to page_cgroup without it, so just try_lock it: | |
343 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
344 | */ | |
52d4b9ac KH |
345 | pc = lookup_page_cgroup(page); |
346 | if (!trylock_page_cgroup(pc)) | |
66e1707b | 347 | return; |
52d4b9ac | 348 | if (pc && PageCgroupUsed(pc)) { |
2680eed7 | 349 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 350 | spin_lock_irqsave(&mz->lru_lock, flags); |
894bc310 | 351 | __mem_cgroup_move_lists(pc, lru); |
2680eed7 | 352 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 | 353 | } |
52d4b9ac | 354 | unlock_page_cgroup(pc); |
66e1707b BS |
355 | } |
356 | ||
58ae83db KH |
357 | /* |
358 | * Calculate mapped_ratio under memory controller. This will be used in | |
359 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
360 | */ | |
361 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
362 | { | |
363 | long total, rss; | |
364 | ||
365 | /* | |
366 | * usage is recorded in bytes. But, here, we assume the number of | |
367 | * physical pages can be represented by "long" on any arch. | |
368 | */ | |
369 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
370 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
371 | return (int)((rss * 100L) / total); | |
372 | } | |
8869b8f6 | 373 | |
6c48a1d0 KH |
374 | /* |
375 | * prev_priority control...this will be used in memory reclaim path. | |
376 | */ | |
377 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
378 | { | |
379 | return mem->prev_priority; | |
380 | } | |
381 | ||
382 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
383 | { | |
384 | if (priority < mem->prev_priority) | |
385 | mem->prev_priority = priority; | |
386 | } | |
387 | ||
388 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
389 | { | |
390 | mem->prev_priority = priority; | |
391 | } | |
392 | ||
cc38108e KH |
393 | /* |
394 | * Calculate # of pages to be scanned in this priority/zone. | |
395 | * See also vmscan.c | |
396 | * | |
397 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
398 | * (see include/linux/mmzone.h) | |
399 | */ | |
400 | ||
b69408e8 CL |
401 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
402 | int priority, enum lru_list lru) | |
cc38108e | 403 | { |
b69408e8 | 404 | long nr_pages; |
cc38108e KH |
405 | int nid = zone->zone_pgdat->node_id; |
406 | int zid = zone_idx(zone); | |
407 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
408 | ||
b69408e8 | 409 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 410 | |
b69408e8 | 411 | return (nr_pages >> priority); |
cc38108e KH |
412 | } |
413 | ||
66e1707b BS |
414 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
415 | struct list_head *dst, | |
416 | unsigned long *scanned, int order, | |
417 | int mode, struct zone *z, | |
418 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 419 | int active, int file) |
66e1707b BS |
420 | { |
421 | unsigned long nr_taken = 0; | |
422 | struct page *page; | |
423 | unsigned long scan; | |
424 | LIST_HEAD(pc_list); | |
425 | struct list_head *src; | |
ff7283fa | 426 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
427 | int nid = z->zone_pgdat->node_id; |
428 | int zid = zone_idx(z); | |
429 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 430 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 431 | |
cf475ad2 | 432 | BUG_ON(!mem_cont); |
1ecaab2b | 433 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 434 | src = &mz->lists[lru]; |
66e1707b | 435 | |
072c56c1 | 436 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
437 | scan = 0; |
438 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 439 | if (scan >= nr_to_scan) |
ff7283fa | 440 | break; |
52d4b9ac KH |
441 | if (unlikely(!PageCgroupUsed(pc))) |
442 | continue; | |
66e1707b | 443 | page = pc->page; |
66e1707b | 444 | |
436c6541 | 445 | if (unlikely(!PageLRU(page))) |
ff7283fa | 446 | continue; |
ff7283fa | 447 | |
4f98a2fe RR |
448 | /* |
449 | * TODO: play better with lumpy reclaim, grabbing anything. | |
450 | */ | |
894bc310 LS |
451 | if (PageUnevictable(page) || |
452 | (PageActive(page) && !active) || | |
453 | (!PageActive(page) && active)) { | |
454 | __mem_cgroup_move_lists(pc, page_lru(page)); | |
66e1707b BS |
455 | continue; |
456 | } | |
457 | ||
436c6541 HD |
458 | scan++; |
459 | list_move(&pc->lru, &pc_list); | |
66e1707b | 460 | |
4f98a2fe | 461 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
462 | list_move(&page->lru, dst); |
463 | nr_taken++; | |
464 | } | |
465 | } | |
466 | ||
467 | list_splice(&pc_list, src); | |
072c56c1 | 468 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
469 | |
470 | *scanned = scan; | |
471 | return nr_taken; | |
472 | } | |
473 | ||
f817ed48 KH |
474 | /* |
475 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
476 | * oom-killer can be invoked. | |
8a9f3ccd | 477 | */ |
f817ed48 KH |
478 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
479 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) | |
8a9f3ccd BS |
480 | { |
481 | struct mem_cgroup *mem; | |
7a81b88c | 482 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
8a9f3ccd | 483 | /* |
3be91277 HD |
484 | * We always charge the cgroup the mm_struct belongs to. |
485 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
486 | * thread group leader migrates. It's possible that mm is not |
487 | * set, if so charge the init_mm (happens for pagecache usage). | |
488 | */ | |
7a81b88c | 489 | if (likely(!*memcg)) { |
e8589cc1 KH |
490 | rcu_read_lock(); |
491 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
492 | if (unlikely(!mem)) { |
493 | rcu_read_unlock(); | |
31a78f23 BS |
494 | return 0; |
495 | } | |
e8589cc1 KH |
496 | /* |
497 | * For every charge from the cgroup, increment reference count | |
498 | */ | |
499 | css_get(&mem->css); | |
7a81b88c | 500 | *memcg = mem; |
e8589cc1 KH |
501 | rcu_read_unlock(); |
502 | } else { | |
7a81b88c KH |
503 | mem = *memcg; |
504 | css_get(&mem->css); | |
e8589cc1 | 505 | } |
8a9f3ccd | 506 | |
7a81b88c | 507 | |
addb9efe | 508 | while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { |
3be91277 | 509 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 510 | goto nomem; |
e1a1cd59 BS |
511 | |
512 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
513 | continue; |
514 | ||
515 | /* | |
8869b8f6 HD |
516 | * try_to_free_mem_cgroup_pages() might not give us a full |
517 | * picture of reclaim. Some pages are reclaimed and might be | |
518 | * moved to swap cache or just unmapped from the cgroup. | |
519 | * Check the limit again to see if the reclaim reduced the | |
520 | * current usage of the cgroup before giving up | |
521 | */ | |
66e1707b BS |
522 | if (res_counter_check_under_limit(&mem->res)) |
523 | continue; | |
3be91277 HD |
524 | |
525 | if (!nr_retries--) { | |
f817ed48 KH |
526 | if (oom) |
527 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
7a81b88c | 528 | goto nomem; |
66e1707b | 529 | } |
8a9f3ccd | 530 | } |
7a81b88c KH |
531 | return 0; |
532 | nomem: | |
533 | css_put(&mem->css); | |
534 | return -ENOMEM; | |
535 | } | |
8a9f3ccd | 536 | |
f817ed48 KH |
537 | /** |
538 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
539 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
540 | * @gfp_mask: gfp_mask for reclaim. | |
541 | * @memcg: a pointer to memory cgroup which is charged against. | |
542 | * | |
543 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
544 | * memory cgroup from @mm is got and stored in *memcg. | |
545 | * | |
546 | * Returns 0 if success. -ENOMEM at failure. | |
547 | * This call can invoke OOM-Killer. | |
548 | */ | |
549 | ||
550 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
551 | gfp_t mask, struct mem_cgroup **memcg) | |
552 | { | |
553 | return __mem_cgroup_try_charge(mm, mask, memcg, true); | |
554 | } | |
555 | ||
7a81b88c KH |
556 | /* |
557 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
558 | * USED state. If already USED, uncharge and return. | |
559 | */ | |
560 | ||
561 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
562 | struct page_cgroup *pc, | |
563 | enum charge_type ctype) | |
564 | { | |
565 | struct mem_cgroup_per_zone *mz; | |
566 | unsigned long flags; | |
567 | ||
568 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
569 | if (!mem) | |
570 | return; | |
52d4b9ac KH |
571 | |
572 | lock_page_cgroup(pc); | |
573 | if (unlikely(PageCgroupUsed(pc))) { | |
574 | unlock_page_cgroup(pc); | |
575 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
576 | css_put(&mem->css); | |
7a81b88c | 577 | return; |
52d4b9ac | 578 | } |
8a9f3ccd | 579 | pc->mem_cgroup = mem; |
508b7be0 KH |
580 | /* |
581 | * If a page is accounted as a page cache, insert to inactive list. | |
582 | * If anon, insert to active list. | |
583 | */ | |
c05555b5 | 584 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 585 | |
072c56c1 | 586 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 587 | |
072c56c1 | 588 | spin_lock_irqsave(&mz->lru_lock, flags); |
f817ed48 | 589 | __mem_cgroup_add_list(mz, pc, true); |
072c56c1 | 590 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
52d4b9ac | 591 | unlock_page_cgroup(pc); |
7a81b88c | 592 | } |
66e1707b | 593 | |
f817ed48 KH |
594 | /** |
595 | * mem_cgroup_move_account - move account of the page | |
596 | * @pc: page_cgroup of the page. | |
597 | * @from: mem_cgroup which the page is moved from. | |
598 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
599 | * | |
600 | * The caller must confirm following. | |
601 | * 1. disable irq. | |
602 | * 2. lru_lock of old mem_cgroup(@from) should be held. | |
603 | * | |
604 | * returns 0 at success, | |
605 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
606 | * | |
607 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
608 | * new cgroup. It should be done by a caller. | |
609 | */ | |
610 | ||
611 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
612 | struct mem_cgroup *from, struct mem_cgroup *to) | |
613 | { | |
614 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
615 | int nid, zid; | |
616 | int ret = -EBUSY; | |
617 | ||
618 | VM_BUG_ON(!irqs_disabled()); | |
619 | VM_BUG_ON(from == to); | |
620 | ||
621 | nid = page_cgroup_nid(pc); | |
622 | zid = page_cgroup_zid(pc); | |
623 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
624 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
625 | ||
626 | ||
627 | if (!trylock_page_cgroup(pc)) | |
628 | return ret; | |
629 | ||
630 | if (!PageCgroupUsed(pc)) | |
631 | goto out; | |
632 | ||
633 | if (pc->mem_cgroup != from) | |
634 | goto out; | |
635 | ||
636 | if (spin_trylock(&to_mz->lru_lock)) { | |
637 | __mem_cgroup_remove_list(from_mz, pc); | |
638 | css_put(&from->css); | |
639 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
640 | pc->mem_cgroup = to; | |
641 | css_get(&to->css); | |
642 | __mem_cgroup_add_list(to_mz, pc, false); | |
643 | ret = 0; | |
644 | spin_unlock(&to_mz->lru_lock); | |
645 | } | |
646 | out: | |
647 | unlock_page_cgroup(pc); | |
648 | return ret; | |
649 | } | |
650 | ||
651 | /* | |
652 | * move charges to its parent. | |
653 | */ | |
654 | ||
655 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
656 | struct mem_cgroup *child, | |
657 | gfp_t gfp_mask) | |
658 | { | |
659 | struct cgroup *cg = child->css.cgroup; | |
660 | struct cgroup *pcg = cg->parent; | |
661 | struct mem_cgroup *parent; | |
662 | struct mem_cgroup_per_zone *mz; | |
663 | unsigned long flags; | |
664 | int ret; | |
665 | ||
666 | /* Is ROOT ? */ | |
667 | if (!pcg) | |
668 | return -EINVAL; | |
669 | ||
670 | parent = mem_cgroup_from_cont(pcg); | |
671 | ||
672 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); | |
673 | if (ret) | |
674 | return ret; | |
675 | ||
676 | mz = mem_cgroup_zoneinfo(child, | |
677 | page_cgroup_nid(pc), page_cgroup_zid(pc)); | |
678 | ||
679 | spin_lock_irqsave(&mz->lru_lock, flags); | |
680 | ret = mem_cgroup_move_account(pc, child, parent); | |
681 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
682 | ||
683 | /* drop extra refcnt */ | |
684 | css_put(&parent->css); | |
685 | /* uncharge if move fails */ | |
686 | if (ret) | |
687 | res_counter_uncharge(&parent->res, PAGE_SIZE); | |
688 | ||
689 | return ret; | |
690 | } | |
691 | ||
7a81b88c KH |
692 | /* |
693 | * Charge the memory controller for page usage. | |
694 | * Return | |
695 | * 0 if the charge was successful | |
696 | * < 0 if the cgroup is over its limit | |
697 | */ | |
698 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
699 | gfp_t gfp_mask, enum charge_type ctype, | |
700 | struct mem_cgroup *memcg) | |
701 | { | |
702 | struct mem_cgroup *mem; | |
703 | struct page_cgroup *pc; | |
704 | int ret; | |
705 | ||
706 | pc = lookup_page_cgroup(page); | |
707 | /* can happen at boot */ | |
708 | if (unlikely(!pc)) | |
709 | return 0; | |
710 | prefetchw(pc); | |
711 | ||
712 | mem = memcg; | |
f817ed48 | 713 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
7a81b88c KH |
714 | if (ret) |
715 | return ret; | |
716 | ||
717 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 718 | return 0; |
8a9f3ccd BS |
719 | } |
720 | ||
7a81b88c KH |
721 | int mem_cgroup_newpage_charge(struct page *page, |
722 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 723 | { |
cede86ac LZ |
724 | if (mem_cgroup_subsys.disabled) |
725 | return 0; | |
52d4b9ac KH |
726 | if (PageCompound(page)) |
727 | return 0; | |
69029cd5 KH |
728 | /* |
729 | * If already mapped, we don't have to account. | |
730 | * If page cache, page->mapping has address_space. | |
731 | * But page->mapping may have out-of-use anon_vma pointer, | |
732 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
733 | * is NULL. | |
734 | */ | |
735 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
736 | return 0; | |
737 | if (unlikely(!mm)) | |
738 | mm = &init_mm; | |
217bc319 | 739 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 740 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
741 | } |
742 | ||
e1a1cd59 BS |
743 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
744 | gfp_t gfp_mask) | |
8697d331 | 745 | { |
cede86ac LZ |
746 | if (mem_cgroup_subsys.disabled) |
747 | return 0; | |
52d4b9ac KH |
748 | if (PageCompound(page)) |
749 | return 0; | |
accf163e KH |
750 | /* |
751 | * Corner case handling. This is called from add_to_page_cache() | |
752 | * in usual. But some FS (shmem) precharges this page before calling it | |
753 | * and call add_to_page_cache() with GFP_NOWAIT. | |
754 | * | |
755 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
756 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
757 | * charge twice. (It works but has to pay a bit larger cost.) | |
758 | */ | |
759 | if (!(gfp_mask & __GFP_WAIT)) { | |
760 | struct page_cgroup *pc; | |
761 | ||
52d4b9ac KH |
762 | |
763 | pc = lookup_page_cgroup(page); | |
764 | if (!pc) | |
765 | return 0; | |
766 | lock_page_cgroup(pc); | |
767 | if (PageCgroupUsed(pc)) { | |
768 | unlock_page_cgroup(pc); | |
accf163e KH |
769 | return 0; |
770 | } | |
52d4b9ac | 771 | unlock_page_cgroup(pc); |
accf163e KH |
772 | } |
773 | ||
69029cd5 | 774 | if (unlikely(!mm)) |
8697d331 | 775 | mm = &init_mm; |
accf163e | 776 | |
c05555b5 KH |
777 | if (page_is_file_cache(page)) |
778 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 779 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
c05555b5 KH |
780 | else |
781 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
782 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
e8589cc1 KH |
783 | } |
784 | ||
d13d1443 KH |
785 | #ifdef CONFIG_SWAP |
786 | int mem_cgroup_cache_charge_swapin(struct page *page, | |
787 | struct mm_struct *mm, gfp_t mask, bool locked) | |
788 | { | |
789 | int ret = 0; | |
790 | ||
791 | if (mem_cgroup_subsys.disabled) | |
792 | return 0; | |
793 | if (unlikely(!mm)) | |
794 | mm = &init_mm; | |
795 | if (!locked) | |
796 | lock_page(page); | |
797 | /* | |
798 | * If not locked, the page can be dropped from SwapCache until | |
799 | * we reach here. | |
800 | */ | |
801 | if (PageSwapCache(page)) { | |
802 | ret = mem_cgroup_charge_common(page, mm, mask, | |
803 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
804 | } | |
805 | if (!locked) | |
806 | unlock_page(page); | |
807 | ||
808 | return ret; | |
809 | } | |
810 | #endif | |
811 | ||
7a81b88c KH |
812 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
813 | { | |
814 | struct page_cgroup *pc; | |
815 | ||
816 | if (mem_cgroup_subsys.disabled) | |
817 | return; | |
818 | if (!ptr) | |
819 | return; | |
820 | pc = lookup_page_cgroup(page); | |
821 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
822 | } | |
823 | ||
824 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
825 | { | |
826 | if (mem_cgroup_subsys.disabled) | |
827 | return; | |
828 | if (!mem) | |
829 | return; | |
830 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
831 | css_put(&mem->css); | |
832 | } | |
833 | ||
834 | ||
8a9f3ccd | 835 | /* |
69029cd5 | 836 | * uncharge if !page_mapped(page) |
8a9f3ccd | 837 | */ |
69029cd5 KH |
838 | static void |
839 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 840 | { |
8289546e | 841 | struct page_cgroup *pc; |
8a9f3ccd | 842 | struct mem_cgroup *mem; |
072c56c1 | 843 | struct mem_cgroup_per_zone *mz; |
66e1707b | 844 | unsigned long flags; |
8a9f3ccd | 845 | |
4077960e BS |
846 | if (mem_cgroup_subsys.disabled) |
847 | return; | |
848 | ||
d13d1443 KH |
849 | if (PageSwapCache(page)) |
850 | return; | |
851 | ||
8697d331 | 852 | /* |
3c541e14 | 853 | * Check if our page_cgroup is valid |
8697d331 | 854 | */ |
52d4b9ac KH |
855 | pc = lookup_page_cgroup(page); |
856 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
857 | return; | |
b9c565d5 | 858 | |
52d4b9ac | 859 | lock_page_cgroup(pc); |
d13d1443 KH |
860 | |
861 | if (!PageCgroupUsed(pc)) | |
862 | goto unlock_out; | |
863 | ||
864 | switch (ctype) { | |
865 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
866 | if (page_mapped(page)) | |
867 | goto unlock_out; | |
868 | break; | |
869 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
870 | if (!PageAnon(page)) { /* Shared memory */ | |
871 | if (page->mapping && !page_is_file_cache(page)) | |
872 | goto unlock_out; | |
873 | } else if (page_mapped(page)) /* Anon */ | |
874 | goto unlock_out; | |
875 | break; | |
876 | default: | |
877 | break; | |
52d4b9ac | 878 | } |
d13d1443 | 879 | |
52d4b9ac KH |
880 | ClearPageCgroupUsed(pc); |
881 | mem = pc->mem_cgroup; | |
b9c565d5 | 882 | |
69029cd5 KH |
883 | mz = page_cgroup_zoneinfo(pc); |
884 | spin_lock_irqsave(&mz->lru_lock, flags); | |
885 | __mem_cgroup_remove_list(mz, pc); | |
886 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 887 | unlock_page_cgroup(pc); |
fb59e9f1 | 888 | |
69029cd5 KH |
889 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
890 | css_put(&mem->css); | |
6d12e2d8 | 891 | |
69029cd5 | 892 | return; |
d13d1443 KH |
893 | |
894 | unlock_out: | |
895 | unlock_page_cgroup(pc); | |
896 | return; | |
3c541e14 BS |
897 | } |
898 | ||
69029cd5 KH |
899 | void mem_cgroup_uncharge_page(struct page *page) |
900 | { | |
52d4b9ac KH |
901 | /* early check. */ |
902 | if (page_mapped(page)) | |
903 | return; | |
904 | if (page->mapping && !PageAnon(page)) | |
905 | return; | |
69029cd5 KH |
906 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
907 | } | |
908 | ||
909 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
910 | { | |
911 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 912 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
913 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
914 | } | |
915 | ||
d13d1443 KH |
916 | void mem_cgroup_uncharge_swapcache(struct page *page) |
917 | { | |
918 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
919 | } | |
920 | ||
ae41be37 | 921 | /* |
01b1ae63 KH |
922 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
923 | * page belongs to. | |
ae41be37 | 924 | */ |
01b1ae63 | 925 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
926 | { |
927 | struct page_cgroup *pc; | |
e8589cc1 | 928 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 929 | int ret = 0; |
8869b8f6 | 930 | |
4077960e BS |
931 | if (mem_cgroup_subsys.disabled) |
932 | return 0; | |
933 | ||
52d4b9ac KH |
934 | pc = lookup_page_cgroup(page); |
935 | lock_page_cgroup(pc); | |
936 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
937 | mem = pc->mem_cgroup; |
938 | css_get(&mem->css); | |
e8589cc1 | 939 | } |
52d4b9ac | 940 | unlock_page_cgroup(pc); |
01b1ae63 | 941 | |
e8589cc1 | 942 | if (mem) { |
01b1ae63 | 943 | ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem); |
e8589cc1 KH |
944 | css_put(&mem->css); |
945 | } | |
01b1ae63 | 946 | *ptr = mem; |
e8589cc1 | 947 | return ret; |
ae41be37 | 948 | } |
8869b8f6 | 949 | |
69029cd5 | 950 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
951 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
952 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 953 | { |
01b1ae63 KH |
954 | struct page *target, *unused; |
955 | struct page_cgroup *pc; | |
956 | enum charge_type ctype; | |
957 | ||
958 | if (!mem) | |
959 | return; | |
960 | ||
961 | /* at migration success, oldpage->mapping is NULL. */ | |
962 | if (oldpage->mapping) { | |
963 | target = oldpage; | |
964 | unused = NULL; | |
965 | } else { | |
966 | target = newpage; | |
967 | unused = oldpage; | |
968 | } | |
969 | ||
970 | if (PageAnon(target)) | |
971 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
972 | else if (page_is_file_cache(target)) | |
973 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
974 | else | |
975 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
976 | ||
977 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 978 | if (unused) |
01b1ae63 KH |
979 | __mem_cgroup_uncharge_common(unused, ctype); |
980 | ||
981 | pc = lookup_page_cgroup(target); | |
69029cd5 | 982 | /* |
01b1ae63 KH |
983 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
984 | * So, double-counting is effectively avoided. | |
985 | */ | |
986 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
987 | ||
988 | /* | |
989 | * Both of oldpage and newpage are still under lock_page(). | |
990 | * Then, we don't have to care about race in radix-tree. | |
991 | * But we have to be careful that this page is unmapped or not. | |
992 | * | |
993 | * There is a case for !page_mapped(). At the start of | |
994 | * migration, oldpage was mapped. But now, it's zapped. | |
995 | * But we know *target* page is not freed/reused under us. | |
996 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 997 | */ |
01b1ae63 KH |
998 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
999 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1000 | } |
78fb7466 | 1001 | |
c9b0ed51 KH |
1002 | /* |
1003 | * A call to try to shrink memory usage under specified resource controller. | |
1004 | * This is typically used for page reclaiming for shmem for reducing side | |
1005 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1006 | */ | |
1007 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
1008 | { | |
1009 | struct mem_cgroup *mem; | |
1010 | int progress = 0; | |
1011 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1012 | ||
cede86ac LZ |
1013 | if (mem_cgroup_subsys.disabled) |
1014 | return 0; | |
9623e078 HD |
1015 | if (!mm) |
1016 | return 0; | |
cede86ac | 1017 | |
c9b0ed51 KH |
1018 | rcu_read_lock(); |
1019 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
1020 | if (unlikely(!mem)) { |
1021 | rcu_read_unlock(); | |
1022 | return 0; | |
1023 | } | |
c9b0ed51 KH |
1024 | css_get(&mem->css); |
1025 | rcu_read_unlock(); | |
1026 | ||
1027 | do { | |
1028 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 1029 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
1030 | } while (!progress && --retry); |
1031 | ||
1032 | css_put(&mem->css); | |
1033 | if (!retry) | |
1034 | return -ENOMEM; | |
1035 | return 0; | |
1036 | } | |
1037 | ||
d38d2a75 KM |
1038 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
1039 | unsigned long long val) | |
628f4235 KH |
1040 | { |
1041 | ||
1042 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1043 | int progress; | |
1044 | int ret = 0; | |
1045 | ||
1046 | while (res_counter_set_limit(&memcg->res, val)) { | |
1047 | if (signal_pending(current)) { | |
1048 | ret = -EINTR; | |
1049 | break; | |
1050 | } | |
1051 | if (!retry_count) { | |
1052 | ret = -EBUSY; | |
1053 | break; | |
1054 | } | |
bced0520 KH |
1055 | progress = try_to_free_mem_cgroup_pages(memcg, |
1056 | GFP_HIGHUSER_MOVABLE); | |
628f4235 KH |
1057 | if (!progress) |
1058 | retry_count--; | |
1059 | } | |
1060 | return ret; | |
1061 | } | |
1062 | ||
1063 | ||
cc847582 KH |
1064 | /* |
1065 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1066 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1067 | */ | |
f817ed48 | 1068 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 1069 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 1070 | enum lru_list lru) |
cc847582 | 1071 | { |
f817ed48 | 1072 | struct page_cgroup *pc, *busy; |
cc847582 | 1073 | unsigned long flags; |
f817ed48 | 1074 | unsigned long loop; |
072c56c1 | 1075 | struct list_head *list; |
f817ed48 | 1076 | int ret = 0; |
072c56c1 | 1077 | |
b69408e8 | 1078 | list = &mz->lists[lru]; |
cc847582 | 1079 | |
f817ed48 KH |
1080 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1081 | /* give some margin against EBUSY etc...*/ | |
1082 | loop += 256; | |
1083 | busy = NULL; | |
1084 | while (loop--) { | |
1085 | ret = 0; | |
1086 | spin_lock_irqsave(&mz->lru_lock, flags); | |
1087 | if (list_empty(list)) { | |
1088 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 1089 | break; |
f817ed48 KH |
1090 | } |
1091 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1092 | if (busy == pc) { | |
1093 | list_move(&pc->lru, list); | |
1094 | busy = 0; | |
1095 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
1096 | continue; | |
1097 | } | |
9b3c0a07 | 1098 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
f817ed48 KH |
1099 | |
1100 | ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE); | |
1101 | if (ret == -ENOMEM) | |
52d4b9ac | 1102 | break; |
f817ed48 KH |
1103 | |
1104 | if (ret == -EBUSY || ret == -EINVAL) { | |
1105 | /* found lock contention or "pc" is obsolete. */ | |
1106 | busy = pc; | |
1107 | cond_resched(); | |
1108 | } else | |
1109 | busy = NULL; | |
cc847582 | 1110 | } |
f817ed48 KH |
1111 | if (!ret && !list_empty(list)) |
1112 | return -EBUSY; | |
1113 | return ret; | |
cc847582 KH |
1114 | } |
1115 | ||
1116 | /* | |
1117 | * make mem_cgroup's charge to be 0 if there is no task. | |
1118 | * This enables deleting this mem_cgroup. | |
1119 | */ | |
c1e862c1 | 1120 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1121 | { |
f817ed48 KH |
1122 | int ret; |
1123 | int node, zid, shrink; | |
1124 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1125 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1126 | |
cc847582 | 1127 | css_get(&mem->css); |
f817ed48 KH |
1128 | |
1129 | shrink = 0; | |
c1e862c1 KH |
1130 | /* should free all ? */ |
1131 | if (free_all) | |
1132 | goto try_to_free; | |
f817ed48 | 1133 | move_account: |
1ecaab2b | 1134 | while (mem->res.usage > 0) { |
f817ed48 | 1135 | ret = -EBUSY; |
c1e862c1 KH |
1136 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1137 | goto out; | |
1138 | ret = -EINTR; | |
1139 | if (signal_pending(current)) | |
cc847582 | 1140 | goto out; |
52d4b9ac KH |
1141 | /* This is for making all *used* pages to be on LRU. */ |
1142 | lru_add_drain_all(); | |
f817ed48 KH |
1143 | ret = 0; |
1144 | for_each_node_state(node, N_POSSIBLE) { | |
1145 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
1ecaab2b | 1146 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1147 | enum lru_list l; |
1ecaab2b | 1148 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
f817ed48 KH |
1149 | for_each_lru(l) { |
1150 | ret = mem_cgroup_force_empty_list(mem, | |
1151 | mz, l); | |
1152 | if (ret) | |
1153 | break; | |
1154 | } | |
1ecaab2b | 1155 | } |
f817ed48 KH |
1156 | if (ret) |
1157 | break; | |
1158 | } | |
1159 | /* it seems parent cgroup doesn't have enough mem */ | |
1160 | if (ret == -ENOMEM) | |
1161 | goto try_to_free; | |
52d4b9ac | 1162 | cond_resched(); |
cc847582 KH |
1163 | } |
1164 | ret = 0; | |
1165 | out: | |
1166 | css_put(&mem->css); | |
1167 | return ret; | |
f817ed48 KH |
1168 | |
1169 | try_to_free: | |
c1e862c1 KH |
1170 | /* returns EBUSY if there is a task or if we come here twice. */ |
1171 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1172 | ret = -EBUSY; |
1173 | goto out; | |
1174 | } | |
c1e862c1 KH |
1175 | /* we call try-to-free pages for make this cgroup empty */ |
1176 | lru_add_drain_all(); | |
f817ed48 KH |
1177 | /* try to free all pages in this cgroup */ |
1178 | shrink = 1; | |
1179 | while (nr_retries && mem->res.usage > 0) { | |
1180 | int progress; | |
c1e862c1 KH |
1181 | |
1182 | if (signal_pending(current)) { | |
1183 | ret = -EINTR; | |
1184 | goto out; | |
1185 | } | |
f817ed48 KH |
1186 | progress = try_to_free_mem_cgroup_pages(mem, |
1187 | GFP_HIGHUSER_MOVABLE); | |
c1e862c1 | 1188 | if (!progress) { |
f817ed48 | 1189 | nr_retries--; |
c1e862c1 KH |
1190 | /* maybe some writeback is necessary */ |
1191 | congestion_wait(WRITE, HZ/10); | |
1192 | } | |
f817ed48 KH |
1193 | |
1194 | } | |
1195 | /* try move_account...there may be some *locked* pages. */ | |
1196 | if (mem->res.usage) | |
1197 | goto move_account; | |
1198 | ret = 0; | |
1199 | goto out; | |
cc847582 KH |
1200 | } |
1201 | ||
c1e862c1 KH |
1202 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1203 | { | |
1204 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1205 | } | |
1206 | ||
1207 | ||
2c3daa72 | 1208 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1209 | { |
2c3daa72 PM |
1210 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
1211 | cft->private); | |
8cdea7c0 | 1212 | } |
628f4235 KH |
1213 | /* |
1214 | * The user of this function is... | |
1215 | * RES_LIMIT. | |
1216 | */ | |
856c13aa PM |
1217 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1218 | const char *buffer) | |
8cdea7c0 | 1219 | { |
628f4235 KH |
1220 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
1221 | unsigned long long val; | |
1222 | int ret; | |
1223 | ||
1224 | switch (cft->private) { | |
1225 | case RES_LIMIT: | |
1226 | /* This function does all necessary parse...reuse it */ | |
1227 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
1228 | if (!ret) | |
1229 | ret = mem_cgroup_resize_limit(memcg, val); | |
1230 | break; | |
1231 | default: | |
1232 | ret = -EINVAL; /* should be BUG() ? */ | |
1233 | break; | |
1234 | } | |
1235 | return ret; | |
8cdea7c0 BS |
1236 | } |
1237 | ||
29f2a4da | 1238 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1239 | { |
1240 | struct mem_cgroup *mem; | |
1241 | ||
1242 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
1243 | switch (event) { |
1244 | case RES_MAX_USAGE: | |
1245 | res_counter_reset_max(&mem->res); | |
1246 | break; | |
1247 | case RES_FAILCNT: | |
1248 | res_counter_reset_failcnt(&mem->res); | |
1249 | break; | |
1250 | } | |
85cc59db | 1251 | return 0; |
c84872e1 PE |
1252 | } |
1253 | ||
d2ceb9b7 KH |
1254 | static const struct mem_cgroup_stat_desc { |
1255 | const char *msg; | |
1256 | u64 unit; | |
1257 | } mem_cgroup_stat_desc[] = { | |
1258 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1259 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1260 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1261 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1262 | }; |
1263 | ||
c64745cf PM |
1264 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1265 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1266 | { |
d2ceb9b7 KH |
1267 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1268 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1269 | int i; | |
1270 | ||
1271 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1272 | s64 val; | |
1273 | ||
1274 | val = mem_cgroup_read_stat(stat, i); | |
1275 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1276 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1277 | } |
6d12e2d8 KH |
1278 | /* showing # of active pages */ |
1279 | { | |
4f98a2fe RR |
1280 | unsigned long active_anon, inactive_anon; |
1281 | unsigned long active_file, inactive_file; | |
7b854121 | 1282 | unsigned long unevictable; |
4f98a2fe RR |
1283 | |
1284 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1285 | LRU_INACTIVE_ANON); | |
1286 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1287 | LRU_ACTIVE_ANON); | |
1288 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1289 | LRU_INACTIVE_FILE); | |
1290 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1291 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1292 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1293 | LRU_UNEVICTABLE); | |
1294 | ||
4f98a2fe RR |
1295 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1296 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1297 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1298 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1299 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1300 | ||
6d12e2d8 | 1301 | } |
d2ceb9b7 KH |
1302 | return 0; |
1303 | } | |
1304 | ||
c1e862c1 | 1305 | |
8cdea7c0 BS |
1306 | static struct cftype mem_cgroup_files[] = { |
1307 | { | |
0eea1030 | 1308 | .name = "usage_in_bytes", |
8cdea7c0 | 1309 | .private = RES_USAGE, |
2c3daa72 | 1310 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1311 | }, |
c84872e1 PE |
1312 | { |
1313 | .name = "max_usage_in_bytes", | |
1314 | .private = RES_MAX_USAGE, | |
29f2a4da | 1315 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1316 | .read_u64 = mem_cgroup_read, |
1317 | }, | |
8cdea7c0 | 1318 | { |
0eea1030 | 1319 | .name = "limit_in_bytes", |
8cdea7c0 | 1320 | .private = RES_LIMIT, |
856c13aa | 1321 | .write_string = mem_cgroup_write, |
2c3daa72 | 1322 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1323 | }, |
1324 | { | |
1325 | .name = "failcnt", | |
1326 | .private = RES_FAILCNT, | |
29f2a4da | 1327 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1328 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1329 | }, |
d2ceb9b7 KH |
1330 | { |
1331 | .name = "stat", | |
c64745cf | 1332 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1333 | }, |
c1e862c1 KH |
1334 | { |
1335 | .name = "force_empty", | |
1336 | .trigger = mem_cgroup_force_empty_write, | |
1337 | }, | |
8cdea7c0 BS |
1338 | }; |
1339 | ||
6d12e2d8 KH |
1340 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1341 | { | |
1342 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1343 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1344 | enum lru_list l; |
41e3355d | 1345 | int zone, tmp = node; |
1ecaab2b KH |
1346 | /* |
1347 | * This routine is called against possible nodes. | |
1348 | * But it's BUG to call kmalloc() against offline node. | |
1349 | * | |
1350 | * TODO: this routine can waste much memory for nodes which will | |
1351 | * never be onlined. It's better to use memory hotplug callback | |
1352 | * function. | |
1353 | */ | |
41e3355d KH |
1354 | if (!node_state(node, N_NORMAL_MEMORY)) |
1355 | tmp = -1; | |
1356 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1357 | if (!pn) |
1358 | return 1; | |
1ecaab2b | 1359 | |
6d12e2d8 KH |
1360 | mem->info.nodeinfo[node] = pn; |
1361 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1362 | |
1363 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1364 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1365 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1366 | for_each_lru(l) |
1367 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1368 | } |
6d12e2d8 KH |
1369 | return 0; |
1370 | } | |
1371 | ||
1ecaab2b KH |
1372 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1373 | { | |
1374 | kfree(mem->info.nodeinfo[node]); | |
1375 | } | |
1376 | ||
c8dad2bb JB |
1377 | static int mem_cgroup_size(void) |
1378 | { | |
1379 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
1380 | return sizeof(struct mem_cgroup) + cpustat_size; | |
1381 | } | |
1382 | ||
33327948 KH |
1383 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1384 | { | |
1385 | struct mem_cgroup *mem; | |
c8dad2bb | 1386 | int size = mem_cgroup_size(); |
33327948 | 1387 | |
c8dad2bb JB |
1388 | if (size < PAGE_SIZE) |
1389 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 1390 | else |
c8dad2bb | 1391 | mem = vmalloc(size); |
33327948 KH |
1392 | |
1393 | if (mem) | |
c8dad2bb | 1394 | memset(mem, 0, size); |
33327948 KH |
1395 | return mem; |
1396 | } | |
1397 | ||
1398 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1399 | { | |
c8dad2bb | 1400 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
1401 | kfree(mem); |
1402 | else | |
1403 | vfree(mem); | |
1404 | } | |
1405 | ||
1406 | ||
8cdea7c0 BS |
1407 | static struct cgroup_subsys_state * |
1408 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1409 | { | |
1410 | struct mem_cgroup *mem; | |
6d12e2d8 | 1411 | int node; |
8cdea7c0 | 1412 | |
c8dad2bb JB |
1413 | mem = mem_cgroup_alloc(); |
1414 | if (!mem) | |
1415 | return ERR_PTR(-ENOMEM); | |
78fb7466 | 1416 | |
8cdea7c0 | 1417 | res_counter_init(&mem->res); |
1ecaab2b | 1418 | |
6d12e2d8 KH |
1419 | for_each_node_state(node, N_POSSIBLE) |
1420 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1421 | goto free_out; | |
1422 | ||
8cdea7c0 | 1423 | return &mem->css; |
6d12e2d8 KH |
1424 | free_out: |
1425 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1426 | free_mem_cgroup_per_zone_info(mem, node); |
c8dad2bb | 1427 | mem_cgroup_free(mem); |
2dda81ca | 1428 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1429 | } |
1430 | ||
df878fb0 KH |
1431 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1432 | struct cgroup *cont) | |
1433 | { | |
1434 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
c1e862c1 | 1435 | mem_cgroup_force_empty(mem, false); |
df878fb0 KH |
1436 | } |
1437 | ||
8cdea7c0 BS |
1438 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1439 | struct cgroup *cont) | |
1440 | { | |
6d12e2d8 KH |
1441 | int node; |
1442 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1443 | ||
1444 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1445 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1446 | |
33327948 | 1447 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1448 | } |
1449 | ||
1450 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1451 | struct cgroup *cont) | |
1452 | { | |
1453 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1454 | ARRAY_SIZE(mem_cgroup_files)); | |
1455 | } | |
1456 | ||
67e465a7 BS |
1457 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1458 | struct cgroup *cont, | |
1459 | struct cgroup *old_cont, | |
1460 | struct task_struct *p) | |
1461 | { | |
1462 | struct mm_struct *mm; | |
1463 | struct mem_cgroup *mem, *old_mem; | |
1464 | ||
1465 | mm = get_task_mm(p); | |
1466 | if (mm == NULL) | |
1467 | return; | |
1468 | ||
1469 | mem = mem_cgroup_from_cont(cont); | |
1470 | old_mem = mem_cgroup_from_cont(old_cont); | |
1471 | ||
67e465a7 BS |
1472 | /* |
1473 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1474 | * in effect owned by the leader | |
1475 | */ | |
52ea27eb | 1476 | if (!thread_group_leader(p)) |
67e465a7 BS |
1477 | goto out; |
1478 | ||
67e465a7 BS |
1479 | out: |
1480 | mmput(mm); | |
67e465a7 BS |
1481 | } |
1482 | ||
8cdea7c0 BS |
1483 | struct cgroup_subsys mem_cgroup_subsys = { |
1484 | .name = "memory", | |
1485 | .subsys_id = mem_cgroup_subsys_id, | |
1486 | .create = mem_cgroup_create, | |
df878fb0 | 1487 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1488 | .destroy = mem_cgroup_destroy, |
1489 | .populate = mem_cgroup_populate, | |
67e465a7 | 1490 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1491 | .early_init = 0, |
8cdea7c0 | 1492 | }; |